AgentDrivenDevelopment.com’s Survive

Chapter 1: Twelve Days

August 2026

Executive Play: Quarantine The Proof

The demo should have been a favor, which is how the most dangerous things usually entered Kedran: politely, late on a calendar invite, with no one important feeling obligated to prepare.

It was August 2026, Central Indiana was pressing its heat against the fifth-floor glass, and the air conditioning in the internal demo room had finally begun to win. I sat in the second row with cafeteria coffee in a paper cup and a spare HDMI adapter in my jacket pocket, because no one who has survived three acquisitions trusts a conference room to have the right cable.

Kedran Industrial Holdings had offices in more than thirty countries and people who spoke more languages than that. We had tens of thousands of repositories, a shelf of acquired technology stacks, and several thousand engineers if you counted the divisions that only remembered central IT existed after a plant dashboard or batch job broke. We were public, profitable, respectable, and large enough that a truth could lose its shape before it crossed two reporting lines.

The company had started in the late nineteenth century building machined parts for railroads, then taken a generation to get electricity fully into its own main factory. A wartime government contract pulled Kedran out of regional usefulness and into the supply chains that eventually made us international. Our oldest story was still the truest one: Kedran adopted the future, but only after finding a way to make the future fit the building.

We were also the kind of company where I could send the same message through four collaboration tools and still have a vice president ask whether I had posted it in the right place. This was the organization people expected me to make AI-ready.

That morning, one truth walked in carrying a laptop.

His name was Devon M. He was a staff engineer, thirty-two or close enough, with rolled sleeves, black sneakers, and a faded college shirt from West Lafayette under a zip hoodie. Devon wore his college gear the way some men wore wedding rings: not as decoration, but as a settled fact. I had gone to college in Bloomington, which meant I noticed and had the good manners not to open with it while he was discovering that enterprise technology begins with a projector that cannot find God.

“Need a minute?” I asked.

“It’s the adapter.”

I handed him mine. A CIO with a spare adapter looks humble in a way no leadership book can teach, and I had learned over the years that usefulness at small scale buys a kind of grace at large scale. Know the cafeteria cashier’s name, ask about the knee brace, return calls from nonprofit directors, carry the adapter. Let people see you solve the tiny problem before they watch you avoid the expensive one.

People liked me. I worked at it, and I meant most of it.

Victor came in while Devon was still fighting the screen. Our CEO did not like arriving early because early made him available, but he had a gap before analyst prep and the demo had acquired just enough AI odor to become unavoidable. He had run plants before he ran Kedran, and some part of him still looked like a man who could hear a bad bearing from across a factory floor. The rest of him had been trained by public-company life to survive every sentence he spoke.

He took the chair beside me.

“Tom.”

“Victor.”

“This the AI thing?”

“One of them.”

“Good. No surprises.”

He said it the way other people said good morning.

Dana K., our CISO, sat on my other side with a tablet open and the expression of someone who had seen enough innovation to know where the bodies were buried. A field-service lead, two business leaders, and a few engineers filled the rest of the room. I knew their names; I also knew the room would soon turn their names into functions if the demo became dangerous enough.

Devon got the screen up and took one step away from the laptop.

“This is rough,” he said.

There was no title slide, no strategic frame, no animation of a rocket leaving a launchpad. Just a staff engineer warning the adults that the screen had not been blessed.

“How rough?” Victor asked.

“Visually or legally?”

The room laughed because the sentence was funny until it became evidence.

Devon pulled up a plain white browser window with Kedran gray buttons and a header that said Parts Quote Assistant. It looked like software made by people trying to solve a problem before anyone could schedule a meeting about the problem.

“We picked service-parts quoting because the intake queue is buried,” he said. “Reps are copying part numbers out of PDFs and old bulletins. Customers call with equipment history, regional price rules, and a harvest window, and we make them wait while five systems disagree with each other. We wanted to see if call notes plus equipment context could get a rep to a quote draft.”

Dana asked whether he had touched production data.

“Read-only staging copy for installed base and parts,” Devon said. “Masked pricing table. Live inventory through the service API, availability only.”

Her pen stopped. “Who approved the live inventory path?”

Devon did not look at me, which was his first mistake.

“We used a service token from the platform team.”

In a company our size, half the world runs on favors with better names: temporary access, legacy exception, shared service account, one engineer helping another because the approved path is a hallway full of locked doors and payroll still has to run on Friday. Devon had just said the quiet part into a microphone, and the room changed around him.

Victor turned his head half an inch toward me.

I did not stop the demo. Stopping it would have made the problem larger, and letting it continue kept the room breathing. Also, I wanted to see the tool work. I have always loved the moment when a machine finally does what people have been talking around for months.

“Show us the flow,” I said.

Devon entered an equipment ID, pulled up a compressor assembly outside Peoria, and pasted a field note with the grammar service reps use when they are typing in a truck with one thumb.

Customer hears grinding during startup. Belt replaced in May. Wants quote for full assembly if failure likely before harvest window.

The tool paused for four seconds, then built a quote draft.

The quote was ugly. The table had the social grace of a parts catalog, and the language sounded as if it had been raised by a warranty database, but the parts were right, the stock availability matched the field-service lead’s notes, and the tool flagged a revised bracket our normal process missed often enough to have its own angry email template.

The field-service lead leaned forward. “It caught the bracket.”

Devon nodded. “From the service bulletin.”

“That bulletin isn’t in the parts system.”

“No. PDF library.”

Nobody moved. The silence mattered more than the model or the screen, because a business person had just realized the unofficial tool saw the work more clearly than the official system.

She took the keyboard and made it harder: older assembly, messier history, a customer note no product manager would ever include in a requirement document. Six seconds later, the tool produced another quote draft.

“That’s not bad,” she said.

At Kedran, that sentence could carry more force than applause.

Victor leaned back, and I watched him do the math. Six to nine months for the formal version if intake ranked it high enough, architecture had capacity, procurement and legal moved, the business funded the work, and no division president decided the shared platform was a central IT land grab.

Devon and three engineers had done it in twelve days.

I asked the question because I needed the room to hear the number from him.

“How long?”

Devon swallowed. “Twelve days.”

No one wrote it down. They did not have to. The number wrote itself on the walls.

“How many people?”

“Four engineers,” he said. “Plus field context. Not full-time.”

“Nights?”

He gave the embarrassed shrug good engineers use when they have accidentally revealed affection for a problem.

The feature was small, ugly, and probably carrying six policy violations under its coat. Kedran made physical products with software and controllers buried inside them, and we outsourced too much of that work because our own delivery system could not reliably ship fast and clean at the same time. Devon had done in twelve days what our approved roadmap had wrapped in two quarters of prioritization language, and the danger came from the fact that it worked.

I had seen Kedran do this before. The old Agile and DevOps artifacts were still in the system: maturity decks, coach rosters, deployment dashboards, ceremony scores, and lessons-learned folders nobody opened unless a new program needed ancestral permission. AI moved faster than either of them had, but Kedran had grown older too.

I had been the good guy in those rooms, or tried to be. I protected engineers when a release missed because the approval path had become a second product. I found budget for platform teams everyone praised and no business unit wanted to fund. I sat beside directors whose spouses had cancer, kids in college, parents in assisted living, and told them the reorganization would not be as reckless as the rumor made it sound. I believed, for longer than I should have, that enough patience and enough executive air cover could make a system choose against its own protection.

The DevOps transformation had cured me of the clean version of that belief. We had spent more than two hundred million dollars over five years if you counted tools, coaches, platform contracts, consulting work, training weeks, internal labor, and the quiet backfill no one liked putting on a slide. The final board update called it modernization. The operating readout was harder to frame: we shipped slower in the systems that mattered, carried more headcount around every release, and had worse uptime in the places where the new controls created another handoff before anyone was allowed to fix the old one. The system had not rejected the money. It had eaten the money and used it to become larger.

People outside the room imagine authority as a switch. The CIO should flip it. The CxO suite should demand it. The board should order it from the top and make the old machinery obey. I had sat in enough prep rooms to know the limits. Victor could demand urgency. Mara could move budget. I could sponsor, escalate, protect, hire, threaten, persuade, and take the blame. The board could ask sharper questions than management wanted to answer. None of that gave a product team the right to cross a business-unit P and L, retire a platform tied to warranty revenue, ignore a plant freeze, change workforce planning, or turn a quarter into a rebuilding season. A public company can survive bad architecture longer than it can survive a bad quarter, and everyone with real authority knows which one shows up first.

I began asking questions, calmly, because sharp questions create defendants and calm questions create adults. Which model? Commercial. Which code tool? A second vendor, used locally. What went into the prompts? Documentation, service bulletins, schema notes, sample call text. Who anonymized the notes? The team did, mostly by hand. Which repositories? Two read-only clones, one new service, one temporary front end. Legal approval? No. Security review? No. Architecture review? No. Platform backlog? No. Product intake? No.

“Manager know?”

Devon looked at the floor. “Parts of it.”

The room laughed again, the kind of laugh people use when the elevator drops one floor too fast.

I could have made Devon the problem. Unauthorized access. Unreviewed AI tooling. Unapproved data path. Weekend work. Hero culture. Every phrase was true, and every one would have fit neatly in an incident memo.

The issue was that Devon was not the issue.

He had done the exact thing leadership claimed to want: found a customer pain, pulled in users, used the tools at hand, and built a working tool before the organization could turn the idea into a document. Punish him, and the builders would learn the lesson. Celebrate him, and the enterprise would learn the wrong one. Ignore him, and Victor would still have the number.

Twelve days.

Victor said my name quietly.

“Tom.”

One word, and inside it were all the questions he could not ask in front of the room. Is this good? Is this bad? Is this ours? Will it become a board question, a headline, a savings promise, or a workforce rumor by lunch?

I stood.

“Devon, first, thank you,” I said. “This is real work against a real pain point. You and the team made the work visible, and that matters.”

His shoulders dropped half an inch.

“Second, we need to treat this as a controlled learning artifact, not a deployed product. There are data, security, IP, authorization, and support questions we have to answer before this leaves the room.”

The security chief looked at me. She knew what I was doing, and she also knew the words were correct.

“Of course,” Devon said. “We weren’t trying to ship it.”

He meant that. I believed him. Most trouble in large companies begins with decent people not trying to cause trouble.

Victor asked, “But it works?”

The field-service lead answered before I could. “It works enough to make my reps mad if they see it and can’t use it.”

That was the cleanest sentence in the room, so I smiled.

“Then we should be careful who we make mad.”

The room laughed one last time. It helped. A room can absorb almost anything if you let it laugh once on the way down.

Afterward, people stood too quickly. Nobody wants to be first to panic or last to understand after a dangerous demo. Victor pulled me aside near the credenza where the stale cookies lived.

“How contained is that?”

“Contained enough for today.”

He looked back toward the room, where Devon was unplugging his laptop with the careful movements of a man handling evidence. Victor’s face had gone plant-manager flat, which meant he was already sorting the thing into risk, cost, and what he could say to the board without creating a promise.

“Should I be excited or worried?” he asked.

“Both, but not in public.”

“Can we use it?” he asked.

“Not yet.”

“Can we talk about it?”

“Only carefully,” I said. “And not broadly.”

He did not like that, but he liked surprises less. “Board?”

“Not as a feature,” I said. “As a signal that we need a responsible path.”

He nodded once. “No surprises.”

“No surprises,” I said.

The security chief joined us with a list already forming: prompts, logs, architecture, repositories, access path, vendor terms, and any customer data that may have touched a provider. She wanted answers from Devon directly, not through four management layers, and she did not want the demo used as an excuse to freeze all AI work.

“Agreed,” I said.

She left with the posture of a person carrying live ammunition in a paper bag.

I found Devon in the hallway, coiling the borrowed HDMI cable so badly that I took it from him and did it myself.

“Sorry,” he said.

“For the cable or the controlled substance you brought to a quarterly review?”

He smiled despite himself. “Both.”

“And the shirt.”

He looked down at the college logo. “I wondered how long you could hold out.”

“I am a senior executive,” I said. “We are trained to delay hard conversations.”

Up close, he looked younger than thirty-two, not because of his face, but because he still believed useful work should be allowed to matter quickly.

“You did good work,” I said.

“It doesn’t feel like that’s what happened.”

“Both can be true.”

He leaned against the wall. “We weren’t trying to bypass everybody. The normal path is impossible. Field service has been asking since last winter. The platform team said maybe Q2 next year if Components didn’t pull capacity. Reps are copying part numbers out of PDFs. We thought, with the new tools, maybe we could just see.”

“And now we have seen.”

“So what happens?”

I knew what could not happen. We could not let four engineers become the proof that Kedran’s delivery model was optional, and we could not let Victor stand before the board with a twelve-day miracle and no answer for why the same company needed nine months to produce lesser things. A prototype could not become a referendum on several thousand engineers, tens of thousands of repositories, a run of acquisitions, and a public-company calendar that turned every hard question into next quarter’s problem.

We also could not pretend it had failed.

“First, you give Security everything they asked for,” I said.

“Okay.”

“Second, you stop expanding it. No new users. No new data. No quiet improvements over the weekend.”

His face confirmed the weekend plan.

“Third, document what you did. Not a slide deck. A build note. Decisions, shortcuts, assumptions, where the tools helped, where they hallucinated, where our process would have stopped you.”

“That could be useful.”

“It will be.”

He waited, and I put a hand on his shoulder. Rare for me at work. He looked like a man who had shown adults a working motor and been told the extension cord was a governance failure.

“Devon,” I said, “do not show this to anyone else yet.”

He searched my face. “Because of Security?”

“Partly.”

“Legal?”

“Partly.”

“Because it makes people uncomfortable?”

He had named the part everyone else was trying to route through policy.

“Because once proof gets loose,” I said, “people start making promises before they understand the cost.”

He nodded. He wanted to believe I was protecting the work.

I wanted to believe it too.

When he left, I stood alone in the hallway with the adapter in my hand. Downstairs, payroll ran, plants got parts, distributors got answers, and families counted on all of it without knowing our repository count or the favors that kept ancient systems breathing.

I put the adapter back in my jacket pocket and went to find Dana before the evidence learned to travel without me.

Chapter 2: No Surprises

September 2026

Executive Play: Build A Narrative Without An Obligation

Two weeks after Devon’s demo, I printed my organization chart and spread it across the conference table in my office.

The chart was already wrong by the time the printer stopped. Kedran kept people data in three systems, and none of them agreed after the third level, but paper still made a mess feel more honest because you could put a hand on it. Fifteen hundred people sat under my name across Central Indiana, acquired offices near plants, a few remote clusters that still used the old company chat, and enough dotted-line arrangements to keep HR employed forever.

Delivery towers. Platform groups. Service owners. Architects. Program managers. Reliability teams. Security liaisons. Data engineers who were mine on paper and the business unit’s in every meeting that mattered.

I stood over the chart with a blue pen and a cup of coffee my assistant had bullied the cafeteria into making stronger. If trust was the missing platform, maybe the honest plan was to send all fifteen hundred of us to a ropes course for a year and call the invoice an operating model.

I laughed once, alone in my office, then took that thought and left it out of the deck.

The board technology committee had added AI-assisted software development to the September agenda. Victor wanted a story before the pre-read closed. Mara wanted the story not to become a number. Dana wanted the story not to become a breach. I wanted the story not to become a mandate, because a mandate would force questions Kedran had spent twenty years learning how to postpone politely.

I knew the old script because I had helped perform parts of it. Agile and DevOps had arrived with useful ideas and serious people; years later, their most durable remains were ceremony metrics, maturity dashboards, coaching contracts, and phased rollout language. The new vocabulary had spread faster than the power needed to make the vocabulary matter.

At my level, power is mostly calendar control. It is not the inspirational kind people write about after they leave industry and find a good photographer. It is deciding who gets thirty minutes before the board book locks, which question is answered live and which one becomes a follow-up, which work is framed as urgent and which work is framed as sequenced. A lot of careers turn on the difference between those two words.

By September, I had started thinking of the AI work as having two zones. The safe zone was not where nothing changed. That would have bored Victor, insulted the engineers, and wasted the opportunity. The safe zone was where everyone saw enough of what they needed to see: the board saw discipline, the CEO saw motion, Finance saw no premature savings claim, Security saw controls, engineers saw permission to learn, and I saw another quarter in which the proof did not become a public-company demand. The unsafe zone was smaller and more dangerous. It began at the sentence that said the tools worked well enough to expose the operating model. In that zone, I did not become a hero. I became the predecessor people described kindly after he was gone, and the work became a cautionary tale with a new owner and a smaller budget.

I had blocked ninety minutes with Victor and Mara in the eighth-floor prep room because that room had no windows and no casual foot traffic. The wall screen was too large, the table was too polished, and the speakerphone in the middle looked as if it had been designed by a person who hated voices.

Before I went upstairs, Hannah texted a med school loan disclosure and wrote, No rush.

No rush from a child means the opposite. She was twenty-five and decent in a way that made the cost harder to complain about, so I told her I would look that night and then looked again at the chart. Fifteen hundred names, one med school bill, one pregnant daughter, a Central Indiana mortgage, an inherited Ludington cabin that always needed another repair, and a compensation plan that rewarded clean quarters more reliably than clean truth.

Nobody pays you seven figures to panic in public.

Victor was already in the prep room when I arrived. He had a legal pad open with nothing written on it, which meant he wanted the meeting to feel hands-on without creating discoverable artifacts. Mara came in two minutes later with a silver laptop and the expression she wore when a topic had not yet become expensive but clearly wanted to.

Mara E. was the CFO and, in most rooms, the closest thing I had to another adult who understood how little room there was between prudence and evasion. She knew timing, promises, analyst memory, margin guidance, and the shape of fear. She did not need the AI work killed. She needed it made financially harmless until the company knew what it could afford to say.

“Tom,” Victor said. “Where are we?”

I connected my laptop. The first slide came up:

Responsible AI-Assisted Development: Board Discussion Frame.

No image. No lightning bolt. Russell M. would have added one if I had let him near it.

“We have three separate facts,” I said. “First, AI-assisted development is real enough that our engineers are already using it. Second, the risk profile is unsettled across source code, customer data, IP, and vendor retention. Third, productivity claims will mislead us if we put enterprise numbers on coding acceleration before we understand the full delivery system.”

Mara looked up. “Full delivery system.”

“Coding is one part of the work,” I said. “Review, test, security, architecture, release, funding, adoption, customer impact. If one part accelerates and the rest stays fixed, the benefit can disappear before it reaches a customer or a financial statement.”

“That sounds like a reason not to promise savings.”

“It is.”

Victor tapped his pen against the legal pad. “I don’t want a science project.”

“Agreed.”

“I don’t want a security issue.”

“Also agreed.”

He looked at the screen instead of at me. “And I don’t want Evelyn asking why four engineers can build in twelve days what we budgeted six months for.”

So Devon’s demo had stayed with him too. I had hoped it would; I had also hoped it would not.

“Then we should not frame it as a productivity breakthrough,” I said.

Victor frowned. “But it might be one.”

“Locally, yes. In bounded conditions, with direct user access, informal authority, and no production obligations, it might be exactly that. At enterprise scale we do not know yet, and saying more gives the board a number the operating model cannot defend.”

That was all true. Truth is useful when it also buys time.

Mara leaned back and ran the sequence before Victor could. If the tools worked, investors and analysts would ask when it showed up in headcount. If we said not in headcount, they would ask about vendor spend. If we said not vendor spend, they would ask why we were investing. She did not need to dramatize it; the questions had the clean inevitability of quarterly math.

“Option value,” I said.

She gave me half a smile. “That’s my phrase.”

“I learned from the best.”

People sometimes think executive teams spend their time trading politically correct statements. Some do. The better ones develop a language for not forcing decisions before anyone wants the decision forced, and Mara and I could be honest because neither of us needed the other to pretend the danger was fake. Victor needed a board story that sounded like movement without becoming a promise he could be measured against by March.

I moved to the second slide.

Five bullets sat there in Kedran blue:

Kedran will pursue AI-assisted development responsibly.

Initial work will be bounded to approved pilots.

Security, IP, and data controls will lead adoption.

Metrics will mature as pilots clarify valid baselines.

No workforce, productivity, or operating-model commitments are implied at this stage.

Victor read the slide twice.

“That last bullet.”

“Important.”

“Too defensive?”

“Only if someone wants it to be.”

Mara rested her fingers on the laptop lid. “I like it. It keeps Compensation from asking whether we are baking savings into next year’s plan.”

“It also keeps Legal from asking why we signaled productivity impact before control readiness,” I said.

Victor looked between us. “I need a verb more active than ‘we are being careful.’”

“We can give them activity.”

I clicked again.

Phase One: Controlled Pilots.

The word controlled did most of the work. Controlled meant safe. Pilots meant small. Phase One implied a Phase Two, which moved the dangerous questions into a future with better lighting and fewer named owners.

“We identify a limited number of teams,” I said. “We select use cases where data exposure is manageable. Dana’s team is inside the process from the start. We create usage policy, approved tools, logging, exception review, and a clean path for stopping anything that touches production before the controls are ready. Then we capture learning on developer experience, quality, cycle-time indicators, and governance gaps.”

“Indicators,” Mara said.

“Not commitments.”

“Good.”

Victor flipped a blank page on his legal pad and still did not write. “Who leads it?”

“I am considering Lena O.”

Mara nodded. “Developer Experience.”

“Yes.”

“People like her.”

“They do.”

“Does she have the authority?”

The speakerphone hummed in the silence. I could have said no, and the clean answer would have ruined the room. Nobody at Kedran had that kind of authority, not really, because authority moved differently when revenue was nearby. If Lena had the power to change funding, architecture, release calendars, and business-unit roadmaps, the pilot would no longer be a pilot. It would be a reorganization with a model provider attached.

“She has enough credibility to convene the right people,” I said.

Mara understood the answer. Victor did too, though he liked the shape of it better.

“Lena is good,” he said. “Engineers trust her.”

“Product leaders return her calls.”

“Dana?”

“Dana will have concerns.”

“Dana always has concerns.”

“Dana is paid to notice the thing everyone else would rather discover after it is expensive.”

Victor accepted that because it was kind and useful. Dana would slow this work with real reasons, which made the slowing durable.

Mara looked back at the bullets. “What about the rogue demo?”

“We do not call it rogue.”

“What do we call it?”

“An internal proof point that surfaced the need for a governed adoption path.”

She laughed through her nose. “An outside advisor would charge us two hundred thousand dollars for that sentence.”

“I saved the company money.”

“Put it in your self-review.”

Victor did not laugh. He had gone back into board prep, where humor becomes a luxury item.

“Can I say we have seen promising early results?” he asked.

“Yes.”

“Can I say AI may improve developer productivity?”

“May, yes. Will, no.”

“Can I say we are moving quickly?”

“Responsibly.”

He grimaced. “That’s a slow word.”

“It’s a public-company word.”

Mara closed her laptop halfway, then opened it again. “It is also the word that keeps a reporter from asking why engineers put customer data into an unapproved tool.”

Victor pointed the pen at her. “That did not happen.”

“We do not know that yet,” she said.

He looked at me, and for a second I could see the plant manager still inside the CEO, the man who hated being told a machine might have failed when no alarm had sounded.

“Dana is reviewing,” I said. “So far, nothing material. But we should not get ahead of the facts.”

The line settled over the table with the clean weight of ethics and delay.

We spent the next forty minutes shaping language. Victor wanted urgency. Mara wanted containment. I wanted sequencing. Adopt became pursue. Productivity became capability. Savings became option value. Operating model became governance. The sentence “AI-assisted development could materially alter software delivery economics” became “AI-assisted development may create meaningful opportunities to improve developer experience, quality, and delivery effectiveness over time.”

Over time is a beautiful phrase. It contains no date and offends almost no one.

At one point, Victor stood and walked to the wall screen.

“I know what Evelyn will ask,” he said. “She’ll ask whether this changes how fast we ship software.”

Evelyn N. was the board member we prepared for even when she was not on the agenda. Former founder, former CTO, still carrying enough builder memory to ask simple questions and wait through the fog.

“What do I say?”

“You say early evidence suggests coding tasks can accelerate in controlled settings, but enterprise delivery speed depends on end-to-end flow. We are using pilots to understand where the bottlenecks move.”

Victor looked at the slide. “That’s true?”

“Yes.”

“Is it satisfying?”

“No.”

He turned back. “Can we make it satisfying?”

“We can make it responsible.”

Mara lowered the laptop lid halfway, which meant she was done unless a sentence foolish enough to reopen it entered the room.

“Victor,” she said, “a satisfying answer creates a financial question. A responsible answer creates a follow-up.”

I wrote her sentence down.

Victor sat down. “No surprises.”

“That’s the frame,” I said. “We are not saying no to AI. We are not promising savings. We are not changing workforce planning. We are not bypassing security. We are telling the board the technology is real and Kedran is taking a disciplined path.”

“And if they ask for metrics?”

“Metrics are part of Phase One.”

Mara added the line that earned her bonus: “Baseline quality varies across acquisitions. We need a credible measurement foundation before externalizing numbers.”

I wrote it down. Sometimes the CFO gives you a gift and you do not insult her by leaving it in the air.

The meeting ended with action items, which are how executives reassure themselves that thought has become motion. I would revise the deck. Mara would review the financial language. Victor would use responsible path in analyst prep and let people believe the phrase had arrived naturally. Dana would receive a separate briefing so she did not learn from the board pre-read that she had been made a pillar of the strategy.

As we stood, Victor touched my arm, and the CEO voice dropped for half a second. The plant manager underneath it was still there, the man responsible for a machine too large to hold and too dangerous to ignore.

“Tom. You believe this is real?”

“Yes,” I said.

“Not hype?”

“Not just hype.”

“And Kedran can do it?”

There are questions where the honest answer is a paragraph and the useful answer is a sentence.

“Kedran can learn how to do it responsibly,” I said.

He nodded. It was enough for him because enough was what he had asked me to bring.

After they left, I stayed in the prep room and looked at the final slide: assess, pilot, scale. Every transformation deck I had ever seen had some version of that path. The steps were not wrong, which made them harder to distrust.

My phone buzzed. Hannah had sent the loan disclosure and a photo of a vending machine dinner: pretzels, iced tea, an apple with a bruise.

Eat protein, I texted.

She wrote back: Yes CIO.

I opened the deck notes and put Mara’s sentence beneath the metrics slide, where Evelyn would see it and Victor could read it aloud without sounding evasive. Then I sat for another minute with the remote in my hand, the wall screen still glowing, the legal pad untouched except for the small crescent Victor’s pen had pressed into the top sheet.

I was not a villain. That would have been simpler and less accurate. I was a man doing the job the institution had taught me to do: keep the company steady, keep the board calm, keep the CEO covered, keep the promises soft enough to survive the quarter.

The lane opened in front of me.

Responsible AI-assisted development.

Bounded pilots.

Security and IP controls first.

Metrics as maturity improves.

No workforce implications.

No productivity commitment.

No operating-model change.

No surprises.

By the time I walked back to my office, the chart was still spread across the conference table. Fifteen hundred names waited under my pen. I folded the paper carefully along the creases and slid it into the drawer where I kept old budget drafts, thank-you notes from nonprofit boards, and a football parking pass I could never remember to put in the car.

The path was a politically correct statement that also happened to be usable.

I could put it in the board book.

Chapter 3: Cabin Math

Late 2026

Executive Play: Make Delay Sound Like Stewardship

The boy at the robotics table had the motor wired backward.

He was eight, maybe nine, with one front tooth missing and both shoulders locked up near his ears because he was trying not to cry in public. The basement gym at the Doverline Youth Program was full of Saturday noise: sneakers on old varnish, folding chairs scraping, a baby losing patience, donors in linen jackets pretending they could hear the director over the coffee urn. In the middle of it all, the robotics table had become a small island of defeat.

The motor sat in a plastic chassis with two wires, red and black, crossed through a breadboard in a pattern that made sense only if you believed electricity admired confidence.

“It hates me,” the boy said.

“It doesn’t know you well enough yet.”

His mother looked relieved that an adult in a blazer had decided to crouch down and become useful. I was there for the donor breakfast, not the robotics table. My name badge said Thomas V., Kedran Industrial Holdings. I had already shaken hands with trustees, a pediatric surgeon, and a foundation officer who wanted to discuss digital access in after-school programs. I liked all of it: the building, the controlled noise, the way every good program was really a maintenance problem disguised as hope.

But the boy’s motor mattered more to me in that moment than the coffee and the donors.

“May I?”

He handed me the chassis. I moved the red wire one row over, put the car back in front of him, and asked him to try again. The wheels spun so hard the little frame jumped against his palm.

“I fixed it,” he said.

“You did.”

He had not, not exactly. I had moved the wire. But he had stayed at the table long enough to let current find a path, and sometimes that is the part a child needs credited.

The program director caught my eye from beside the donor wall and smiled. She was used to me wandering off from adult conversations. In Central Indiana, you can be anonymous or useful, and I had chosen useful early because it made life better and because being known is a kind of insurance. At charity breakfasts, school fundraisers, and neighborhood dinners, people called me Tom and asked about Claire, the kids, Kedran, the offensive line. I liked it. That is not noble. It is also not nothing.

By noon, I had written two checks, promised to introduce the program’s technology lead to someone in our network group, and spent ten minutes with a junior staff member who had been assigned the seating chart and looked as if table placement had taken a year off her life.

“You did fine,” I told her.

“Table six was supposed to have the board chair.”

“Table six survived.”

“Barely.”

“That is the standard most of us work to.”

She laughed and looked less alone. I meant the line more than she knew.

From the program, I drove north to meet Claire at the cabin outside Ludington. Late summer in Michigan has a way of making adulthood feel temporarily negotiable. The fields outside the city flattened into hours, then the air changed, the pines thickened, and by the time I crossed into Mason County my phone had enough missed messages to remind me that distance had solved nothing. It had only put the work in my pocket.

Claire had driven up the night before with groceries, clean sheets, and the quiet efficiency of a former hospital administrator. When I pulled into the gravel drive, she was on the porch in bare feet, reading glasses on top of her head, iced tea sweating on the rail.

“You fixed a machine at the program,” she said.

“How do you know?”

“You look pleased with yourself in a very specific way.”

“Small robot. Wrong wire.”

“Ah.”

She kissed me and took the garment bag from my hand because she knew I would hang it on a chair and call it unpacking.

The cabin sat a few blocks from the water, close enough that the wind could turn the screen door into percussion. My grandfather had built it before I was born with borrowed tools, uneven lumber, and the practical ambition of a man who wanted a place to clean perch and get out of town. It was less retreat than fishing shack with bedrooms: knotty pine, sloping porch, soft spot near the back door, a fuse box that made every electrician sigh before opening it.

It had no mortgage. That did not make it free. The roof, the septic line, the insurance, the winter damage, the property taxes, and the dock boards all found me with the persistence of people who knew I could pay. The place was emotionally older than my executive life, which made it harder to sell and easier to turn into another obligation.

That afternoon Lauren and her husband arrived from Chicago with a cooler and the new gravity of a pregnancy that entered the room before she did. Lauren was thirty-one, my oldest, and already carried one hand near her stomach when she paused in a doorway. I had been a father for three decades, and there was a private unfairness in watching your child become someone’s mother. The future had started making appointments without asking me.

Hannah called while I was lighting the grill. The school had changed a disbursement date. Probably fine, she said, which is a phrase children learn from parents who use manageable too often.

“Send me the email,” I said. “I’ll look.”

“Dad, I can handle it.”

“I know.”

“Then why do you always say send it?”

I watched the grill flame catch and the smoke move sideways in the lake wind.

“Because I can read faster than you can worry.”

She went quiet for half a beat. “That’s a good dad answer.”

“I have a few.”

She laughed and went back to an anatomy chapter I did not want described before dinner.

By evening, the cabin was full in the old way. Lauren and her husband argued kindly about baby names. Claire made a salad with more herbs than any salad needed. The dog clicked across the floor. A baseball game ran on mute, and for several hours I was good at being there. I asked Lauren how she was sleeping and listened to the answer. I did not check email during dinner. I ignored a text from our son making fun of our Wi-Fi, which had been installed by a local provider whose support model appeared to be one man named Gary and weather.

After dinner, we walked to the water. Lauren moved slowly, and I slowed with her without making a ceremony of it. Claire noticed. Claire noticed everything.

The lake was restless and pewter-colored under a low sky. Far up the shoreline, the old coal-fired ferry still made its crossing, hauling people and cars across Lake Michigan like an argument nobody had won. I had watched it for years, and it bothered me and comforted me in equal measure. By every clean theory of progress, it should have been gone before my children were born: coal-fired, slow, expensive, beloved by some, criticized by others, maintained by people who knew more about stubborn machinery than most consultants knew about change. The obvious modern answer was to convert it, retire it, replace it, or write a column about why somebody else should pay to make the old thing match the new century. The harder answer was the one operators understood. The boat still moved people and cars. The tickets still sold. The economics of changing the power plant did not become good just because the existing one offended the future.

And still it crossed. Morning and evening. Smoke, whistle, wake.

Old systems do not vanish because better ones exist. They vanish when the replacement can carry the work and pay for its own disruption. I did not say that out loud to Lauren, because she was describing a crib she liked, but the sentence had become part of my private theology of work.

Later, after everyone went to bed, I sat at the kitchen table with my laptop open and the lights low. The cabin made its night sounds: ice maker, settling wood, the soft rush of the dishwasher. From upstairs came one heavy thud, probably the dog deciding gravity had turned against him.

I had three windows open.

Hannah’s tuition portal.

Our brokerage account.

A spreadsheet named household planning because I did not want to name it how long can we keep doing this.

People outside executive life imagine money becomes simple once the number gets large. Some parts do. I wrote checks to youth programs and neighborhood fundraisers without moving money between accounts first, and that was real privilege; I tried not to insult it by pretending otherwise. But I had made more money than anyone in my family history and still could not stop needing the next year to come in clean.

The Central Indiana house. The Ludington cabin. Hannah’s med school. Help we had given Lauren and her husband. The college funds I wanted to open before the baby had a Social Security number. Taxes. Club dues. Charitable commitments that had begun as generosity and become identity. A lifestyle built around the assumption that the next grant would vest, the next bonus would pay, the next year would behave.

The spreadsheet was not melodrama. Hannah had roughly eighty-six thousand dollars due over the next school year once tuition, fees, insurance, and rotation expenses stopped pretending to be separate ideas. The cabin roof quote was thirty-one five, not counting the porch boards my grandfather had installed with more confidence than drainage. Property tax and insurance on the two houses ran just under thirty thousand. The club was twenty-two before meals, which was absurd until you remembered how Central Indiana business still moved through rooms where someone knew your locker number. The Doverline Youth Program pledge was seventy-five, already promised. My next equity vest, if I stayed clean through March, was large enough to make all of that feel manageable and late enough to make every principle sound negotiable after midnight.

I did not say bonus in public. I said responsibility.

Claire came downstairs in a robe and stood behind me long enough that I closed the brokerage tab without thinking.

“That subtle?” she asked.

“I was done with it.”

“Of course.”

She took a glass from the cabinet and filled it at the sink.

“Hannah?”

“Some disbursement timing. I will call the school Monday.”

“She can call the school.”

“I know.”

“Tom.”

I looked at her.

“Let her be capable.”

“She is capable.”

“Then let her feel it sometimes.”

Hannah was only the visible subject. Marriages have footnotes only the two people can see.

Claire sat across from me. The kitchen light caught the silver in her hair. She was fifty-six and still had the calmest angry face I had ever known. Hospital administration had taught her how to absorb panic without becoming it, and marriage to me had given her too many chances to practice.

“How bad is work?” she asked.

“Work is not bad.”

“Then why are you doing math at midnight?”

“Because math is quieter than email.”

She did not smile.

I closed the laptop.

“We had a team build a working tool in twelve days,” I said.

“Useful?”

“Good enough to matter.”

Claire held the glass with both hands and waited. She had a gift for removing padding from language without raising her voice.

“If the tools work, the coding part gets faster,” I said. “But at Kedran, coding is not the whole job. We have architecture review, security review, release calendars, plant windows, acquisition exceptions, business-unit funding, pricing rules, service obligations, and tens of thousands of repositories spread across acquisitions that still remember their original systems. If a team writes code in twelve days and then waits four months to get it reviewed, deployed, funded, or accepted, what changed?”

“The excuse?”

I laughed because she had earned it.

“Maybe.”

“What happens if the AI work succeeds?”

I looked toward the dark window. At night it reflected the kitchen back at me: Claire, the closed laptop, my own face looking older than the Tom people saw in conference rooms.

“If it succeeds in the fantasy version, I get quoted in a magazine and some recruiter calls about a cleaner job with fewer plants and better economics,” I said. “I also have a better chance of winning the lottery, and I do not buy tickets.”

Claire did not laugh. She knew when I was using humor to reach the sentence I did not want to say.

“In the real version, the bureaucracy notices that the proof is dangerous before the board notices that it is useful. People think executives are in charge because we get the offices and the parking spots, but most days the machine is in charge: funding, architecture, business-unit vetoes, procurement, security review, release windows, capital rules, and the memory of the last failed transformation. If I push hard enough to make AI change the operating model, the system will protect itself, and the people carrying the proof will become the cautionary tale.”

Including me. Especially me. I did not say that part.

“Which was?”

“Let him announce the future until he became inconvenient, then thank him for his service.”

I looked back at the laptop. Hannah’s loan disclosure was still waiting behind the closed lid.

“So the board asks why delivery is still slow. Finance asks when savings show up. The business units ask why central IT is touching their work. Security asks whether we changed risk faster than controls. My own people ask what happens to their jobs, and the teams doing the work ask for authority I cannot give them without starting a fight nobody has agreed to have.”

“And you?”

“I keep the company from hurting itself.”

Claire held my eyes. “Are you protecting the company or protecting the life?”

The sentence landed softly because she loved me. That made it worse.

“The life is part of the responsibility,” I said.

She nodded once. Not agreement. Inventory.

I started to tell her the electricity story then. I had read it years earlier and carried it around like a permission slip: factories did not become modern the minute the motor arrived; for years they bolted new power onto old layouts and wondered why the economics lagged. Thirty years, give or take, for the operating model to catch up.

“Tom,” Claire said.

I stopped because she had heard the shape before.

“You are about to compare yourself to industrial history.”

“Only briefly.”

“That is never true.”

Claire looked tired then, not of me exactly, but of the beautiful way I could arrange a defense.

“That is very reasonable,” she said.

“It is.”

“And convenient.”

I did not answer. There was no answer that would improve the room.

Outside, the wind pushed against the cabin. Somewhere upstairs, Lauren coughed. I thought of the baby, not yet born, already changing the math. I thought of Hannah reading loan disclosures under fluorescent lights. I thought of the boy at the robotics table, the motor jumping when current finally went where it was supposed to go.

I wanted to be that man, the one who moved the wire and made the wheels turn.

At Kedran, every wire ran through a committee, a contract, a plant exception, a security policy, a capital plan, and someone’s memory of the last time a change went badly.

Claire reached across the table and put her hand over mine.

“I am not saying burn it down.”

“I know.”

“I am asking whether you still know the difference between stewardship and waiting.”

That was the kind of question she asked. No speech. No accusation big enough for me to fight. Just a practical hole in the floor.

“Most days,” I said.

It was the truest answer I had.

She squeezed my hand and stood.

“Come to bed soon.”

“I will.”

She went upstairs.

I reopened the laptop after a minute, not the brokerage account this time. Work notes. A blank page. The cursor blinked as if it had all night.

I wrote:

Year of the Pilot.

The phrase sat there, harmless and perfect.

It gave Victor motion, Mara containment, Dana a control frame, engineers permission to learn, and me time. A year is a long time in a budget. A pilot is a small thing in a boardroom. Together they sounded like prudence.

I added three bullets.

Bounded teams.

Security first.

Metrics mature over time.

The three bullets looked supportive enough to forward and narrow enough to survive.

I closed the laptop and sat in the dark kitchen a little longer.

Up the shoreline, whether I could hear it or not, the old ferry kept crossing.

Chapter 4: The Perfect Leader

November 2026

Executive Play: Give Visibility Instead Of Power

By the first cold week of November, I had stopped looking for a person who could win.

That was the private version. The version I used in meetings was better. I said we needed credibility across engineering, product, security, and the business units. I said the first year would succeed or fail on trust. I said this could not feel like central IT pushing another program onto tired people. Every sentence was true, which made the private version easier to hide.

At Kedran, the useful politically correct statements were usually built out of true parts.

I had seen the shape before. The CIO before me had done it with Agile and DevOps while I watched from one level down, and the artifacts still lived in our file share: old standup health checks, coach assignments, maturity dashboards, and release-readiness templates that looked official enough to outlast the questions they never answered.

I kept a short list on a yellow pad. I still used paper for decisions that made me ashamed because it slowed my hand down.

Eight names.

Several of them worked for me directly, and several of them wanted my job.

Not crudely. Not in a way I could punish without looking smaller than the office required. Ambition at that level wears good manners. It asks useful questions in steering meetings, volunteers for enterprise problems, compliments your judgment in front of the CEO, and makes sure the board sees its fingerprints on anything that might become the future.

Technically, I could fire any one of them. The org chart said so. The org chart did not show who played golf with which division president, who had helped Victor through the last plant outage, who Mara trusted on capital planning, or who Evelyn might call if she wanted an unvarnished view of technology. Give the AI transformation to the wrong ambitious direct report, and suddenly the work becomes a succession platform. Remove that person after the platform gets attention, and the story is no longer about performance. It is about whether Tom V. is protecting Kedran or protecting Tom V.

That is another lesson people miss about executive jobs. You can be powerful and still trapped. The cage is nicer, and the calendar invites come from people with better titles.

Two were too senior and would ask for money before anyone had agreed this was a money problem. One had enough political capital to pull a business-unit president into a fight, which meant he would lose loudly and Victor would ask why I had let the noise start. One was brilliant and impossible. One would turn the pilot into a vendor selection process by Thanksgiving. One would scare the engineers. One would bore them.

Then there was Lena O.

Lena ran Developer Experience, which meant she owned the places where engineers complained and nobody powerful had to listen too closely: build environments, documentation, developer portals, community meetings with bad coffee and useful questions. If a team could not get a local environment running, Lena’s people usually knew why. If a product leader needed engineers to stop hating a tool, they called her before they called me.

People liked her. That was not a small thing in a company where every change had a scar attached to it. Engineers trusted her because she had been one and still remembered what a broken build did to a Tuesday. Product managers answered her emails because she did not waste their time. Security did not roll its eyes when she entered the room. Architecture respected her enough to say no in complete sentences.

She had no P and L, no authority over business-unit roadmaps, no power to move release dates, no right to tell Celia N. what counted as priority work, no budget large enough to frighten Finance, and no executive mandate that could survive contact with the plant presidents.

She was perfect.

She also remembered the Agile rollout accurately: posters, team renamings, sprint metrics that never touched the funding queue. That made her credible. It also meant she would know when I was rebuilding the same cage with newer wire.

The cafeteria cashier that morning wore a football scarf over her Kedran polo. Her son had started at West Lafayette that fall, and I asked how he was doing before she could ask whether I wanted the oatmeal again.

“Homesick,” she said. “But he won’t admit it.”

“Good,” I said. “Means he likes you.”

She laughed and gave me an extra banana. That was the kind of thing people noticed about me. I noticed them first.

I loved Kedran in the way you can love a city block or a church basement: not the logo, not the investor letter, but the people who kept showing up under bad lighting. By the time I reached the sixth floor, I had already begun calling Lena the humane choice.

I found her at eleven in a conference room on the sixth floor, sitting on the edge of a table while a dozen engineers argued about build times. She had one boot hooked around a chair leg and a marker in her hand. On the whiteboard behind her, someone had written in block letters:

WHY DOES THIS STILL TAKE 45 MINUTES?

Lena was not performing calm. She had the marker uncapped, a paper cup of coffee near her knee, and the patience of someone who had already decided the answer was not going to be a tool name.

“If the answer is ‘because the old build server,’ that is not an answer,” she said. “That is a museum label.”

The room laughed. She did not let them stay there.

She walked them through the ownership map. Agent images sat with Platform. The dependency cache sat with Platform too, theoretically, though nobody could name the person who could approve a change. The product team owned the pain, Security owned one of the controls, Infrastructure owned the old runners, and every group in the room owned enough of the problem to explain why it could not move.

“Then let’s stop calling it a tooling problem,” she said, and capped the marker hard enough for the sound to carry into the hall.

I stood outside the glass and watched her do the dangerous thing naturally. She followed work to authority. Most executives did not. Most of us followed work until it reached a boundary, then gave the boundary a respectable name.

After the meeting, she saw me and smiled as if she had caught me stealing office supplies.

“Tom,” she said. “If you heard the last five minutes, I deny all of it.”

“You heard worse from me fifteen years ago.”

“I heard cleaner profanity.”

We walked toward the elevators. She carried the marker with her by accident, the problem still in her hand.

“You have time this afternoon?” I asked.

“For you, yes. For another listening tour, no.”

“Not a listening tour.”

She looked at me then. She was quick; that was one of the reasons I liked her.

“AI pilot.”

“Year of the Pilot,” I said.

“That’s already a slogan.”

“Russell hasn’t gotten to it yet, so technically it is only a phrase.”

She laughed, but not much.

Russell M. came by my office after lunch with three members of his team and two pages of language. He had heard enough from Victor’s staff to know the pilot was getting a name. Russell could turn an earthquake into badges, manager talking points, and a microsite draft by the time the coffee cooled.

“We want to be careful with the workforce framing,” he said. “Curiosity, confidence, responsible use, skill mobility. We do not want people hearing replacement.”

“They will hear replacement.”

“Of course they will. But we can give managers better words.”

That was Russell at his best. He understood that language did not remove fear; it gave middle managers a sentence to use while fear did its work.

He had Lena’s name on his list too.

“People trust her,” he said.

“Yes.”

“She is not threatening.”

He meant it as praise. I heard the whole sentence.

When Lena came in at four, the sky outside my windows had gone the flat gray Central Indiana gets in November, the kind of sky that makes every office tower look like it owes money. I had asked my assistant to leave the door open because closing it would make the conversation feel ceremonial. Lena closed it anyway.

“You want me to run it,” she said.

“I want you to help us shape it.”

“That is worse.”

“Probably.”

She sat without being asked. I had known Lena for nine years. I had promoted her twice. When her son had appendicitis during a release weekend, I called the hospital and then told her manager that any meeting with her name on it was canceled until she said otherwise. She never forgot that. Neither did I.

That is the trouble with being good to people. The debt runs both ways.

I gave her the clean version first. The board wanted responsible movement on AI-assisted development. Victor wanted no surprises. Mara wanted no headcount promises and no productivity claims we could not defend. Dana needed security controls before source code went near a vendor’s model. The business units needed to see value before they gave capacity.

Lena listened with her hands folded in her lap.

“Do you want a pilot,” she said, “or do you want a story about a pilot?”

I should have admired the question without resenting it. I did admire it. I also resented it.

“I want a pilot that can survive here.”

“Then give me one real product outcome.”

She opened her tablet, not to present but to keep herself honest. “One team with enough authority to move. We pick a feature tied to a real backlog, map the value stream from intake to production, baseline it honestly, and give them ninety days. Tooling, security, architecture, release path, business owner, support owner. Everything in the room. If AI changes the economics, we will see it. If it does not, we will see that too.”

I looked past her to the framed photograph on my wall from a Doverline Youth Program fundraiser three years earlier. I was in a blue blazer, kneeling beside a boy in a dinosaur sweatshirt while he showed me a circuit made from copper tape. He had been worried the LED would not turn on before his mother came back from the restroom.

Small circuits were honest: power, connection, light.

“Which team?” I asked.

“Service parts quoting.”

I let out a breath through my nose.

“Devon’s team?”

“They already proved the shape of it.”

“They proved a shadow version of it.”

“They proved the process is the constraint.”

I did not answer right away. The room was too warm. There was a message from Hannah on my phone about a loan disbursement form and another from Victor’s chief of staff asking whether I could join board prep at seven-thirty the next morning. Nobody pays you seven figures to panic in public, and nobody pays you seven figures to let a capable director walk herself into a fight with no shield.

“What are you asking for?” I said.

“Mandate authority over the pilot teams.”

She wanted three pilots, not one, because one team would be dismissed as a science project before the first status readout. The pattern was clean: one customer-facing feature, one internal workflow, one legacy modernization slice. At least one would touch revenue, architecture review would sit inside the team for ninety days, and Security would have a named person with forty-eight-hour decision expectations instead of a mailbox that swallowed urgency.

“And the business resistance?” I asked.

“You personally resolve it.”

She had come prepared. I could see the notes on her tablet, the few lines she had written for herself instead of a deck. I appreciated that, and appreciating it did not make the answer easier.

I did not say no. No creates a clean surface. People can push against it. People can quote it back to you later. No makes you the obstacle, and there is only so much use in becoming one more obstacle in a building full of them.

“We need to co-create the governance model,” I said.

Her face changed by maybe an inch.

“Tom.”

“I mean it. If this looks like IT carving out exceptions for itself, the business will reject it before the first sprint.”

“Then tell the business it matters.”

“Forced participation creates antibodies.”

“So does fake participation.”

I smiled because she was right and because smiling gave me half a second.

“We need the business to pull, Security designed in from the start, Architecture in the tent, and Finance comfortable that this is a learning investment rather than a hidden productivity target. If we move too fast, we turn the pilot into a referendum before it has enough evidence to defend itself.”

She looked down at the marker still in her hand. Black Expo, chisel tip, borrowed from a room where everyone had known the problem and no one had owned the authority.

“You are going to give me a steering committee.”

“I am going to give you executive sponsorship.”

“And a steering committee.”

“Yes.”

She kept her eyes on me while I gave her the rest of the package: program budget within the charter, candidate teams to be identified, release authority to be defined by work type. Every answer had enough substance to sound like support and enough conditional language to preserve the system around it.

She sat back. “That is no in four different suits.”

I could have laughed with her. I did not.

“It is not no,” I said. “It is how we keep you from being the person everyone blames when a pilot breaks a system they were already scared to touch.”

That landed because it was partly love.

Lena was not naive. She knew Celia would not let central IT near plant scheduling in the middle of a distributor push. She knew Dana’s concerns were real. She knew Legal could not ignore model retention because engineers were excited. She knew a single production incident tied to AI would freeze the program for a year and put her name in rooms she could not enter.

“I am not afraid of being blamed,” she said.

“You should be. Not because you cannot handle it. Because blame at this place is never proportional to authority.”

She looked at me for a long time.

“Is that why you are picking me?”

There are questions you answer with truth and still avoid the truth.

“I am picking you because engineers trust you and because you will tell me when the work is becoming theater.”

“Will you listen?”

“Yes.”

“Will you move?”

I looked at the photograph again. The LED had turned on eventually. The boy had shouted for his mother like he had discovered fire.

“I will move what can be moved.”

Lena nodded once. Not agreement. Inventory.

We spent the next forty minutes sketching the official shape. Lena would be named enterprise lead for the Year of the Pilot. Russell would help with manager messaging and learning language. Dana would nominate a security partner. Architecture, Legal, Procurement, Finance, and two business-unit presidents would sit on the steering group. We would build a pilot charter by January, select teams after the charter, and brief Victor before the holiday break.

It sounded responsible because it was responsible.

It also meant the pilot could not outrun the company.

When she stood to leave, she set the marker on my desk.

“You stole that from the sixth floor,” I said.

“Evidence,” she said. “Do not lose it.”

After she left, Russell stopped by again. He wanted to know if we had a leader.

“We do.”

“Lena?”

“Lena.”

He smiled with obvious relief.

“Perfect.”

I waited until he was gone before I wrote the word on my yellow pad.

Perfect.

Then I drew a box around it, because that is what executives do when they have turned a person into a plan.

Chapter 5: The Charter

January 2027

Executive Play: Charter Away The Authority

The first draft of the pilot charter was eight pages long and had one dangerous sentence in it.

Lena wrote the sentence herself.

Each pilot team will own a feature from intake through production.

It sat near the bottom of page two, plain enough to be missed by anyone hunting for governance language and sharp enough to cut through everything we had been saying since Devon’s demo. No adjectives softened feature. No maturity-stage phrase softened own. No dependency map, council review, or release-governance reference stood after production to give the organization a place to hide.

I read it alone at my kitchen table on the first Sunday in January while Claire made coffee and the house clicked in the cold. The Christmas boxes were still stacked by the basement door because I had promised to carry them down before New Year’s and then let three budget reviews, one security incident, and a Doverline Youth Program board call become my excuse.

“Bad?” Claire asked.

“Good,” I said.

She poured coffee into the mug Hannah had given me for Father’s Day. It said TRUST ME, I READ THE SYLLABUS. Claire had never liked executive documents; she said they sounded like people trying to be forgiven before they committed the offense.

“Good is worse?”

“At Kedran, usually.”

She leaned against the counter with her own mug and waited. Claire did not chase me when I was arranging a truth into a version I could live with. After thirty years she knew the difference between silence and evasion, and most Sunday mornings she was kind enough to let me keep the lesser one.

“Lena wants the pilot teams to own delivery,” I said. “Real features, with release authority attached.”

“Isn’t that the point?”

I looked at the sentence again. “That is the problem.”

The charter process began two days later in the executive conference room on eleven. The room had a long white table, two wall screens, twelve chairs, and a view over the parking garage where snow had been pushed into gray ridges beside the elevators. By eight-fifteen the coffee urn was empty, the fruit tray was untouched, and everyone had a printed copy because the outside advisor liked paper in meetings. He said paper slowed people down in useful ways. I thought it gave consultants a surface to tap with a pen.

The advisor wore a navy suit with no tie and the kind of shoes that looked comfortable only if you were billing by the hour. His firm specialized in enterprise AI operating model readiness, a phrase I disliked the first time I heard it and paid for anyway because Victor wanted external confidence and Mara wanted a vocabulary she could put in a board deck.

He had a gift for making delay look like design.

“What we are building here,” he said, “is responsible acceleration.”

Russell wrote it down.

Lena did not.

Dana K. sat three chairs to my left with the draft open and a blue pen in her hand. Dana was our CISO, and she had the expression of a person who had been invited to another meeting where Security would be blamed for gravity. Her concerns were real: source-code exposure, prompt leakage, model retention, vendor terms, regulated customer data, audit trails. None of those words were fake, and that was what made them useful. A fake concern is easy to defeat. A real concern can be placed wherever authority is trying to stand.

Architecture sent two directors instead of the chief architect. Procurement sent a senior manager who had never met a new tool she could not turn into a comparison matrix. Finance sent Victor’s favorite controller, who spoke rarely and always with the calm of a prison warden. Legal sent counsel who could make “we should be thoughtful” sound like a padlock closing.

Lena brought no entourage. She had her laptop, a notebook, and the same black marker from my office. She put the marker beside the draft as if the sentence on page two might need to defend itself.

I opened the meeting.

“The goal is to leave January with a charter we can brief to Victor, Mara, and the board technology committee,” I said. “It needs to create real learning, and it has to be bounded enough that the organization can support it without turning the pilot into an incident.”

I needed every word in that sentence. Real learning kept Lena in the room. Bounded kept Dana and Legal from reaching for the brakes. Support sounded generous enough that no one had to say permission. Incident reminded everyone that executive courage has lawyers, auditors, insurers, and a quarterly calendar sitting beside it.

The advisor began with principles. The pilot would be human-centered, risk-aware, business-partnered, security-led, outcome-informed, and learning-oriented. Russell liked human-centered. Procurement liked approved tool patterns. Finance liked learning investment. Legal liked risk-aware. Dana liked none of it yet.

Lena waited until the second hour, after the first round of adjective polishing had made everyone feel helpful.

“Before we wordsmith,” she said, “we need to decide whether teams own outcomes or tasks.”

The room settled into the colder kind of attention.

The advisor smiled at her. He smiled at conflict as if it had validated the invoice. “Say more.”

“If pilot teams get fragments of work, we will learn whether developers enjoy the tools,” Lena said. “That may be worth knowing, but it is not the question Kedran needs answered. The question is whether AI-assisted development changes delivery when a team can move a feature through the system: intake, design, build, review, security, release, and support. If all of those pieces remain outside the team, the bottleneck will move, and we will call the result a tool finding.”

Dana clicked her pen once. “Security cannot be outside the team if the team is using external models against source code.”

“Agreed,” Lena said. “Name a security partner for each pilot team.”

“For every team?”

“For every team that touches code.”

Procurement tapped the margin of page four. “We have not selected an enterprise-standard tool.”

“Then the pilot can test within an approved shortlist.”

Finance looked up from the cost page. “What is the spending envelope?”

“Small enough to fit inside Tom’s program budget,” Lena said, “and large enough that we are not pretending volunteers can transform a company after hours.”

I looked down at my copy so nobody could see me smile.

Lena was good in rooms because she did not raise her voice or perform impatience. She kept returning the conversation to the work, and the repetition made people uncomfortable in a way they could not easily punish.

Architecture objected first to the sentence itself.

One of the directors had been at Kedran for twenty-two years and had the soft, weary politeness of a man who had survived every new operating model by asking where the integration pattern was documented.

“The intake-through-production language creates an implication of bypassing established governance,” he said.

“It creates an implication of ownership,” Lena said.

“Ownership has to exist within enterprise architecture.”

“Then put enterprise architecture inside the pilot team.”

The director looked at me, not at Lena. That was how authority showed itself in a room. People answered the person who could make their quarter harder.

I did not rescue either of them.

“Let’s capture both needs,” I said.

The advisor leaned in. “Maybe the sentence becomes: pilot teams will coordinate across the delivery lifecycle in partnership with existing governance bodies.”

Lena looked at him as if he had removed the engine from a truck and called it safer.

“Coordinate is not own,” she said.

“It may be the right verb for this phase,” the advisor said.

This phase was another clean place to bury time.

I recognized the mechanism because I had used it before. Agile had not been fake at Kedran. DevOps had not been fake either. Good people had done serious work with both, and some of that work still paid the bills every Friday night when plants closed orders and batch jobs crawled across old systems. The changes failed to alter the company because the charter always stopped before the money, architecture, release authority, and business-unit operating model. We added coaches, standardized tools, wrapped governance in program language, measured maturity, and let the hard questions become the next phase. My predecessor, whom I had criticized for years, had not been confused. He had understood the building.

Over the next three weeks, the charter became a hotel towel after too many washings: softer, thinner, easier to fold, and less able to do its job.

Dana added a section on source-code handling. I supported it because she was right. No regulated data in prompts. No proprietary code in unapproved tools. Named pilot repositories only. Logging required. Access reviews. Model-retention attestations. A security partner for each use case. The section was practical, and because it was practical it slowed everything with clean hands.

Legal added vendor constraints and language about prompts as possible business records. Procurement added approved-tool language and a comparison step in case any division wanted a different assistant. Architecture added review checkpoints at design, integration, and pre-release. Finance added spend controls, capitalization guidance, and a line requiring any productivity claim to be reviewed before publication. Russell added learning objectives, manager conversation guides, and a confidence pulse.

The advisor added a diagram.

They always added a diagram.

The diagram showed a circle of functions around the pilot team: Security, Legal, Architecture, Product, Finance, Procurement, HR, Business Sponsor. In the middle was a smaller circle labeled AI Delivery Pod. It looked balanced. It looked mature. It also looked like a person trying to run while eight people held the same belt.

Lena kept fighting for her sentence.

We argued about it in conference rooms, on calls, in tracked comments, and once in the hallway outside a board prep session while Victor’s chief of staff waited with a binder under her arm.

“Each pilot team will own a feature from intake through production,” Lena said, reading from her phone. “It is not poetry. It is the minimum viable sentence.”

“It overpromises authority we have not aligned,” I said.

“Authority is what we are testing.”

“Tool use is what the board approved.”

“No,” she said. “Tool use is what we put in the safe version.”

We stood beside a wall of framed photographs from Kedran’s plant safety awards: men and women in hard hats, forklift banners, zero lost-time incident signs, supervisors smiling under fluorescent lights. Every picture was a reminder that speed was not the only virtue in an industrial company. A late feature could embarrass us. A bad release into scheduling or pricing could move steel, labor, inventory, and customer promises in the wrong direction.

“I am trying to keep the work alive,” I said.

“You are keeping the language alive.”

The line stayed with me because it was both unfair and precise.

Midway through January, Evelyn N. asked for the baseline.

It happened on a board technology committee prep call that should have lasted twenty minutes and ran almost an hour. Evelyn had founded a software company before selling it to a larger one that knew how to say synergy without embarrassment. She understood code, incentives, and the smell of a metric chosen after the fact.

Victor was on the call for the first ten minutes. Mara stayed for all of it. I had the advisor in the room with me, off camera, which was a mistake because he kept writing phrases on a legal pad and sliding them toward me.

Evelyn did not ask whether we had a policy. She did not ask whether employees were excited. She did not ask if the tools were safe. She asked for the current lead time by application class.

I had expected it. Expecting a punch does not make your ribs immune.

“We have partial views,” I said.

She moved down the list without looking away: deployment frequency, merge cycle time, defect escape rate, approval wait time, repos retired, teams with end-to-end release authority. For each one I gave the version of an answer that was true enough to survive minutes: inconsistent across environments, repository-level in some platforms but not all, varied by business unit and system type, no single queueing model, retirements countable but active dependency harder.

When she reached release authority, the advisor had the sense to stop tapping his pen.

He slid a phrase toward me: MATURITY BASELINE WORKSTREAM.

I ignored it.

“Low,” I said.

Evelyn waited.

“And not consistently defined,” I added, because I could not help myself.

Mara looked down. She knew what I was doing and why it would work.

“So how will you know whether the pilot improves delivery?” Evelyn asked.

“We will establish baselines as part of the pilot,” I said. “The data is fragmented across acquisitions, toolchains, and delivery models. Some of that fragmentation is precisely what the pilot needs to surface.”

The statement was true, which made it the strongest kind of politically correct statement and more useful than a clean answer. Seven acquisitions had left us with old ticketing systems, inherited source-control organizations, plant systems that exported CSVs to shared drives, and homegrown release trackers that had once been temporary. I could not give Evelyn a clean enterprise baseline because no clean enterprise existed underneath the word enterprise.

The practical advantage was harder to say. Without a baseline, one rough pilot could not become a verdict. Without a verdict, the program could continue. Without hard measurement, we could use story, sentiment, policy maturity, and use-case discovery until the clock gave us better weather.

Evelyn knew enough to hear the empty space.

“Do not let measurement become Phase Two,” she said.

“Understood.”

I wrote it down. Writing a request down in a meeting is not the same as agreeing to let it happen.

The final charter was approved on a Friday afternoon when everyone was tired and the snow had turned to rain. The page count had grown from eight to sixteen, plus appendices, and the dangerous sentence had disappeared so gradually that no one except Lena could point to the hour it died. My assistant had printed the signature page on heavier paper because Victor liked documents that felt final when you turned them over.

The version in front of us looked harmless enough:

Kedran Industrial Holdings
AI-Assisted Development Pilot Charter
Version 1.0 – Board Technology Committee Pre-Read

Date and Context: January 28, 2027. Establishes the Q1 operating charter for the Year of the Pilot following executive review of approved AI-assisted development demonstrations and board direction to pursue responsible learning within existing Kedran controls.

Scope: Up to six nominated technology teams may execute approved AI-assisted development use cases against named repositories, workflows, and internal users. Use cases may include code explanation, test generation, documentation support, developer search, defect triage, support summarization, and other low-risk engineering activities approved through the pilot intake process.

Authority: Pilot teams may use approved AI-assisted development tools, collect delivery observations, propose process changes, and request time-bound support from Security, Architecture, Product, Finance, Procurement, HR, and Legal. Pilot teams will execute approved AI-assisted development use cases in coordination with product owners, security partners, architecture review, and established release governance.

Authority Reserved: The pilot does not grant exception authority over business-unit priorities, annual funding, architecture standards, production release gates, security controls, customer commitments, vendor contracting, workforce planning, or support ownership. Final decision rights remain with the currently accountable executive, business sponsor, or control function.

Exclusions: The pilot will not change pricing logic, quote commitments, plant scheduling, revenue recognition, regulated customer data handling, customer-facing obligations, job architecture, headcount plans, or enterprise productivity forecasts without separate executive approval. No productivity claim may be published, forecast, or used in workforce planning without Finance and Legal review.

Funding: FY27 pilot spending will not exceed $4.8 million from the CIO AI program cost center, including approved tooling, consulting support, enablement materials, security logging, and limited team backfill. Business-unit participation will be absorbed within existing operating plans unless separately approved through the annual funding process.

Success Measures: The pilot will report approved-use-case completion, security and policy compliance, team participation, manager confidence, training completion, baseline availability, cycle-time observations where comparable data exists, defect findings, and narrative case studies for the June Technology Committee readout. Enterprise delivery acceleration, cost reduction, repo retirement, and headcount impact are not success measures for this phase.

Governance: Weekly AI Delivery Pod meetings will manage use-case progress. A biweekly cross-functional working group will resolve intake, tooling, and control questions. The monthly AI Steering Council, chaired by the CIO, will include Security, Architecture, Legal, Procurement, Finance, HR, Product, and named business sponsors. Exceptions will follow standard escalation paths. The Board Technology Committee will receive quarterly progress summaries.

Operating Model: AI Delivery Pods operate as learning environments inside Kedran’s current delivery and accountability model. Teams will continue using assigned scrum masters, agile coaches, sprint planning, daily standups, story estimation, retrospectives, QA validation handoffs, and established release ceremonies unless an exception is approved through standard governance. Product ownership, budget authority, architecture approval, production release go/no-go, incident response, business-unit prioritization, and support commitments remain with existing owners. Pilot recommendations are advisory unless adopted through standard governance.

It was a beautiful document in the way a locked display case can be beautiful: everything had a label, and nothing dangerous could be touched. It moved the pilot into the safe zone by turning authority into a request path and delivery into learning evidence.

Dana signed off because the guardrails were real. Legal signed off because the vendor language was conservative. Procurement signed off because tools would move through the approved channel. Finance signed off because spending stayed inside my cost center and productivity claims required review. Architecture signed off because the checkpoints remained intact. Russell signed off because he had his learning story. The advisor congratulated the room on cross-functional alignment, then underlined a sentence in his own copy as if the document had surprised even him with its wisdom.

Lena did not sign right away.

She sat at the far end of the table with the paper in front of her and the black marker between her fingers. Around us, people packed laptops and made the relieved small talk of a meeting that had removed every sharp corner without leaving fingerprints.

I walked down to her end of the table.

“We preserved the intent,” I said.

She looked at the page. “No. We preserved the program.”

I could have argued. I could have told her that the program had to exist before it could become brave, that Dana’s guardrails would keep us out of a visible incident, that Architecture’s checkpoints would keep the acquired stacks from surprising us, that Finance’s language would protect us from a headcount panic, and that Victor would never tolerate a pilot that smelled like a workforce promise in the same quarter he had to brief margin pressure.

All of it was true, and all of it was arranged in my favor.

“Sign it anyway,” I said.

She looked up.

“That is the first direct thing you have said all month.”

“Maybe I was saving it.”

She signed.

The marker left a small black dot on her thumb. She rubbed it once, then gave up.

When the room emptied, I gathered the extra copies and put them in the recycling bin myself. My assistant would have done it, but I wanted a minute with the table cleared, the chairs pushed back, and the rain tapping the glass. The charter was responsible and thoughtful, the kind of document that could keep us out of obvious trouble while making the most important lesson impossible to reach.

It had also taken the one sentence that might have taught us a dangerous lesson and made it premature.

Chapter 6: The Pilot Year

2027

Executive Play: Use Volunteers To Contain Evidence

Lena asked for three teams.

Not six. Not a portfolio. Three teams, each with real work and enough authority to move it: Devon M.’s service-parts quoting work rebuilt properly, Mei S.’s warranty triage work tied to actual service outcomes, and one modernization slice inside Celia N.’s Components Division where the pain was close enough to money that no one could mistake it for training.

She came to my office on a Tuesday morning in February with a single page. I respected that. The more important the ask, the less paper it should need, and Lena knew better than to bury a decision under a deck full of blue boxes.

“Give me those three,” she said, “and I can give you useful evidence by June.”

I read the page though I did not need to. A good one-page request is dangerous because it removes the comfortable work of asking the author to simplify.

“Celia will not agree to distributor ordering in February,” I said.

“She will if you ask.”

“No. She will say yes to my face and send someone from her staff to make the scope harmless.”

“Then do not let her.”

Some people think executive authority is a hammer you keep in a drawer because you lack courage. Sometimes it is. More often it is wired to the fire alarm, the sprinkler system, payroll, three banks, and the operating plan everybody signed before they knew what the year would require. Celia’s division was in the middle of the distributor push that made their year. Pricing logic, plant scheduling, order commitments, rebate calculations, regional exceptions, and two acquired platforms nobody had fully integrated all ran hot through March. A bad week there did not become a learning. It became missed shipments, field escalations, angry distributors, and Victor asking why the AI pilot had touched the one place everyone knew not to touch before quarter close.

Lena knew all of that. She also knew the difference between a reason and a permanent condition.

“If we avoid every place with pain,” she said, “we will prove the medicine is safe for healthy people.”

“We are not avoiding pain. We are sequencing risk.”

She sat down then, tired of standing over a man who could agree with her indefinitely.

“Tom.”

“Yes.”

“You are going to give me volunteers.”

I did not answer quickly enough.

She set the page flat with two fingers. “Six volunteer teams. Different readiness levels. Low-risk work. Good people. Lots of enthusiasm. Mixed results. Then everyone gets to say the pilot surfaced learnings.”

It would have been easier if she were wrong more often.

“Volunteers matter,” I said.

“So does authority.”

“Volunteer teams reduce resistance.”

“They reduce evidence.”

I looked past her to the office door. My assistant had left a folder clipped to the handle, which meant a signature was due before lunch and someone had already waited through two reminders.

“I am giving you a path the company will accept,” I said.

“No,” Lena said. “You are giving the company a path it can digest.”

The call for volunteers went out the next week.

Russell’s team wrote it in warm language: Kedran builders, responsible AI, learning by doing, the Year of the Pilot. Managers were invited to nominate teams interested in approved AI-assisted development patterns for bounded use cases. The phrase bounded use cases appeared four times in nine paragraphs. I asked Russell to cut two of them. He cut one.

We held the information session in the auditorium and online because Kedran was too large for any one room to tell itself a story. The back rows filled with engineers who had brought laptops and managers who had brought expressions of qualified support. Remote attendees blinked in from plant offices, home desks, conference rooms left over from an acquisition, links forwarded from another division, and one old chat workspace nobody in central IT had managed to kill without starting a governance argument. We wanted AI to change software delivery, and half the company still needed four collaboration tools to schedule a meeting.

Lena described the pilot honestly enough that the old pinch of admiration and concern came back.

“This is not training by another name,” she said. “We are going to learn where the tools help, where they do not, and where our own system prevents the benefit from reaching production.”

Dana walked through the security guardrails from the side screen. Approved tools only. Approved repositories only. No regulated data. No customer secrets. No model paste-and-pray. She did not use that last phrase, though I wrote it in my notebook because I liked it.

Then Devon raised his hand from the fifth row. He wore a West Lafayette quarter-zip because Devon appeared to own no clothing that did not involve West Lafayette, which made me like him and distrust his judgment in equal measure. I had gone to Bloomington. Some differences cannot be bridged by governance.

“Can our team volunteer with the quoting work from the demo?” he asked.

Half the room turned toward me. Lena did not. She kept looking at Devon.

“You can submit it,” she said.

She had given the right answer. It also meant I would spend the afternoon making sure the answer became no without becoming no.

After the session, Devon came up to me while people were closing laptops and checking for coffee.

“Tom. We can do it clean this time. No shadow access. No field-rep workaround. We know what we did wrong.”

“I know you do.”

“Then let us put it in the pilot.”

He was not arrogant, which made him harder. He was a staff engineer who had seen a door open and did not understand why everyone senior had immediately started discussing the hinges.

“Submit a slice that does not touch live quoting,” I said. “Decision support, search, explanation. A feature that helps service reps without changing price, promise date, or order flow.”

“That is not the thing.”

“It is the thing the company can let you touch in February.”

He looked disappointed, then embarrassed by being disappointed in front of me. I hoped, for his sake, that he would learn somewhere smaller how executives could be warm while denying you the thing you came for.

“What you built mattered,” I said.

“Then why are we hiding from it?”

I could have told him we were not hiding. I could have talked about Celia, Dana, Victor, Mara, the board, quarter close, distributor calls, audit trails, release governance, and a run of acquisitions still welded to our ribs. I could have told him large companies do not change because a young engineer is right in a room.

“Because proof needs a place to land,” I said.

He nodded as if that meant progress. Maybe it did. Maybe I had become fluent in sentences that sounded like bridges and worked like fences.

The nominations came in over ten days. Twenty-seven teams applied, which Russell treated as a sign of cultural readiness and I treated as a sign that managers had learned to volunteer for work that might later become mandatory.

Lena and I reviewed the list in a small room near the security floor because Dana had a gap between meetings and refused to walk upstairs for what she called “another enthusiasm harvest.” Mei S. joined by video from a plant conference room, her camera aimed at a whiteboard full of defect categories and the corner of a vending machine.

Mei managed a warranty and service platform team. She had good engineers, a brittle backlog, and the careful patience of someone who spent most of her week explaining why small changes in service workflows became large arguments in finance.

“I can give you four engineers at fifty percent for six weeks,” she said.

Lena shook her head. “That gives us two engineers.”

“It gives you four names,” Mei said. “Which matters in our resource model.”

Middle managers at Kedran carried transformation like weather. They were praised for supporting the future and graded on the old commitments.

Celia did not attend our review. She sent a deputy who had the smooth face of a man trained never to create a quote. He thanked us for the opportunity, praised the responsible approach, and said Components was excited to participate in areas aligned with the distributor push.

“Plant scheduling?” Lena asked.

The deputy smiled. “Not in this phase.”

“Pricing logic?”

“Not in this phase.”

“Distributor ordering?”

“We would recommend adjacent enablement.”

“Meaning?”

“Knowledge retrieval for support documentation. Possibly test generation for non-production components of the distributor portal.”

Lena looked at me.

“That seems like wise sequencing,” I said.

Business-unit presidents at Kedran refused things this way: fully supportive, excited to partner, not in this phase. No one had to say no if the calendar, operating plan, and scope language could say it for them.

By the end of the week, we had six teams.

Only two looked anything like the evidence Lena had asked for: Mei’s warranty triage work and Devon’s service-parts decision-support slice, carefully severed from live quoting. The other four sat squarely in the safe zone: documentation, internal workflow cleanup, test generation far from release authority, and legacy code explanation that involved reading old code rather than changing it. On the steering slide, the portfolio looked balanced by risk, function, and business unit. In practice, it had been balanced to make sure no single team could prove too much.

Each team also came with the furniture Kedran had acquired during the Agile years: a scrum master, an agile coach assigned at least part-time, mandatory standups, story points, retrospectives, and a QA handoff that still sat between the people who built the work and the people allowed to say it was ready. Engineers argued, uselessly, that estimating a discovery spike against an AI tool with Fibonacci numbers did not make the work more predictable. They were right, and their rightness did not survive the operating model.

The kickoff was held on the last Monday of February in the training room on four. Lena stood at the front in a black sweater, sleeves pushed up, marker in hand. Dana sat near the door with her laptop open. Russell’s team had made table cards that said PILOT TEAM in Kedran blue.

I moved through the room before we started. I asked one engineer about his father’s knee surgery. I asked a QA lead how her daughter liked the small college. I thanked Mei for making capacity. I told Devon I was glad he was there.

These were not tactics. I cared. That made the rest harder to look at directly.

“You are here because you volunteered,” Lena said. “That matters. It means you are willing to do real work in public before the company knows exactly how to support you.”

People nodded. Volunteer courage feels clean from inside the room. Nobody has forced you. Nobody has promised it will work. You raise your hand, and for a moment the system seems less heavy because your hand is in the air.

The first month was mostly permission logistics. Lena needed tool access and got Procurement asking for a vendor comparison. She needed source-code policy and got Dana, Legal, and three model-retention memos. She needed real backlog items and got documentation cleanup, test scaffolding, and an internal reporting tool no customer would ever see. She needed engineering capacity and got names at twenty percent. At Kedran, twenty percent meant a person attended your meetings and did their real work at night.

I never told her no.

I told myself a direct no would create a villain and a martyr, and Lena was too good to waste as either. A direct yes would move the work into the unsafe zone, where Procurement, Security, Finance, Architecture, and the business units would discover at the same time that a pilot had become an operating-model referendum. My job was to keep her moving where the organization could see courage without having to surrender power.

In April, she brought me a one-page request for three business-unit commitments. She had printed it because she knew I treated paper with more respect than attachments.

“I am not asking for the factory floor,” she said. “I am asking for work that matters.”

The football helmet on my credenza faced the window. A photo from a Doverline Youth Program gala sat beside it, Claire and me with our badges crooked, both of us tired and happy. That morning I had helped the program director find a vendor after their registration system failed during enrollment week. The call had given me a cleaner hour than most of my calendar did.

“Industrial Equipment may give you the service portal,” I said.

“If you ask him.”

“I will talk to him.”

“Will you tell him he has to make capacity?”

I tapped the request with my pen. “That is a different conversation.”

“It is the same conversation.”

She was right, of course. In a small company, the same leader could decide priority, team, release, and risk. At Kedran, those decisions lived in different rooms and defended themselves under different names.

The business-unit calls went exactly as I expected. Industrial Equipment offered knowledge-base search but would not move a revenue commitment. Celia approved documentation around pricing exceptions and nothing that smelled like approval flow. Controls offered legacy code access if Dana signed off, Legal signed off, Architecture signed off, and Aaron B. had veto rights over every repository. It was permission shaped like a hedge maze.

When I summarized the responses for Lena, I used the language executives use when the content is disappointing and the tone must not be.

“Industrial Equipment is open to knowledge-base search. Celia will support documentation around pricing exceptions. Controls is willing to explore access with guardrails.”

Lena sat across from me and did not write anything down.

“Those are the softest possible versions of the requests.”

“They are available versions.”

“Because nobody had to move.”

“Because revenue leaders need confidence before they put the pilot near commitments.”

“How do they get confidence if the pilot never touches commitments?”

I had an answer. I always had an answer, which was one of the things that made me good at the job and less proud of it than people thought.

“Through sequencing,” I said. “We build the practices, prove the controls, and earn the right to move closer to production.”

She looked at the framed photo behind my desk, then back at me.

“I spend half my week asking people to let us learn,” Lena said.

“That is where we are.”

“No. That is where you are keeping us.”

Irritation moved through me, quick and private. I had made her visible. I had protected her budget. I had stopped Russell from turning the pilot into a badge program before it had any substance. I had given Dana cover to approve a bounded policy. I had absorbed Victor’s impatience and Mara’s questions about workforce implications. I had done a dozen things Lena never saw, and the fact that she was right did not make me enjoy hearing it.

I also had not forced anyone to move.

That evening, I drove to a neighborhood cleanup on the near north side. I was late and still wearing dress shoes, so I ruined them in wet mulch and pretended not to care. A Kedran engineer from the platform group told me his father had started chemo again. I stood beside a pile of trash bags and asked the questions you ask when there are no good questions: which hospital, which doctor, how his mother was sleeping, whether his manager knew.

On the drive home, I called my assistant and told her to move his Friday review.

I was not a bad man. I knew that much in the way he answered the phone two days later and said, “Thank you, Tom,” before I had finished explaining. I knew it in the notes from program staff, the scholarship breakfasts, the quiet checks, and the calls I returned when no one with a larger title would.

I also knew I had learned how to make a system say no while I remained the person who cared.

The next steering meeting was in Conference Room 14B, which had a screen that tinted everyone blue. Lena presented the pilot status with no drama. Tooling access was partially approved. Vendor terms remained under review. Security policy had a bounded path. Business-unit work was being scoped. Engineering allocation was uneven. Risk review was ongoing.

Dana asked for model-retention language to be attached to the team charters. Finance asked whether pilot labor was being tracked as program cost or business-unit contribution. Architecture asked whether generated code would trigger additional review gates. Legal asked whether prompts were business records. Russell asked if we wanted a recognition moment for participating teams.

By the end, the pilot had not been rejected by anyone. It had simply acquired more gravity.

Afterward, Lena waited until the room emptied. Dana lingered by the door, then caught my eye and left.

Lena closed her notebook.

“Do you want this to work?”

I could have said yes. It would have been easy and mostly true.

Instead I gave the answer that had become my shelter.

“I want it to work the right way.”

Lena nodded once, not because she agreed, but because she finally understood the size of the room I had put her in.

Chapter 7: The Success Memo

Late 2027

Executive Play: Declare Success Without Delivery Metrics

Devon M. was kneeling under a table with a power strip in his hand when I arrived for the July pilot review.

Lena had scheduled the meeting in the old product lab on the fourth floor because she wanted the teams near the artifacts. The lab still had gray carpet from a renovation nobody could date, movable tables, monitors on carts, and a tray of muffins that had appeared without ownership. At Kedran, food and responsibility often arrived by the same route: anonymously, with a purchase order attached somewhere downstream.

“This is not in your job description,” I said.

“Most useful work isn’t.”

He said it lightly enough for the room, but there was a wire in it. Devon had become famous inside a circle small enough to hurt him. He had built the twelve-day quoting demo, watched the company treat it as a controlled learning artifact, and then returned to normal sprint planning as if nothing had happened except that more people now knew his name.

“Still West Lafayette underneath?” I asked, nodding at the black Kedran fleece over his shirt.

He grinned. “Always.”

“That is the sort of persistence organizations should monitor.”

He laughed, and for a moment we were just two men trying to keep a meeting from beginning with a dead monitor. I asked about his mother in Columbus, who had been in and out of the hospital that spring. He looked surprised that I remembered and told me his sister was doing most of the paperwork.

“Tell her to call me if she needs help with the benefits maze,” I said.

I meant it. That was the uncomfortable part then, and it remains the uncomfortable part now. I was not pretending to care about Devon’s mother so he would tolerate what the company did to his work. I cared about Devon’s mother, and I was also going to let the company do what it did.

By then, the pilot had produced enough evidence to disappoint everyone with a clean conscience. Failure would have been easier. A failed program gives you a closed file, a respectful postmortem, and one slide about lessons learned before everyone goes back to the work that was waiting anyway. The Year of the Pilot had helped real people do real work, which meant nobody could kill it without sounding careless, and it had also shown, with the small cruelty of practical evidence, that the coding tool was not the thing holding Kedran in place.

Lena came in carrying a canvas bag heavy with printouts. She had cut her hair shorter since spring, and the new cut made her look less like someone assigned to lead a pilot and more like someone who had survived one.

“Please do not help with AV,” she said.

“I am excellent at plugging things in.”

“You are excellent at making people think plugging things in is leadership.”

I laughed because it was funny, and because she had earned the right to say it in front of witnesses.

The first team to present was Mei S.’s warranty and service group. Mei stood beside two engineers and walked us through the claims triage tool they had built in four weeks. The old process made a service rep search three systems, read the claim text, hunt for matching repair bulletins, and guess which escalation queue would accept ownership without sending the case back after lunch. The pilot version pulled the claim text, suggested likely categories, surfaced similar cases, and drafted a first response for a human to clean up.

It was ugly in the Kedran way, gray buttons and table borders and labels written by people who had never been asked to sell anything. It also worked.

“Before the pilot, we estimated twelve weeks for a first usable release,” Mei said. “We got an internal version in four.”

Her second slide had numbers on it because Mei knew the Industrial Equipment president would not accept adjectives. The workflow handled roughly twenty-nine hundred claims a week. Fourteen percent bounced at least once between service, warranty, and engineering because the first category was wrong or incomplete. The pilot version cut first-pass review by about three minutes on the sample set, which sounded small until you multiplied it across the month-end pile.

The Industrial Equipment president stood at the back with his arms folded over a vest from a plant visit.

“Production?”

Mei glanced at Lena, then back at him. “Not yet. We need integration approval and queue ownership confirmed.”

He nodded as if the answer had always been waiting there. It had.

Lena grouped the rest of the evidence the way executives can hold it: test generation, documentation, legacy code explanation, and low-risk workflow automation. Regression coverage on one brittle distributor portal component moved from forty-two percent to sixty-eight, and a generated test caught a discount edge case before it reached QA. The engineers were proud, and they deserved to be proud.

Then someone asked whether defect escape had dropped.

“Too early to say,” the team lead said, after a pause long enough to let the room know he wanted a better answer.

The legacy code story was useful for a different reason. A team had spent six weeks trying to use the tool against a repository that would not build on any current laptop. The readme pointed to a retired build server, and Aaron B. had to explain that the repository only looked dead because the part people cared about ran through a nightly file exchange nobody wanted to rename, replace, or admit was still load-bearing.

The AI assistant could explain functions once the files were open. It could suggest tests. It could even translate a few thousand lines of old Java into code a younger engineer could read without swearing at the monitor.

None of it could be merged.

One engineer put the finding cleanly: “The tool is useful after the archaeology. It does not do the archaeology.”

Another team generated code for a small reporting feature and then discovered the ownership boundary ran through the middle of the change. Their product owner owned the screen. A platform team owned the data service. Security owned the access pattern. QA owned the final validation queue, and the queue was nine business days long because the same testers were also covering quarter-end fixes. Nobody owned the customer waiting for the report. The code was fine, which made the approval path look worse.

On one table, beside the laptops, someone had left the old Agile maturity scorecard because the scrum masters were using the same template for pilot ceremonies. Standup health was green. Story-point hygiene was yellow because engineers kept refusing to estimate research as if it were a bolt order. Retrospective completion was green. QA handoff compliance was green. It was all real, and none of it answered the question the demos had put in the room.

Victor arrived twenty minutes late with Russell beside him and two communications people behind them. He could make a person feel seen for ninety seconds and then forget the sentence by the elevator; that was not malice, exactly, but executive attention is a narrow pipe.

“Show me the good stuff,” he said.

Lena showed him the good stuff, and she made him stand close enough to see the rest. She moved through the internal tool in four weeks, the test coverage increase, the documentation gains, the legacy code explanations, and the comments from developers that Russell copied with a pinched mouth.

I got three days back.

I finally understood a module no one owns.

This would matter more if we could merge faster.

The tool made the coding part less miserable and the waiting part more obvious.

Victor liked the first half of every sentence.

“This is encouraging,” he said.

“It is,” Lena said. “But the pattern is clear. Coding accelerates first. Review does not. Release does not. Funding does not. Ownership does not.”

Russell looked down at his notes. He had been trying all morning to find a phrase that would carry that without cutting anyone.

“So the bottleneck moves,” Victor said.

Lena did not take the gift. “No. We revealed where it already was.”

The communications people stopped typing.

After the review, Lena and I went back to my office with her draft report. She had written it herself, without consultant polish or internal communications perfume. The title was plain enough to make trouble:

Pilot Findings: AI-Assisted Development Improves Local Throughput, Enterprise Bottlenecks Remain

“You know that title is a board fight,” I said.

“It is a finding.”

“It is also a conclusion.”

“Findings lead to conclusions.”

“At universities.”

The report was good. It said the tools improved individual and team-level work where environments were stable, ownership was clear, and review paths were known. It said documentation and test generation showed the strongest early benefit. It said legacy systems created setup friction that consumed pilot time. It said business-unit capacity constraints prevented meaningful production commitments. It said Kedran lacked reliable baseline metrics across lead time, deployment frequency, merge cycle time, approval wait time, and defect escape.

It did not accuse anyone because the table of facts was accusation enough.

“If we send this as written,” I said, “Evelyn will ask why we are not reorganizing around end-to-end product teams by Monday.”

“Would that be bad?”

“It would be impossible.”

“Impossible or inconvenient?”

“At Kedran, those words are cousins.”

She looked past me toward the window. Below us, a delivery truck backed into receiving while two workers in orange vests guided it with hand signals. They were careful and exact, the kind of workers who make a headquarters possible without ever being invited upstairs to explain their method.

“What do you want the report to say?” she asked.

“The truth.”

“Tom.”

“The truth in a form the company can use.”

“That means a smaller truth.”

My phone buzzed on the desk. Hannah had sent a photo of anatomy flashcards and the message, I have become a person who highlights highlighters. I smiled before I could stop myself, then put the phone face down.

“Here is what I think we can defend,” I said, and took a pen to the top page.

Strong developer sentiment. Significant use-case discovery. Early productivity indicators. Improved test coverage in participating teams. Important security and operating-model learnings. Phase Two metric development.

Lena watched the pen move as if it were cutting fabric.

“You are taking the delivery sentence out.”

“I am moving it into operating-model learnings.”

“That is where findings go to lose weight.”

“It keeps the door open.”

“The door to what?”

“Scale.”

She leaned back. “Scale without authority.”

I hated how much of the conversation she could now skip.

The board technology committee met two weeks later in a room with better coffee and worse patience. Evelyn N. came in person, which meant she wanted to ask questions without giving anyone the mercy of a frozen video screen. Victor opened with the CEO version of restraint: encouraged, not declaring victory, learning responsibly. Mara sat beside him with her tablet flat on the table. She had already marked the spend as a learning investment in the quarter’s internal forecast: seven million dollars direct, thirteen million in labor and opportunity cost if you counted people’s time honestly, twenty million all in. None of that appeared on the first page.

I presented the results.

I told them about service claims triage, test coverage, documentation improvement, developer confidence, and the policy gaps in source-code handling, approved tool patterns, vendor terms, and operating-model dependencies. I did not say enterprise delivery had improved, because it had not, and even I had limits on what I would let a slide imply.

Evelyn waited until slide fourteen.

“Did lead time improve?”

The room changed by inches. Victor stopped moving his pen. Mara looked at me instead of the slide.

“At the team level, we saw early indicators in selected workflows,” I said.

“That is not what I asked.”

No, it was not. Evelyn wanted enterprise lead time, deployment frequency, approval wait time, and teams with authority to ship without external boards. I gave her the truth in the order least likely to start a fire: sample too small, repository data inconsistent, application classes fragmented by business unit, no clean baseline across participating teams, authority outside the pilot scope.

“So what improved?”

I thought of Mei’s face when the claims tool found the right cases, the engineer saying the tool did not do the archaeology, and Lena in my office watching a hard truth become a phrase that could get through committee.

“The coding work improved where the surrounding system allowed it,” I said. “The pilot also identified the conditions required for that improvement to scale.”

Evelyn tapped her pen once on the table.

“That is a careful answer.”

“It is a careful program.”

Victor stepped in before care could turn into indictment. “The important thing is that we are building capability without creating operational risk.”

Mara added the sentence everyone needed her to add. “And without making premature commitments about productivity or workforce implications.”

No headcount story. No margin promise. No analyst bait. The hardest measurement had been moved far enough into the future to give the room its oxygen back.

By November, Mara and I were writing the success memo in a finance conference room with no windows. She said windows made people perform optimism. I thought that was unfair to windows, but Mara had earned her theories.

She had the first draft open on the wall screen.

“I started with the facts nobody can argue with,” she said.

Three hundred twelve engineers engaged.

Forty-eight use cases identified.

Nineteen demos completed.

Eighty-seven percent participant satisfaction.

Eleven policy gaps resolved.

Four approved tool patterns.

One enterprise AI development playbook drafted.

No clean delivery metric.

The last line was not in the memo. It sat in my notebook because I still had enough private honesty to write down what the official language had to step around.

“Twenty million all in,” Mara said.

“Do we put that in?”

“Not on page one.”

“Page three?”

“Appendix.”

“Coward.”

“Adult.”

The direct spend was just over seven million dollars: vendor trials, the outside advisor’s firm, training material, security review, legal time, tooling, environment work, and travel disguised as collaboration. The internal cost was uglier. Engineers sat in pilot workshops instead of finishing committed backlog. Managers prepared demos. Security and architecture took review cycles. Business-unit people attended alignment meetings they had not asked for. Lena’s team burned weeks on coordination, translation, and morale. If you counted labor honestly, and Mara did when no one was asking her to speak publicly, the opportunity cost was about thirteen million.

“No macro benefit,” Mara said.

She had not typed it. She was looking at me.

“None I can defend,” I said.

Mara drew a small table on the legal pad between us because she trusted rows more than adjectives.

Enterprise lead time: no comparable movement.

Revenue systems: no material release-frequency change.

Acquisition integration backlog: unchanged except for two documentation wins nobody could book.

Operating margin: no attributable signal.

Defect escape: too noisy to claim, and Legal would hate the word claim anyway.

“That means no.”

“That means not enough data.”

She smiled a little. “You are already in meeting voice.”

At noon, Mara took a Treasury call and I checked my phone under the table like a teenager. Lauren had sent a sonogram photo. The baby looked like a weather system. I kept the gray blur on my screen longer than I meant to.

There was a time when I thought making enough money would reduce the number of claims life could make on me. That was before mortgages, tuition, parents, adult children, taxes, charity tables, the quiet cost of being the person who is expected to answer, and the Ludington cabin my grandfather built, closer to a fishing shack than a vacation place, with soft porch boards and a roof that needed attention before it needed charm. I had made more money than my father could have imagined. I still counted vesting dates.

Mara came back in and caught me looking.

“Grandchild?”

“Apparently the baby has my nose.”

“The baby has fog.”

“That is what I said. I was overruled.”

She sat down and softened for half a second, which was how Mara showed mercy.

“You need three more years, right?” she said.

I looked at her.

“At least I am not the only one who can do math,” she said.

The hardest paragraph was measurement. It had to say we did not have mature metrics without sounding as if we had spent a year avoiding them. It had to make the absence of a baseline into a discovery rather than a failure of basic program design.

My first version was too honest:

The pilot confirmed that Kedran lacks consistent delivery metrics across repositories, business units, and application classes, limiting our ability to quantify enterprise throughput impact in Year One.

Mara read it once.

“True and indicting.”

Her version had the smoothness of a table with no edges:

The pilot surfaced the need for a scalable measurement foundation that reflects Kedran’s varied application portfolio, acquisition history, and risk profiles.

“That sentence has no blood,” I said.

“Exactly.”

We compromised badly, which is how most executive language is born:

The pilot surfaced important measurement gaps across Kedran’s varied application portfolio. Closing those gaps will be a core component of Phase Two, enabling leadership to track productivity, quality, and delivery impact at scale.

It was defensible. It was true enough. It moved the question that mattered into a future where I could manage the room before the room managed me.

Lena came at three.

I had asked her to review the memo before Victor saw the final. That was more than politics. She deserved to see what would be said in her name, and she had carried the year with a smile that grew thinner each quarter.

She sat with the printed memo and a blue pen. Mara and I let her read.

“You removed the bottleneck finding,” she said.

“It is in the operating-model learnings section.”

“The words are there. The finding is gone.”

She turned a page. “You say 312 engineers engaged. That includes people who attended one tool policy session.”

“It is participation.”

“It is attendance.”

“The appendix breaks down activity types.”

“The board will not read the appendix.”

She turned another page, and the pen stopped beside the measurement paragraph.

“You say Phase Two will develop scalable metrics. We asked for baseline metrics in January.”

“We did not have reliable data.”

“We still do not.”

“Which is why measurement becomes part of the next phase.”

She looked at me then. Not angry. Clear.

“You understand how perfect that is, don’t you?”

I folded my hands on the table because a moving executive looks guilty even when he is only tired. “I understand how necessary it is.”

“We proved training was possible and avoided proving delivery was possible.”

Mara looked down at her tablet. She had the decency not to rescue me.

“Lena, I am trying to keep this funded.”

“Funded for what?”

“For scale.”

“Without authority.”

Mara finally spoke. “The board will not approve a reorganization off six teams and no clean baseline.”

“I am not asking for a reorganization,” Lena said. “I am asking for one sentence that says throughput did not improve because the system outside the teams did not change.”

Mara looked at me.

I knew I could give Lena that sentence. It would be accurate, and it would also become the sentence Evelyn underlined, Victor minimized, Mara hedged, the business units resisted, and Russell translated into training language by breakfast. Maybe the fight needed to happen there. Maybe a braver man would have written it and let the room decide what kind of company Kedran wanted to be.

I thought of the fifteen hundred people under me, Hannah’s tuition bill, Lauren’s foggy sonogram, Claire at the kitchen counter asking whether I was protecting the company or protecting the life, and my own number, the one I did not say out loud because it was easier to look principled when nobody could see the spreadsheet in my head.

“Not in the board memo,” I said.

Lena uncapped the blue pen and wrote one note in the margin beside the measurement paragraph:

This is the decision.

Then she stood.

“Thank you for letting me review it.”

“Lena.”

She paused at the door.

“I mean that.”

“I know you do, Tom. That has never been the problem.”

Victor liked the memo. He checked the things that mattered to him: no productivity commitment, no workforce implication, enough progress to show momentum, and a sentence he could use when Evelyn asked about lead time.

“No surprises,” he said.

“No surprises.”

The board approved Phase Two funding in principle before Christmas. The Year of the Pilot was officially a success. Communications pulled three sentences for the internal site. Russell’s team made a slide with the number 312 in large type.

I sent Lena a handwritten note after the board meeting. I thanked her for her courage, her patience, and her leadership. I told her the company was better because she had pushed it. Every word was true. I chose good stationery because my mother had raised me to believe a note mattered more when it had weight.

Six weeks later, she came to my office and closed the door.

I knew before she sat down.

“I am resigning,” she said.

She was going to a smaller systems company outside Indiana, about eight hundred people, where the CEO wanted her to rebuild delivery around three product lines. The title was chief product and technology officer.

“Do you trust the CEO?” I asked.

“He gave me release authority in writing.”

I looked at the football helmet on my credenza and the stack of board materials that would become the next year of my life. Somewhere in the building, people were already saying Phase Two as if it were a destination rather than a delay.

“Good,” I said.

At the door, she turned back.

“For what it is worth, I think you did want it to work.”

I nodded.

“Just not enough to make it dangerous.”

After she left, I sat alone for a while. Outside my window, evening settled over the parking lot, and headlights moved in patient lines toward the exits.

The pilot had succeeded. The memo said so.

Lena had succeeded too. That was why she had to leave.

Chapter 8: Chip B.

Early 2028

Executive Play: Hire A Harmless Champion

Lena’s official resignation email arrived three minutes after she left my office, six weeks after the board memo went out.

She wrote it kindly, which made it worse. There was no theatrical paragraph about values, no accusation, no inventory of promises I had made without making. She thanked me, said she had learned a great deal, and confirmed she had accepted the role she had just described to my face: a smaller company where the CEO wanted her to build the engineering operating model from the ground up.

I sat with the email open longer than the words required.

My assistant knocked once and stepped in with a condolence card for me to sign. One of our database administrators in Lafayette had lost his wife after a long illness. I had met her twice.

I signed the card slowly.

“Do we have the son’s name?” I asked.

“Evan,” she said. “Junior. Studying accounting in Bloomington.”

“Call the Kedran Foundation. Ask whether the employee hardship scholarship can cover tuition and mandatory fees through graduation.”

“Through graduation?”

“Through graduation.”

The foundation office called back in eleven minutes. Tuition and mandatory fees, not room and board. The letter would come from the Kedran Foundation, not from me, which was exactly right.

“Do it today,” I said.

Then I texted Mara.

One of our DBAs lost his wife. Son is Evan. Junior accounting student in Bloomington. Foundation is covering tuition. If Finance has internship room, he could use a real summer.

Mara answered three minutes later.

Send me his email.

Twelve minutes later, Mara had sent Evan a note. She led with his mother, not the job, then told him Finance would hold a paid internship for him after the semester if he wanted it. No pressure, no deadline, no cheerful sentence asking him to turn grief into opportunity.

Mara could take ten million dollars out of a budget without blinking. She could also spend three sentences kindly when the room required it.

After that, the office went too quiet, or maybe I did. I thought of Evan in Bloomington, at the same school I had gone to, learning in one afternoon that tuition and summer work had changed shape because people he had never met had decided his father’s work mattered.

This was the part I was good at: a human problem with a name, a fund code, a phone call, an offer no one could turn into an operating-model debate.

When the card came back, I wrote Evan’s name, his mother’s name, and a sentence telling his father that Kedran would stand with the family. I did not make it grand. Grand language is rude around fresh grief.

The good things were not camouflage. I did call people. I sent flowers. I remembered names and bad insurance forms. I had patience for human grief and less courage for institutional pain, which is not a contradiction so much as an executive biography.

After the card left, I went back to Lena’s email.

The official read was simple. The pilot had given her visibility, visibility had brought a better job, and Russell would write language about her leadership in establishing Kedran’s responsible AI development foundation. Victor would say Kedran grew leaders. Communications would find a photograph where she looked optimistic near a whiteboard.

All of that would be true enough.

The private read was cleaner. Lena had wanted the work to be true. I had needed it to be safe.

I was not angry at her. Anger would have been cleaner. What I felt had the sourness of envy and the manners of admiration. If I had been forty, with fewer tuition schedules, fewer vesting dates, and a house that did not yet feel like proof of a life, I might have done exactly what she did: taken the smaller company, the written authority, the chance to build the thing instead of narrating around it. But I was not forty. This was my last rodeo, and Kedran was still moving under me. If I kept my seat, kept my hands light, and did not let the room see how badly I wanted the gate to open, I could ride it out clean.

By ten o’clock, Russell M. was in my office with a legal pad, two coffees, and the face HR leaders make when they want to mourn and recruit at the same time.

“She’s a loss,” he said.

“She is.”

“People are going to read meaning into it.”

“People read meaning into the cafeteria soup schedule.”

Russell smiled because he knew I was trying to be kind to both of us.

He had already drawn three boxes on his pad: internal leader, external advisor, hybrid model. Each box had arrows under it. Russell was very good at arrows. They made uncertainty look like sequence.

“We can promote someone from DevEx,” he said. “Continuity story. Lena built the foundation, next leader scales the learning.”

“That puts somebody in the same chair with the same missing levers.”

“We could give them more levers.”

I looked at him.

Russell took a sip of coffee and put the cup down. “Or we could not.”

Russell could turn almost anything into learning language, but in private he knew where the bones were buried and did not always pretend they were roots.

“We need Year Two to feel different,” I said.

“Scale,” he said.

“Scale without promising throughput.”

He wrote it down, frowned, and crossed it out. “Capability.”

“Better.”

“Capability at scale.”

“There it is.”

The board had accepted Year One because the pilot had been bounded, responsible, and full of activity no one could cleanly dislike. But pilot language expires. It is a bridge word, useful because it lets everyone admit they do not know yet. By the second year, a company like Kedran needs shape: cohorts, rooms, calendars, certificates, leader videos, and a dashboard that turns effort into rectangles.

The artifacts were still in the file share if you knew where to look: Agile health checks, DevOps maturity dashboards, coach utilization reports, sprint ceremony attendance, story-point normalization guidance, retrospective themes, QA handoff scorecards. By the second year of every transformation, people wanted bodies moving through a machine because bodies moving through a machine could be counted.

Lena would have hated that. She would have asked which teams owned value from intake to production. She would have asked who could approve a security exception inside forty-eight hours. She would have asked whether a team that learned to move faster would be allowed to change the release path.

Good questions. Dangerous questions.

I needed a person who would ask loud questions that pointed away from those.

Russell tapped the external advisor box. “I had a call last month with Chip B. Runs AI dojos. Big energy. Manufacturing references. Not our size, but adjacent enough.”

“Dojos.”

“He has a whole model.”

“Of course he does.”

“You will hate him.”

“Why?”

“Because he says things like ‘unlocking the builder inside the enterprise.’”

I closed my eyes.

“I wish,” I said.

Russell waited.

“I mean that. I would love to unlock the builder inside the enterprise. I would love to find the room where the operating model is waiting for one courageous sentence. But nobody has gotten an organization our size to change because a consultant found the right metaphor. Not with a run of acquisitions, tens of thousands of repositories, business-unit P and Ls, plant calendars, security review, architecture councils, release trains, QA handoffs, and five messaging platforms still arguing about where the meeting notes live.”

“That is a little long for the brochure.”

“Does he know any magicians?”

Russell laughed. “No.”

“Then we should keep expectations realistic.”

“People remember him,” Russell said. “He fills a room.”

That afternoon we put Chip in a conference room on thirty-one, the one with the glass wall looking north across the city. From that height it was flat roofs, parking lots, hospital towers, warehouses, a gray strip of highway, and the stubborn usefulness of people going places for reasons no board deck would ever capture.

Chip arrived thirteen minutes early.

I remember the thirteen because I had given myself ten minutes to read the briefing packet and three minutes to regret the meeting. My assistant came in and said, “Your two o’clock is already here. He brought his own markers.”

“That is never good.”

“Three colors.”

Chip B. was forty-six and had the physical stillness of a man who did not believe in chairs. He wore a navy quarter-zip under a sport coat, expensive sneakers, and an expression that suggested every airport had personally tried to slow him down and failed. He also wore two smartwatches, one on each wrist, as if time itself needed redundant tracking. Three dry-erase markers sat in his shirt pocket like surgical tools.

He shook my hand with both of his.

“Tom, I am thrilled,” he said. “Kedran is exactly the kind of enterprise where this matters.”

People who called me Tom without being invited usually wanted access. At work, it made me reach for the folder.

Chip’s firm sounded larger in Procurement than it was. The delivery team was Chip and his business partner, who handled contracts, calendars, and the kind of pleasant silence that makes a rate card feel settled. Chip had never run a business the way Kedran meant business: payroll under margin pressure, inventory stuck on a dock, a plant waiting on a controller, a customer screaming because a shipment missed a window. He had run a practice. Somehow that qualified him for a daily rate that made my chief of staff look at the statement of work as if it had made a sound.

To be fair, Chip was good at sales and pageantry. His rates said that before he did, and the brand-new sports car I had passed in the visitor lot, still wearing its temporary plate, said it with better paint. Many people bought business that way. They mistook polish for proof, noise for confidence, and a price high enough to defend for evidence that someone else had already done the diligence.

Chip did not sit. He went straight to the whiteboard.

“The pilot year proved appetite,” he said, writing APPETITE in blue.

It had proved no such thing. It had proved that six volunteer teams would accept tool access and executive attention if we called it sponsorship. But Chip was not asking for correction. He was arranging momentum, and momentum was what the room had been told to buy.

“Year Two is not about more pilots,” he said. “Pilots are for discovering if the water is wet. You know the water is wet. Year Two is muscle memory.”

He wrote MUSCLE MEMORY in green and underlined it twice.

I disliked him before the first marker cap clicked back on.

Not as a person. That distinction mattered to me. He was not cruel, and he was not stupid. He probably called his mother, tipped hotel staff, and sent follow-up notes faster than his competitors. What he had was the consultant habit of turning other people’s blocked reality into a failure of posture. He squinted while you answered, nodded before you finished, and placed your concern in a quadrant he had drawn before he met you.

“Our six-week dojo model creates an immersive lane,” he said. “Teams come in with real work. We give them tool fluency, prompt patterns, secure workflows, code review accelerators, test generation, documentation automation, and demo discipline. They leave with confidence, reusable assets, and a story leaders can repeat.”

“Do they leave with release authority?” I asked.

No trap. I wanted to hear how he handled the word.

Chip smiled, and for half a second I almost liked him.

“They have to,” he said.

Russell stopped writing.

“Not in the inspirational sense,” Chip said. “Bounded authority. A named path. If a team proves the slice, the team has to be able to release the slice without starting over in every approval room. Otherwise you are not building muscle memory. You are training people to wait with better posture.”

That was the first thing he said that was both completely true and professionally fatal.

“Authority is not usually earned by the team at Kedran,” I said. “It is allocated by funding, architecture, risk, and business ownership.”

“Then the dojo has to demand it,” Chip said.

I let the sentence sit in the room. Today it made Chip sound brave. Later, if I needed to move him out of the center, the language would already be sitting in the notes: Chip was pushing ahead of Kedran’s risk posture, asking for authority the operating model had not assigned, confusing capability with governance. A true sentence can become a personnel plan if you leave it in the room long enough.

I nodded as if weighing a bold idea, which is one of the nicer ways to slow it down.

He drew six boxes: orientation and secure tool setup, code understanding and test generation, refactoring patterns, feature slice, stabilization and narrative, demo day and leadership playback. The model was neat, which mattered more than people admit. A neat model lets executives imagine the work has been made obedient.

“What do you need from executives?” Russell asked.

“Presence. Air cover. Storytelling. And a written release-authority path. Leaders have to signal that this is the way Kedran builds now.”

Not budget control. Not a mandate to change funding. Not the power to make Celia accept a central backlog or Dana waive a control because a cohort felt confident. A path. A phrase. The safe version of an unsafe thing, and Chip did not know the difference.

Relief arrived in me anyway, which tells you more about Kedran than about Chip.

“How do you measure success?” I asked.

“Four levels,” Chip said. “Participation. Confidence. Demonstrated capability. Reusable patterns. We also capture use cases, blockers, and leader sentiment.”

“Throughput?”

“Eventually. But Tom, and this is important, immature systems punish early throughput measurement. You have to build the capability floor first. Otherwise you measure old constraints and call the people slow.”

It was the most useful sentence he said all day, partly because it was not wrong. Measure too early, and the data would be noisy. Push too hard, and business units would resist. Give authority before Security finished model review, and we could create real exposure. Every caution carried a fact underneath it, and facts are excellent hiding places.

Chip showed us sample certificates, demo-day agendas, pulse surveys, and a maturity model shaped like a staircase. He had a slide with a woman in safety glasses standing beside a piece of industrial equipment that did not resemble anything Kedran made.

My chief of staff asked about cost.

Chip gave the direct number first, which was smart: facilities, instructors, vendor seats, training environments, travel, program management, measurement, communications. Eighteen million dollars.

“All in?” my chief of staff asked.

Chip tilted his head.

“Internal labor,” I said. “Six-week cohorts. Backfill. Managers. Demo prep. Steering. Opportunity cost against committed delivery.”

Chip nodded as if this were a known supporting detail instead of the larger number. “We can partner with Finance to model that.”

He could not read the economics.

I do not mean he could not read a spreadsheet. He understood price, statements of work, and the calming effect of a round number with commas in it. But he did not feel labor as money once the people were already on payroll. He did not feel the cost of pulling a warranty team into six weeks of learning while the distributor portal slipped another month. He did not see that the room full of engineers cheering at demo day could be the same room full of engineers returning to a backlog that had not forgiven their absence.

That made him useful.

After the meeting, Russell and I walked him to the elevators. Chip kept talking until the doors opened. He said Kedran had the right ingredients. He said the organization was ready for a visible move. He said, “You are going to need to protect the dojo from skeptics.”

“We have a few,” I said.

“Good,” he said. “Skepticism is just energy without a path.”

When the elevator doors closed, Russell turned to me.

“Well?”

“You were right.”

“You hate him.”

“Within the first five minutes.”

“But?”

I looked back through the glass wall. The whiteboard still had Chip’s staircase on it: participation, confidence, demonstrated capability, reusable patterns. In red, under the boxes, he had written RELEASE AUTHORITY PATH and circled it twice. No one on the executive committee would see that version until I had translated it into governance language. Victor would like the energy. Mara would like that the financial story could be contained as a learning investment. Russell would get certificates, cohorts, and pulse surveys. Communications would get rooms full of people doing visible work.

And Chip B., intense as a fire alarm, would never be invited into the room where Victor and Mara discussed operating margin.

“He’s a better fit for Phase Two,” I said.

Russell knew exactly what I meant and chose to hear the kinder version.

Two days later, I briefed Victor.

“Does this create a headcount question?” he asked.

“No immediate workforce implication.”

“Security?”

“Inside approved patterns. Dana will have signoff.”

“Board story?”

“Year One established responsible foundations. Year Two scales capability through structured cohorts.”

He leaned back. “No surprises.”

“That is the design.”

Victor nodded. “Do it.”

I could have told myself the whole story in noble language, and some of it would have been true. Engineers needed training. They had been left alone with tools they were not supposed to use and policies written by people who had never opened a repository. A dojo would help them. A visible program would reduce shadow work. Chip’s two-person firm would answer questions faster than our committees did, mostly because neither one of them had to ask a plant president for permission. There were good people who would leave those rooms better at their jobs.

Some of them might leave Kedran better too. I thought of Lena’s resignation letter, the careful gratitude, the better title waiting somewhere smaller, and felt the complicated warmth of a man who could be pleased by another person’s escape even when the escape proved the limits of the place he still ran. If a few of my people learned enough in the dojo to get better jobs, I would not resent them for it. I liked watching people grow. I had never liked watching people suffer. The ugly part was not that the program might help them leave. The ugly part was that leaving might be the cleanest way the program helped them.

I believed all of that.

I also believed Chip B. could not hurt me.

That Friday evening, I stopped at the Doverline Youth Program for a donor committee reception before going home. The lobby was noisy with kids from the robotics club and parents trying to steer them toward pizza. Near one table, a little girl had wedged a plastic gear into a bracket and was close to tears because she thought she had broken the kit.

I knelt beside her in my overcoat and worked the gear loose with a coffee stirrer from the refreshment table.

“There,” I said. “No harm done.”

Her mother thanked me. I told her my own kids had broken better things than that and survived.

On the drive home, I thought about Lena, Chip’s staircase, and the clean usefulness of a man who had demanded power in a way that would keep him from ever receiving it.

I had done a useful thing that day for a child with a gear.

I had also hired the perfect man to make sure Kedran looked busy for another year.

Both facts came home with me.

Chapter 9: Six Weeks

2028

Executive Play: Train Capability Without Granting Authority

Chip wanted drums for the first morning.

“Not metaphorical drums,” he said. “Actual drums. Energy matters.”

“We manufacture industrial components,” I said. “If people hear drums at eight-thirty, they will assume the ERP system is down.”

Priya M. laughed from the far end of the table without looking up from her laptop. That was the first thing I liked about her.

Chip paced between the desks with an uncapped marker, hunting for a blank surface even though every whiteboard had already been filled, photographed, erased, and filled again during setup. He was useful in the visible way executives understand. He had energy, a vocabulary, and no authority over capital allocation, release calendars, architecture councils, or anyone’s bonus plan, which made him safe to put in front of anxious rooms.

“This has to feel different,” he said.

Priya closed the laptop halfway. “It will. They get access to the tools.”

“And the mindset.”

“Mostly the tools.”

Priya was thirty-five, a principal engineer, and one of the few people at Kedran who could make engineers trust her without performing intimacy first. She opened the code, pointed at the fault line, and stayed there until the work either moved or admitted why it could not. Engineers respected that. Managers respected the engineers respecting that. Chip respected her the way a consultant respects the person who can make the demo real, which meant he tried to stand close to her whenever anyone from Communications entered the room.

The first cohort came in at eight-thirty. Twenty-four engineers, six managers, a few product people, observers from architecture and security, the assigned scrum masters and agile coaches, and HR staff who had the alert posture of people waiting to capture culture in a photograph. Mei S. had sent four people from warranty and service even though her team already owed a distributor portal release and had a support backlog old enough to vote.

Mei stood by the door with a paper cup of coffee, arms folded, watching her engineers choose seats.

“Thank you for making room,” I said.

“I did not make room. I made a trade I will be asked to explain later.”

Mei did not pretend capacity was a mood.

Chip clapped once from the front of the room. “All right, builders. Let’s begin.”

The first morning was better than I expected, and I had expected enough not to schedule a budget call over it. Chip did ten minutes of theater about why six weeks could change the way teams thought about software. He used his hands too much, and he made “builder” do more work than the word could bear, but the room gave him attention. After years of transformation language that landed on managers and never quite reached the code, even Chip’s volume had a certain mercy in it.

Then Priya took over, and the air changed.

She did not start with a manifesto. She opened an ugly repository from a non-regulated internal service and asked the room what they would be afraid to change. People answered slowly at first: brittle tests, unclear ownership, old dependencies, a build script nobody trusted, a QA handoff that took longer than the change itself. Priya wrote each answer down.

“Good,” she said. “The tool is not here to make your fear disappear. It is here to make the fear more specific.”

I wrote that line down.

By Wednesday, engineers who had come in looking tired were staying after sessions to argue about prompt patterns, test generation, and whether a model’s suggested refactor preserved behavior or merely sounded confident. Cynicism did not vanish. It became useful. People tried the tools, found the edges, improved the prompts, and laughed when the assistant produced code that would have passed a demo and failed a Tuesday.

The dojo worked.

Not as a transformation, and not as a delivery engine, but as a room where people who had been treated like capacity units got to practice their craft with modern tools and a teacher who would not let the software flatter them. The old Agile boot-camp binder had promised the same dignity before Kedran converted it into ceremony compliance, ticket hygiene, and coach utilization. Real things can enter a company and still be digested into the old shape.

The warranty team chose a triage improvement for service claims. There was no glamour in it, which made it a better test. When a distributor submitted a claim with a few bad photos, half a failure note, and a product line chosen from a dropdown the rep had never liked, the assistant would suggest likely categories based on product line, failure language, and repair history. A human would still decide. The tool would route work a little faster and reduce the number of claims that bounced between queues until month-end overtime made everyone generous.

Mei came down to see it with a product manager and an architecture liaison who wore the strained smile of a man trying to support innovation without accidentally permitting it. One engineer showed the flow: a claim came in, the note was messy, the assistant extracted failure terms, linked them to a known issue pattern, and suggested three categories with confidence scores. Modest, explainable, and worth about two minutes a claim if it could sit behind the existing screen.

Mei did the math while she watched. Twenty-seven hundred claims a week. A misroute rate that hovered near sixteen percent when month-end volume spiked. Two minutes off first review if the categories held. A QA handoff that still added six to eight business days before anything touched the service screen. She had spent years learning not to call any of that innovation.

“Can this go behind the service screen?” she asked.

The architecture liaison cleared his throat, and Kedran returned to the room.

“We need to confirm ownership of the categorization logic and whether assisted notes create a records-retention issue,” he said.

Mei looked at him. “We own the workflow.”

“You own the workflow. Platform owns the service screen. Data governance owns the claim history. Security owns access. Legal may have a view on assisted recommendations.”

Chip stepped in with a grin. “This is exactly why we capture enterprise dependencies during the dojo.”

Mei said, “It is not a dependency. It is permission.”

The room went quiet in the expensive way.

I could have forced the issue right there and created a satisfying scene for everyone who wanted proof that the CIO was on the side of builders. I also could have let the liaison become the villain, which would have been unfair and convenient. He was describing the system we had built, not inventing a new one to spite Mei’s team.

“Let’s capture the ownership questions and keep the prototype moving inside the approved environment,” I said.

Mei nodded once and left before she handed the steering committee a sentence it could preserve.

By week three, the pattern had settled. The teams learned faster than the company could decide. They used AI to generate tests, explain old code, draft documentation, sketch small feature slices, and reduce the blank-page panic that had made some engineers feel obsolete before they had even touched the tool. The work improved.

Every cohort still had scrum masters and agile coaches. Standups remained mandatory. Story pointing remained mandatory because removing it would have required a decision from people who did not attend the sessions. Retrospectives produced notes that Russell’s team summarized into themes. QA handoffs stayed where they had been, glowing green on compliance reports and adding days to the path between useful code and usable software.

Quarterly planning still took the entire week. Engineers who had spent six weeks learning how to move faster went right back into rooms where Monday was intake, Tuesday was dependency review, Wednesday was capacity negotiation, Thursday was risk adjustment, and Friday was the ceremonial discovery that every team was already overcommitted. The dojo had changed what they could do with a blank screen. It had not changed the calendar that owned them.

Priya saw it before Chip could convert it into a slide.

On the final Friday of the first cohort, I found her in the hallway outside the training room taking sticky notes off a wall. Each note named a blocker: no owner for shared data service, release calendar fixed monthly, test environment unstable, security review outside team, customer workflow split across two product groups, funding locked to old backlog.

“You are stealing evidence,” I said.

“I am rescuing it.”

“From what?”

“Theme clustering.”

She had learned Kedran quickly.

“The teams are better,” she said.

“Yes.”

“The work is better.”

“Yes.”

“They still cannot merge it.”

I did not answer quickly enough to look innocent.

Priya was not accusing me. She did not have Lena’s wounded belief. She had a builder’s economy of words and the annoying habit of staying near the fact.

“What do you want me to do with that?” I asked.

“Say it.”

I could imagine what the board would do with a sentence like that if it landed without padding. Evelyn would ask for merge cycle time by value stream. Mara would ask which budget mechanism owned the blockers. Victor would ask whether the issue created risk to guidance. None of those questions would be wrong, and every one would turn a six-week training success into an operating-model debate we had not agreed to have.

Chip came around the corner with two people from Russell’s team.

“There you are,” he said. “We are collecting reflection language.”

Priya put the sticky notes into a folder. “We need to talk about merge paths.”

Chip smiled. “Exactly. Merge enablement opportunity.”

Priya looked at him for one full second.

I rescued him because I needed him unbruised. Chip was harmless when he stayed visible, and visibility was his whole value to me.

“Dependency awareness goes into the cohort readout,” I said. “Let’s keep the language consistent.”

There are moments when you hear yourself translate a human fact into an executive phrase and feel the machine accept the coin.

They can write the code now. They still cannot merge it.

By the end of six weeks, the first cohort had produced five demos, twenty-seven reusable prompt patterns, improved tests in a dozen repositories, cleaner documentation for two legacy services, and more confidence than the organization deserved. The engineers clapped for each other in a way that made me proud. Chip took photos in front of the dojo wall, two smartwatches flashing as he lifted both arms. Russell’s team captured quotes. Communications asked for one more shot of the warranty group with laptops open.

Mei’s team went back to its regular work with new skills and the same dates. The distributor portal did not care that her engineer could now generate tests. The release train did not care that the warranty triage slice had looked good in a training environment. The shared service boundary did not care that a room full of people had clapped.

At lunch, Mei asked me the question she had earned.

“What am I supposed to tell my division president?”

I could have said what she wanted me to say. Tell him the company took your people for six weeks and should move the commitment. Tell him the triage slice proved the operating model was blocking value. Tell him Thomas V. will back you.

“Tell him the team built capability we can reuse,” I said. “Tell him the triage work is a candidate for the next release path once ownership is confirmed.”

Mei looked at me over the sneeze guard.

“So, nothing.”

“Not nothing.”

“Tom.”

“Not enough,” I said.

That was the most I could give her in the cafeteria with Finance people two tables away and my own calendar already full of the next steering committee.

I stood in line with a bowl of soup I no longer wanted and watched the room move around me. Engineers with laptops. Finance people in quarter-close fleece vests. A plant manager visiting from Kokomo, eating alone while answering emails. People with mortgages, college savings, parents in hospitals, bad knees, and calendars full of work they did not control.

The dojo had made some of them better.

It had not made any of them freer.

Chapter 10: Green Demo Day

Late 2028

Executive Play: Make Activity The Dashboard

The dashboard was Russell’s idea, so it arrived with friendlier colors than the truth required.

He brought the first version to my office on a Monday in July, printed on eleven-by-seventeen paper because HR leaders still believe a large sheet makes a draft feel collaborative. Chip came with him, carrying enough optimism for the three of us. Mara joined by video from Chicago, where she had spent the morning with investors and had the flat patience of a CFO who had already listened to two men explain macro uncertainty as if they had discovered weather.

The paper had twelve tiles.

Engineers trained. Cohorts completed. Demos delivered. Use cases submitted. Prompt patterns published. Repositories touched. Test cases generated. Developer confidence. Leader attendance. Policy exceptions closed. Satisfaction score. Dependency themes captured.

Every tile was green except leader attendance, which was yellow because two division presidents had sent deputies to demo day.

“That one needs to be green,” Chip said before anyone sat down.

Mara looked up through the screen. “Good morning to you too, Chip.”

He laughed with the cheerful immunity of a man who did not know when he had been corrected. “We have seventy-eight percent senior leader presence across cohort events. That is strong.”

“Then why is it yellow?”

Russell adjusted his glasses. “The draft threshold was eighty.”

“Change the threshold,” Mara said.

No one wrote it down in the moment. We all understood it had happened. Russell would put the new threshold in the definitions page because Mara did not tolerate numbers that could not survive a callback. The move was aggressive, not reckless, which is a distinction public companies pay lawyers to preserve.

I had spent the hour before the meeting with a spreadsheet that did not use friendly colors. Year Two was on pace for eighteen million dollars in direct spend and thirty-four million in internal labor and opportunity cost. Fifty-two million all in, if we counted dedicated dojo staff, Priya’s instructor group, vendor seats, training environments, backfill, travel, manager time, demo prep, Communications support, steering committees, and the usual fog around the cost of people who were already on payroll until a program needed to defend itself.

Fifty-two million dollars needs a story. The dashboard was the story learning to stand.

It was also the safest possible surface for Year Two. A green dashboard gave every function a reflection it could tolerate: HR saw people developed, Finance saw disciplined tracking, Security saw policy closure, Communications saw a visual, Victor saw scale, and the board saw a company behaving like it had a plan. The unsafe dashboard would have asked how much wait had left the system. It would have shown merge time, approval time, release ownership, and the denominator under every impressive numerator. That dashboard would have been more useful and less survivable.

“Walk me through it,” I said.

Russell took the chair beside my small conference table. Chip stayed standing, because he liked his sentences better when gravity could not catch them.

“We are tracking three hundred eighty-six engineers trained, with another one hundred twelve active,” Russell said. “Seven cohorts complete, three in flight. Forty-one demos delivered. Sixty-four prompt patterns published.”

“Repos touched?” Mara asked.

“Five hundred twelve.”

She looked down, probably at her own notes. “Touched means what?”

Chip stepped in. “A repository where a cohort participant applied at least one approved AI-assisted development pattern. Code explanation, test generation, documentation, refactoring analysis, small feature slice, any use that builds the muscle.”

Mara looked at me. Touched did not mean shipped, retired, secured, simplified, or made cheaper to maintain. It meant someone had opened the door and left a fingerprint.

“Define it in the appendix,” I said.

“Do we need the appendix in the board version?” Russell asked.

“Yes,” Mara said, then allowed herself a thin smile. “No one will read it.”

Chip pointed at the confidence tile. “This is the headline. Developer confidence up thirty-one points from cohort start to completion.”

“Self-reported,” Mara said.

“Yes,” Russell said, “and that matters. We have been asking people to adopt tools that can embarrass them, expose them, or make them wonder whether the company is training their replacement. Permission to learn in daylight is a real indicator.”

He believed it. He was also right. The dojo had given people a room where they could be bad at a new tool without being punished for it. That is not a trivial gift in a company where a failed estimate can outlive a marriage.

“What about throughput?” Mara asked.

Chip took a breath. I could see the sentence forming in him, polished and useless.

I answered first.

“Too early for enterprise-level measurement,” I said. “The cohort data is meaningful for capability and adoption. Delivery outcomes are still tangled in intake, review, release governance, data ownership, and business prioritization.”

Mara looked at me through the screen. “You can say that to me. I am asking what we say to the board.”

“We say Phase Two is building the measurement foundation while scaling capability.”

Russell nodded. Chip nodded harder.

Mara did not nod. “Evelyn will ask whether we are faster.”

“In places.”

“She will ask which places.”

“Low-risk internal tools, selected test automation, documentation-heavy workflows. We will have examples.”

“She will ask for lead time.”

“We will tell her repository data remains fragmented across acquisitions and the dojo is surfacing the dependency map needed for mature measurement.”

Mara sat back. “That sentence is a sweater.”

“Warm?”

“It covers a lot.”

On the paper in front of me, every tile promised motion. Engineers trained. Cohorts completed. Demos delivered. Use cases submitted. Prompt patterns published. Repositories touched. Test cases generated. Confidence up. Leaders present. Satisfaction high. Dependencies captured.

I turned the page sideways and wrote the missing tiles in the margin: end-to-end lead time, approval wait removed, teams with release authority, handoffs eliminated, repos retired divided by the full estate. The list ran down the edge of Russell’s friendly colors like contraband. Each measure crossed a boundary where Kedran had decided, years before AI, that responsibility should be divided until no one person could be blamed for time.

My predecessor had used green dashboards for Agile adoption and DevOps maturity, and I had inherited the full museum of their afterlife: laminated ways of working, ceremony attendance, maturity levels, deployment-readiness checklists, platform scorecards, and quarterly slides showing progress that had never quite reached the operating model. Kedran had learned how to turn real practices into safe evidence.

“Move leader attendance to green if the threshold is seventy-five,” I said.

Russell made a note.

“Add dependency themes as its own tile.”

“Green or yellow?” Russell asked.

Priya’s sticky notes had not disappeared. They had become categories.

“Green if themes are captured and assigned.”

Mara leaned toward her camera. “Assigned to who?”

“Operating model working group.”

“Does that exist?”

“It can by Friday.”

She gave me the small CFO look that meant she recognized fresh furniture being built around an old hole.

“Fine,” she said. “Do not staff it like a second dojo.”

After the meeting, I walked Chip and Russell down to the elevator bank. A plant team in steel-toe boots stood beside a wall of framed innovation awards, trying not to look at the posters of people who had once made different promises about different programs.

Chip carried the dashboard under one arm.

“Tom, this is strong,” he said. “You have the story now.”

“We have a story.”

“The board will see scale.”

“The board will see green.”

He missed the difference and kept smiling.

Russell stayed behind when Chip got on the elevator.

“You think it is too soft,” he said.

“I think it is exactly soft enough.”

He folded his copy of the dashboard. “Lena would have hated it.”

“Yes.”

“Priya will hate it.”

“Priya hates fewer things than you think. She just knows where they are.”

The next day, Priya came to the dashboard review with a backpack, wet hair, and the expression of a person who had already decided patience would not be reimbursed.

Chip presented the tiles. Russell framed the people story. Aaron explained the appendix. I watched Priya read the page.

“Where is merge time?” she asked.

The room did not freeze. We were too experienced for that. It became careful.

“Captured under dependency themes,” Chip said.

“That is not merge time.”

“We are not at the maturity point where that metric is stable across cohorts.”

“Then show it unstable.”

I looked at her. “For the board dashboard, we need consistent definitions.”

“For the work, we need to know whether anything changed.”

Chip said, “The cohorts are changing people.”

Priya turned to him. “I know. I am one of the people changing them.”

That landed harder than if she had raised her voice.

“Then call it adoption,” she said. “Do not let people mistake it for delivery.”

No one had written delivery on the page. That was the cleverness of it, and she saw that too.

“Fair,” I said. “The title becomes AI Dojo Scale Dashboard.”

She watched me long enough to decide whether the concession had weight.

“Add a note that production throughput is not measured yet.”

Chip shifted. Russell looked at me.

There are requests that fit in a sentence and still require a room to admit what it is protecting.

“In the appendix,” I said.

Priya nodded once. She knew exactly what she had won and what she had not.

By the end of the week, the dashboard was beautiful. Leader attendance was green. Dependency themes were green. Confidence was green. Cohorts were green. Demos were green. Use cases were green. Repos touched were green. Test cases generated were green. The appendix contained definitions, caveats, and one sentence noting that enterprise production throughput would be measured in a later phase after dependency and data-quality normalization.

Evelyn would find that sentence if she wanted to. I knew she would.

Most people would see the first page.

Demo Day came in October, before production blockers could ruin the room.

Chip arrived at the Kedran learning center before the facilities crew had finished setting out the coffee. I found him under a ladder with three uncapped markers in one hand and a wireless microphone clipped to his belt. He had already moved two poster boards, offended the AV technician, and written RESPONSIBLE ACCELERATION in block letters on a whiteboard nobody had asked for.

“Tom,” he said, as if we had not seen each other in thirty-six hours. “Big day.”

“It is.”

The AV technician was on the floor under the front table with a spool of cable and the expression of someone who had been handed a vision statement instead of an outlet. I knew her father had retired from our Anderson plant after thirty-one years, and I made a point of knowing people whose work became visible only when it failed.

I set my bag down and crouched beside her. “What do you need?”

“Nobody knows where the second HDMI run goes.”

Chip started to answer. I lifted one hand without looking at him.

“Let her talk.”

She pointed at the ceiling track. “It is supposed to feed the side displays, but I am only getting signal on the left. Ten minutes if people stop moving my tables.”

I looked at Chip. “Nobody moves a table.”

He pressed his lips together and nodded as if I had just given him a leadership principle worth laminating.

Chip was exhausting, but he was not malicious. He wanted the day to matter. He believed proof changed organizations if it was presented with enough energy, enough applause, and clean enough typography. I had hired him because he believed that, and because Victor and Mara never would. The useful evangelist in a company this large is not the person who can change the operating model. It is the person intense enough to make the work visible and harmless enough that no executive has to fear him.

By eight-thirty the room had the clean, temporary happiness of a hotel ballroom before the first coffee spill. Demo stations lined the walls. Communications had set up cameras near the stage. A banner hung above the entrance:

KEDRAN AI DOJO SHOWCASE

Someone from Russell’s team had added the small line underneath:

Building responsibly. Learning together.

I hated it less than I expected to.

The teams started arriving in blue badges and uncomfortable shoes. Engineers who had spent most of their careers avoiding executive floors stood next to posters with QR codes, architecture diagrams, screenshots, and metrics that had survived three rounds of Communications review. Priya moved among them with a paper cup of tea, adjusting nothing, asking practical questions.

“Does the search demo still work without the cached index?”

“Show the failing test first. Then show the generated test.”

“Do not say the model understands the code. Say it summarizes patterns. That matters.”

She had no stage voice. Engineers listened to her anyway.

Devon was at the service-parts search station, wearing a sport coat that did not belong to him over another West Lafayette shirt. He saw me looking and tugged at the cuff.

“Borrowed,” he said.

“Good choice. Whoever owns it has longer arms.”

“Still West Lafayette underneath, though.”

“Some of us have standards.”

He laughed, which was what I had wanted. Devon still looked like the man from the twelve-day demo two years earlier, a little older and much more careful around me. He had learned that proof inside Kedran needed handlers.

His team’s demo took a service technician’s half-typed description of a failed pump assembly and returned likely parts, service bulletins, replacement kits, and a confidence score. No magic. Better search over ugly data, with enough guardrails to keep anyone from ordering a part on blind faith.

“Is it live?” I asked.

“In the sandbox.”

“Production path?”

He glanced toward Priya before answering. “Waiting on data ownership for two feeds.”

Devon had learned to wait in complete sentences.

The room had voltage because the work had voltage. I will not pretend otherwise. People stand straighter when they can show working software. Middle managers who had been reciting status for years suddenly had engineers explaining code to the CEO with a laptop open between them, and for a few minutes the old layers seemed less permanent than they were.

Victor arrived at ten with his usual gift for making an entrance look accidental. Chip intercepted him ten feet inside the door.

“Victor, you are going to see the future of delivery today.”

Victor looked at me over Chip’s shoulder.

“No surprises?”

“No surprises,” I said.

That was our whole operating model in two words.

Evelyn arrived late and without an escort. Board members usually traveled through buildings as if gravity applied differently to them; Evelyn came in carrying her own notebook, skipped Chip, took a cup of coffee, and went straight to the automated test generation station.

The engineer there showed her how the tool produced unit tests for the distributor portal. He clicked through a pull request, explained the failing case, and showed how the suggested test had caught a bug in discount logic before it merged.

Evelyn did not smile.

“How often do you deploy this portal?”

“Scheduled release monthly,” he said.

“Did this change that?”

He looked toward his manager. His manager looked toward Priya. Priya did not rescue him.

“No,” he said. “The release calendar is still monthly.”

“Who owns that calendar?”

“Business release governance.”

She wrote a note.

Evelyn always found the place where time went after the demo ended.

Chip hated silence, so he tried to fill it. “Part of the power here is confidence. Teams are finding issues earlier, which creates the runway for future release modernization.”

Evelyn looked at him as if he were a vending machine that had dispensed soup. “I asked who owns the calendar.”

I stepped beside him before he could say partnership again.

“That one sits with the distributor steering group,” I said. “The dojo work improved the quality path. Release frequency is part of the next operating conversation.”

“Is it on a calendar?”

“Not yet.”

“Then it is not a conversation. It is a sentence.”

She moved to the next station.

At eleven-thirty, Russell’s team dimmed the lights and put the crowd in chairs. Chip warmed them up with a five-minute talk that lasted twelve. He called the dojo a movement, a muscle, and a builder revolution. Nobody important believed he could move a budget code, but the engineers clapped because he had spent six weeks caring about their work and because being cared about, even theatrically, is better than being ignored precisely.

Then I walked to the stage.

Nobody pays you seven figures to panic in public. They pay you to give unresolved dependencies a shape the room can hold.

“Two years ago,” I said, “a small Kedran team showed us a result that was both exciting and uncomfortable. They showed us that the way we build software could change faster than the way a company like ours usually changes.”

Devon looked down at his shoes.

“We made a choice then. We were not going to chase novelty. We were not going to pretend risk did not exist. We were also not going to hide behind risk as an excuse to stand still.”

That line got a good response. It deserved one. It was true enough.

“Our next job is to turn these lessons into platform capability,” I said. “We will remove friction where we can, standardize where we should, and keep the controls that protect our customers, our plants, and our people.”

Victor nodded. Mara, near the side wall, was already thinking about the budget language. Evelyn was not nodding. Priya watched me with no visible expression.

After the applause, Victor did two camera takes in front of the service-parts demo, then left for an analyst prep call.

“Good day, Tom,” he said on the way out.

“Thank you.”

“Feels real.”

“Parts of it are.”

He gave me a quick look, then smiled because a CEO can forgive almost any sentence that arrives after a good event.

“Keep it controlled.”

“I will.”

Evelyn appeared beside me after the room started to empty.

“How many are in production?”

No greeting. No softener.

“Of the demos?”

“Yes.”

Chip, who had the gift of arriving at the wrong exact second, inserted himself on my left.

“Four are on track for productionization.”

Evelyn did not turn her head.

“That was not the question.”

Chip’s smile held, but the rest of him froze.

Priya walked up carrying her tea. She had heard enough.

“One is in limited internal use,” she said. “The plant documentation one. It is not connected to production systems. Warranty is waiting on policy for assisted claim notes. Service-parts search is blocked on data ownership for two feeds. Distributor portal tests are useful now, but release frequency did not change. Controls code explanation is approved for developer support, not code changes.”

Chip opened his mouth. I touched his elbow lightly, and he stopped.

“Two of those are blocked by the same approval path that existed before the dojo,” Priya said. “Different labels. Same line.”

It was the cleanest sentence of the day.

Evelyn turned to me. “So the demo proves what?”

I had several answers available. Some were useful. One was honest enough to be dangerous.

If a demo cannot become production, the demo was not the proof.

I did not say that.

“It proves the capability is real,” I said. “It also proves the next constraint is not the individual engineer. Year Three should focus on platform-level friction removal: approved patterns, clearer data ownership, low-risk release pathways, automated review where it fits.”

Evelyn kept her eyes on me. She had been a founder. She knew when a sentence was built to carry a truth and a delay at the same time.

“Send me the production conversion list,” she said.

“Of course.”

“Not the showcase list. The conversion list.”

“I understand.”

The crowd had gotten its celebration. The cameras had gotten their footage. Victor had gotten a story. Chip had gotten a movement. Priya had gotten another proof point that craft was not authority. Evelyn had gotten a list to ask for.

And I had gotten Year Three.

Two weeks later, Mara scheduled a budget meeting at 7:15 in the morning, which meant she wanted the room too tired to perform optimism.

“I need thirty-two million out of technology run rate before the February board package,” she said.

“Good morning to you too.”

She slid the packet across the desk and tapped the second page.

“Contract labor. Duplicate tools. Training vendors. Backfills. Consulting carryover. Pick your nouns.”

The AI Dojo was not the largest number in my budget. It was simply the newest number wearing the most adjectives. Mara had put the three-year bridge on page four, where expensive truths go until someone needs them. Year One: seven million direct, thirteen million in labor and opportunity cost, twenty all in. Year Two: eighteen direct, thirty-four internal, fifty-two all in. Year Three, if the platform plan passed with the cuts wrapped around it: fourteen direct, twenty-four internal, thirty-eight all in. One hundred ten million dollars over three years for a program that had produced real local capability, useful policy, cleaner tooling, and no enterprise throughput benefit either of us could tie to the P and L.

“That page does not go to Victor first,” she said.

“No.”

“But you need to understand it before you sell page one.”

“I do.”

There was no company-level productivity dividend I could defend if Evelyn decided to keep asking for one. Enterprise cycle time had not moved in any way that would survive a hard question. Revenue systems were not shipping faster. The acquisition backlog still sat there, a pile of inherited promises and local exceptions. Review boards still met on Tuesdays, architecture still needed one more readout, and funding still arrived in annual chunks with quarterly rituals attached. The bottleneck had moved in the demos because demos belong to the people in the room. In the company, the bottleneck stayed where Kedran kept bottlenecks, inside permission.

Mara folded her hands over the packet. “You know what this looks like.”

“It looks like you want to cut around the program and let the program share credit.”

She smiled for the first time. “I knew you were rested.”

We consolidated duplicate AI tool trials from eleven to three, then one preferred path. We folded overlapping learning vendors into dojo materials and internal patterns. We reduced external support in enablement groups, delayed backfills officially, and canceled some quietly if volume held. We did not touch plant support, revenue systems, or quarter-close work because even budget courage knows where the live wires are.

“Can we attribute part of it to AI development maturity?” Mara asked.

“Part.”

“How much part?”

“Enough to say the program accelerated tool consolidation, reduced duplicate training spend, and improved internal capability so we can lower dependency on external support in selected areas.”

“Is that true?”

“In selected areas.”

“That,” she said, taking back her pen, “is the most important phrase in finance.”

The cuts did not come from cycle-time gains. We were not doing the same work with fewer people because the dojo had made delivery faster end to end. We were trimming the brush around the road and calling the road improved. The sentence would be defensible at the right altitude, which is the form most durable corporate stories prefer.

At nine, we moved to Victor’s conference room. Chip arrived with a backpack, a scarf he wore indoors, and the look of a man ready to be included in strategy. Priya came in after him without a notebook. She carried one pen and a face that said the meeting had already happened somewhere else.

Victor stood by the window, reading Mara’s one-page summary.

“So this is good news,” he said.

“It is disciplined news,” Mara said.

I walked him through the package with the nouns in the order Mara wanted. In a board deck, the sequence would look causal. In the ledger, it would look adjacent. In a company as large as Kedran, adjacent is often close enough for the people who have to explain February.

Chip almost bounced in his chair. “This is exactly the maturity curve we talked about. Capability compounds, then dependency falls.”

Priya looked at him. “That is not what happened.”

The room quieted around her.

“The program created skill,” Priya said. “That matters. Some teams can do more for themselves, and some tool cleanup should have happened years ago. But the savings are coming from consolidation and budget pressure. We did not remove the wait states that control delivery.”

Victor asked what she meant by wait states.

“Architecture review queues,” she said. “Data ownership decisions. Release governance. Funding approvals. Places where work waits for permission after the team knows what to do.”

Chip lifted both hands, trying to make agreement look like control. “And those are exactly the dependencies the next phase will target.”

“Then say that,” Priya said. “Do not say the savings came from delivery productivity.”

I liked her in that moment more than was convenient. Priya was not dramatic. She was doing what good engineers do when a system starts turning cause into politically correct statements: putting labels back on wires before someone closes the panel.

Victor turned to me.

“Tom?”

I had the choice again, the familiar fork between a sharper truth and a steadier room.

“Priya is right on the technical diagnosis,” I said. “The savings package is not evidence that enterprise cycle time improved. It is evidence that the program gave us enough visibility and internal skill to rationalize the surrounding cost base: duplicate tools, external training, and some contractor dependency. That is real value. It is not the whole value story, and it is not the end-to-end delivery story.”

Mara watched me with approval because I had given Victor a clean enough sentence for the board without putting a board-safe one in Priya’s mouth.

Victor folded the summary. “Can I say the AI development program is producing financial benefits?”

“Yes,” Mara said.

“With discipline,” I added.

“I do not say discipline at investor conferences unless I have to,” Victor said. “They hear layoffs.”

“Then call it productivity foundation,” Chip offered.

Nobody wrote that down.

That afternoon, I called three managers whose contractor lists were about to become targets. I told them before the spreadsheet did. I asked for names, real names, not resource IDs: who had plant knowledge, who had a sick spouse, who was carrying a fragile integration nobody had documented because the business kept postponing the rewrite. I could not save everyone. I could keep the lazy cuts from doing the most damage.

I had spent years taking sharp instructions from above and blunting them before they reached people with mortgages, children, and parents in assisted living.

At six, I drove to the Doverline Youth Program for a donor committee meeting I had forgotten until my calendar buzzed. I was late, still in the same suit, carrying the budget packet in my bag like contraband. The projector refused to connect. Someone made a joke about needing a CIO, so I got under the table with my phone flashlight and found the loose adapter.

The room applauded when the screen came alive.

“Do not encourage him,” Claire said from the far end of the table.

I stayed after to help stack chairs.

On the drive home, she asked, “Bad day?”

“Complicated day.”

“That usually means someone else had a bad day and you found language for it.”

I kept my eyes on Meridian Street. “We are cutting contractor spend. Vendors. Backfills. The usual.”

“Is it because the AI work made people faster?”

I could have given her the board answer. I had practiced it all day.

“Not really.”

She nodded, not satisfied, but appreciative of the more convenient version I had declined to tell.

The next week, Communications drafted the internal article. It praised the AI development program for enabling tool consolidation, strengthening internal capability, and supporting disciplined investment. It included a photograph from Demo Day where Devon stood beside the service-parts search screen, proud and nervous in the borrowed coat.

I approved the article after changing two verbs.

The external story was clean. The private readout was cleaner. Year Two had cost more than Year One, generated better stories, trained more people, made some engineers genuinely better, cut adjacent spend, and still produced no macro benefit I could see.

It had also bought me the one thing no spreadsheet named.

Another year.

Chapter 11: From Pilots To Platform

2029

Executive Play: Use The Numerator

Dana K. came to my office at eight with her own marked copy of the January operating review deck and the look of someone who had found a concealed charge in a hotel bill.

Across the top of the first page, Russell’s team had written the phrase that had arrived in a Tuesday steering committee and immediately learned to stand upright in capital letters:

From Pilots To Platform.

The slide had three columns: Standardize, Simplify, Scale. Under each were the moves we were finally willing to make after two years of pilots, dojos, showcases, confidence surveys, tool trials, risk reviews, steering committees, and the other machinery Kedran used to turn fear into project management.

Some of the moves were real.

We would standardize on one primary coding assistant vendor by midyear. We would create approved patterns for non-regulated repositories so teams did not have to relitigate the same source-code questions every time a developer wanted help writing tests. We would remove two approval gates for low-risk internal applications, launch automated code review for selected repos, fund Priya’s prompt and pattern library because engineers were already sharing useful examples in private channels, and start a repo-retirement campaign, which sounded tidy until you met a repo.

We would not remove scrum masters, agile coaches, standups, story pointing, retrospectives, or QA handoffs. Those were treated as the floor, not the question. The AI work had to fit inside the delivery model Kedran already knew how to audit.

I had marked up my own copy before Dana arrived. “Transformation foundation” became “developer platform maturity.” “Release acceleration” became “low-risk pathway.” “Enterprise-wide” sat there for a while, smug and useful, until I crossed it out and wrote “enterprise-aligned.”

There are phrases you learn to love because they leave room for what you are not doing.

“I see Security owns three blockers,” Dana said.

“Good morning.”

“Tom.”

I took off my glasses. “They are not blockers. They are control decisions.”

“The slide says Security alignment remains the gating factor for broader adoption.”

“In regulated repos and sensitive data paths.”

“That adjective is on page nineteen. The first twelve pages make it sound like my office is the reason teams cannot use tools.”

She was right. Not completely, which is how these arguments survive, but enough. Dana had been the safest shield in the company because her concerns were legitimate: source-code exposure, vendor retention, regulated data, auditability, identity, and the kind of quiet exception that looks harmless until a reporter calls, at which point everybody asks why Security did not stop it.

I had used her office when I needed time. I had also used her office when Security had already given a bounded answer and I wanted the bounds to sound like the point.

“I will change the deck,” I said.

She had come prepared for a longer fight, so the sentence reached her a beat late.

“The first section will say the approved pattern applies to non-regulated repositories with no customer data, no embedded controls, no plant operational dependency, and no export-restricted code,” I said. “Security owns the pattern with Platform. Not the delay.”

“Thank you.”

“You still have to sit through the meeting.”

“I assumed there would be punishment.”

I smiled despite myself. Dana did not, but the room temperature dropped.

“One more thing,” she said from the doorway.

“Go ahead.”

“Do not ask me to approve the phrase ‘AI-safe by design.’”

“I would never.”

“You would if Russell got to you first.”

The operating review was in Victor’s boardroom, where the table was long enough to make agreement look formal. Mara sat to his right, quiet and exact. The COO joined by video from a plant cutover in Ohio, wearing a hard hat and the expression of a person who would kill any central initiative approaching production if it arrived with adjectives instead of a rollback plan. Priya sat near the end of the table. Aaron B. sat beside her with a yellow legal pad, a pen clipped to his shirt pocket, and the patient sadness of a man who knew where old systems were buried.

Chip was not in the room.

That had taken no courage. It took scheduling. He was two floors down running a dojo alumni activation session with thirty managers and a buffet of boxed lunches. Chip was useful in rooms where motion mattered more than authority; in Victor’s room, he would have turned every boundary into a slogan and then been confused when no one funded it. He was harmless because he did not control budgets, release rights, or Victor’s calendar, and because I made sure those facts stayed true.

I started with the phrase everyone already liked.

“Year Three is about moving from pilots to platform,” I said. “We are not declaring victory. We are taking repeatable lessons from the pilot and dojo work and removing friction where the organization can safely absorb change.”

Victor nodded. Safely absorb change was one of his favorite shapes. It sounded active and insured.

It was also the only kind of change I could still make without turning the program into a fight Kedran would win by not having. Safe-zone work was low-risk pathways, approved patterns, automated review where the blast radius was small, vendor consolidation, and repo cleanup that did not interrupt a plant, a distributor, or a quarter. Unsafe-zone work was one real value stream with funding, architecture, security review, release authority, support, and business priority placed under one accountable owner. That was the work Priya wanted. It was also the work that could make every protected room in the company protect itself at once.

I walked through vendor standardization first: eleven tool trials down to one primary vendor and two controlled exceptions by the third quarter. Procurement liked it. Finance liked it. Security could live with it. The business-unit presidents could complain in principle and still get enough seats to keep their local leaders quiet.

Then I put the approved patterns on the screen. Non-regulated code. No plant operations. No customer data. No embedded controls. No export-restricted repositories. It was a paved path for the part of Kedran where the blast radius fit inside a conference room.

The COO unmuted. “So not plant scheduling.”

“Not in wave one.”

“Good.”

She muted again.

That was how much authority the work had over the systems that made money.

Priya presented the automated code review plan after me. She had refused to let Communications touch her slides, which made them clearer and worse looking. The first one showed a pull request with comments from three human reviewers, two automated scanners, one architecture bot, and a final approval that had taken nine calendar days for a twelve-line change.

“This is not quality,” she said. “This is queueing with comments.”

Mara looked up.

Priya clicked to the next slide, a plain table with evidence types down the left side: tests, deploy history, rollback path, ownership, observability, dependency age, policy exceptions. “For selected repos, routine checks can move into automated review. Style, common security patterns, dependency age, test impact, ownership rules, generated test suggestions. But the larger point is deciding what evidence we trust. If a team has strong tests, deploy evidence, observability, rollback safety, and clear ownership, its review path should be different. If it does not, adding another reviewer is theater.”

She had the room then. Even Victor.

The old executive instinct rose in me: praise the sharp point, scope the sharp point, make it survivable.

“That is the direction,” I said. “Wave one focuses on internal applications with mature test coverage and no regulated data.”

Priya looked at me. “That is the safe part of the direction.”

“It is the part we can start.”

“Starting is not the same as changing the model.”

Victor shifted in his chair. Mara looked down at her notes. Dana watched me to see whether I would make Security the reason.

“No,” I said. “It is not. The operating model question is bigger than automated review: funding, release ownership, risk classification, architecture decision rights. We will not solve all of that in this wave.”

It was the most honest sentence in the meeting, and it was also a useful one because the word wave promised another meeting.

Aaron took the repo-retirement section.

I had known Aaron for fifteen years. He was fifty-nine, built like a high school shop teacher, and incapable of making technical debt fashionable. If a system was ugly, he called it ugly. If an ugly system was keeping invoices moving, he said that too. He had saved us from several clean plans by remembering one dirty fact nobody else had written down.

“We have tens of thousands of repositories in indexed source systems,” he began. “That number includes active applications, retired application remnants, vendor forks, build scripts, data export jobs, firmware branches, plant utilities, reporting tools, one-time conversion scripts that became monthly processes, and work inherited through a run of acquisitions that no longer maps cleanly to the org chart.”

Victor smiled at the list. Aaron did not.

“The campaign proposes three retirement classes. Class A: no commits in five years, no deployment record, no known package dependency, and no business owner response after two notices. Class B: no active development, but evidence of scheduled execution, data transfer, reporting, or contractual use. Class C: active or unknown with operational dependency.”

The COO unmuted again. “Give me an example.”

Aaron did not look down. “A repo called mth-end-adjust-old appears dead. Last commit 2018. No deployment record in the central tool. It generates a flat file on the twenty-seventh of every month for a finance process inherited from the 2014 Veylan acquisition. If we remove it, month-end close does not fail loudly. It gets wrong quietly.”

Mara wrote that down.

“Another one,” Aaron said. “Plant dashboard helper from the 2009 Tivorn acquisition. Looks like a deprecated web service. It feeds a local dashboard in Fort Wayne that maintenance supervisors use during third shift. No enterprise owner because the plant built it with a contractor who died in 2016.”

Nobody smiled at that.

“Another one. Export utility tied to a private-label service contract. The contract still requires a nightly file in a format we would not design now. Legal says the obligation runs until renewal in 2031.”

He turned away from the screen. “That is why I do not recommend repo retirement as a number exercise. Some of the mess is neglect. Some is memory.”

That was Aaron at his best. He complicated the easy executive statement without killing the work. He was why I trusted old engineers more than new dashboards.

“What can we retire?” Victor asked.

“A few hundred in the first wave,” Aaron said. “Maybe. If the business owners answer. If they do not, fewer.”

“Out of all of them.”

“Yes.”

The denominator landed in the room, and the room disliked it.

I stepped in. “The first wave is about building the operating pattern: classification, owner confirmation, archival process, dependency checks, recovery procedure, audit evidence. We do not get to rationalization at scale by deleting the loudest things first.”

That was true, which made it especially valuable to me. Complexity was not an excuse I had invented. It was the floor under us. Kedran was made of acquisitions, contracts, local workarounds, plant exceptions, quarter-close rituals, and systems that had outlived the people who could explain them. The simple version of change was too clean for the company underneath it. The careful version was slow enough to retire on.

After the meeting, Priya and Aaron walked with me toward the elevators. Chip was visible through the glass wall of the training room, moving between tables with a marker in his hand while thirty managers watched him tape sticky notes to a wall.

“He thinks he is in the strategy meeting,” Aaron said.

“He is happier where he is,” Priya said.

“Most of us are,” I said.

Priya did not laugh.

In the elevator, Aaron turned to me. “Tom, the repo campaign will be useful if we do not let people make it a trophy hunt.”

“I know.”

“No, you know and you also like trophies.”

He had earned that sentence over fifteen years.

“Fair,” I said.

“Class A first. No heroics. No one percent jokes. No pretending three hundred repos changes the estate.”

“You are taking away my best future slide.”

“Good.”

He was too late.

By August, the number had arrived.

Three hundred and twelve could stand alone. It had weight. It sounded earned. It sounded as if people had gone into the basement with flashlights and come back carrying old machinery in pieces.

On the slide, it looked even better.

312 repositories retired.

The number sat in the center of a white field, Kedran blue around the edge, with a small line beneath it in pale gray: reducing technical drag through responsible platform maturity. Communications had tried cleaner words. Mara asked whether we could prove them. Aaron looked personally wounded by “cleaning up legacy code.” The final line was vague enough to survive.

I was in the tenth-floor conference room with Mara, Chip, Priya, and two people from Communications who had the careful posture of staff members waiting for executives to ruin a slide by liking the wrong thing.

Chip loved the number. He loved all numbers that arrived with applause already attached.

“It gives the story a physical body,” he said.

Mara looked over her glasses. “A physical body.”

“People can feel it.”

Priya looked down at her laptop, and I saw the corner of her mouth tighten. She had been polite to Chip for almost eighteen months, which was one of the great feats of discipline inside the AI program.

“We should be precise about what it means,” she said.

“Of course,” Chip said, too quickly. “Precision is the soul of credibility.”

Mara turned one page in the printed deck. “What does it mean?”

Nobody answered for half a second. That was my cue.

“It means the repo-retirement pattern is working,” I said. “We have a repeatable intake path, dependency review, archival standard, owner signoff, and evidence capture for audit. This first wave was low-risk by design. We are proving the method before we move into deeper rationalization.”

Priya looked at me then. She knew the sentence. She had heard versions of it in security steering, platform council, automation review, and the quarterly operating committee. Behind her, one of the Communications people had a printed platform scorecard with the old DevOps colors still visible under the new AI labels. Kedran rarely threw away a safe template.

Three hundred and twelve had still taken more work than anyone outside the building would believe. Engineers traced dependencies through build files that pointed to servers nobody could find. Old scripts existed because a distributor in Ohio expected a file every night at 2:15. One retired service came back to life because Finance still pulled from it at month end, and another repo had secrets in its commit history old enough to have survived two reorganizations and one general counsel.

Aaron had warned me. “Dead is a legal status. In software, dead is usually a rumor.”

So I was not ashamed of the 312. Some repos were real risk. Some had no owner and no build. Some were obsolete in the way a locked storage room is obsolete until someone remembers the only copy of a contract is inside it. Retiring them was good work. Boring work, careful work, the kind of work that prevents a Saturday call instead of earning a Monday headline.

It was also less than one percent of the estate.

I knew that before Evelyn said it.

We moved from the conference room to board prep at eleven. Victor joined by video from a supplier visit in Charlotte, seated in a hotel business center with his tie loosened and his Kedran lapel pin catching the overhead light. Evelyn was there in person, which meant the room adjusted around her before she spoke.

Russell opened with the human story. He talked about builder culture, improved confidence, the dojo alumni network, and the pride of teams cleaning up the landscape they had inherited. Russell had become excellent at making large-company exhaustion sound like a community garden.

Chip handled the middle. He had a new phrase, “platform fluency,” and he used it nine times in six minutes. The Communications people looked at their copies of the deck with the faces of people watching an appliance smoke.

Then I took the final section.

“The point of Year Three,” I said, “is to make the learning durable. Year One gave us bounded proof. Year Two built broad capability and financial discipline around the program. Year Three is translating that into platform and governance patterns the enterprise can absorb safely.”

I advanced the slide.

312 repositories retired in first low-risk wave.

Victor nodded from Charlotte. “That will play.”

Mara glanced down. She did not like the word play, but she liked it when Victor knew what would calm a room.

Evelyn leaned forward. She had a capped pen in her hand. Evelyn did not take many notes. She used the pen like a metronome, tapping once when she found a weak beat.

“Thomas,” she said, “what is the denominator?”

I took a sip of water, not because I needed it, but because a pause can make a plain fact sound less evasive.

“Forty-seven thousand is the current repository count across the enterprise footprint,” I said.

She did the math in her head. I watched it happen.

“So the campaign retired about two-thirds of one percent.”

“In the first wave,” I said. “Designed intentionally around low operational risk.”

“Does the slide say first wave?”

I turned toward the screen as if I needed to check. It did. Barely. We had made it clearer after Aaron, but not so clear that the number lost its posture.

“We can make it more prominent.”

“Please do.”

Victor shifted on the screen. “The number is still meaningful.”

“It is,” Evelyn said. “I am not criticizing the work. I am asking what it proves.”

That was the dangerous kind of question: fair enough that nobody could treat it as an attack, sharp enough that every person in the room had to pretend to welcome it.

“It proves the operating pattern,” I said. “We now have a tested path for identifying, validating, archiving, and decommissioning dead or redundant assets without creating avoidable production risk.”

“Does it prove Kedran is getting faster?”

The room went quiet in a way that had shape.

“Not by itself,” I said.

It was honest enough to satisfy the moment and narrow enough to leave the larger story standing.

Chip stepped in because Chip had never met a silence he trusted. “What we are seeing, Evelyn, is that repo retirement is one thread in a larger productivity fabric. The dojo cohorts, the automated review, the pattern library, the tool consolidation, they all contribute to a broader shift in delivery posture.”

Evelyn looked at him until he stopped.

For half a second, Chip had my sympathy. He had given us a year and a half of visible motion, and I had used him exactly as I knew I would. He trained people. He annoyed executives. He kept the program bright and harmless. The work mattered. It also was not transformation.

“The fall board materials need to be careful,” Evelyn said. “If the benefit is risk reduction, say risk reduction. If the benefit is learning, say learning. If you are claiming throughput, show me throughput.”

“Agreed,” I said.

I meant it in the way executives mean agreed. I agreed with the principle. The implementation would need sequencing.

After the meeting, Mara walked with me toward the elevators.

“She is going to keep asking,” Mara said.

“She should.”

“That is not an answer.”

“I know.”

We stopped near the elevator bank. Through the glass, I could see the plaza below, the lunch trucks lined up along the curb, employees moving through the day with badges, phones, sandwiches, and problems I would never hear about.

Mara followed my eyes. “You are thinking about the people again.”

“I usually am.”

“Sometimes you use that as cover.”

She said it without cruelty. Mara was one of the few people at Kedran who did not require my better version.

“Sometimes,” I said.

The elevator opened. She stepped in, then held the door with her hand.

“Do not let Chip add an executive immersion SKU to this. Victor will ask what it costs, and then I will have to say no to a man with markers.”

“Noted.”

Priya was waiting for me in the platform room. Six screens showed build pipelines, ticket aging, tool adoption, policy exceptions, and a map of repo ownership that looked like a city drawn by committee. She had a paper cup of coffee and a stack of printed diagrams.

“I want Components,” she said.

No preface. No warmup. That was one reason engineers trusted her.

“For what?”

“A real operating-model pilot. One division. One value stream. Authority, architecture, backlog funding, release governance, metrics. Not training. Not tooling. Not pattern maturity. The actual work.”

I leaned against the table. “Celia is in distributor push.”

“Celia is always booked.”

“That is true.”

“Give me warranty triage in Components. Not plant scheduling. Not pricing. Warranty triage. It touches customers, but it will not shut down a plant.”

“It touches field service, claims, dealer portals, data retention, and the acquisition platform.”

“Exactly. It is real enough to matter.”

She spread the diagrams across the table. She had mapped the path from a field complaint to a warranty decision to a service recommendation to a parts quote. There were twenty-one handoffs in the current process if you counted the places where a ticket waited for someone whose job title had changed since the workflow was written. Priya had circled seven in red.

The page had numbers because Priya had learned how to make executives stay seated. Components averaged a little over six thousand warranty claims a month. The median time from field complaint to disposition was eighteen point six days, and Priya had marked eleven of those as waiting time rather than work time. Dealer credits over thirty days were running about two point four million a quarter, not all of it addressable, but enough that Mara would ask better questions than Victor. A two-day improvement would not save Kedran. It would, however, be visible to people who did not care what tool generated the code.

“These are decision delays,” she said. “Not coding delays. Not AI adoption delays. Decisions.”

I read the page twice. It was good work: specific, irritating, hard to dismiss. It also made clear why Kedran preferred tool adoption numbers. A tool number can be purchased, tracked, and celebrated without asking Celia to give up any authority. A decision delay has an owner.

“Who have you shown?”

“Mei. Aaron. Dana’s architecture security lead. Nobody in Celia’s office yet.”

“Good.”

She gave me a tired look. “That sounded like relief.”

“It was process awareness.”

“Thomas.”

I looked up.

“We have trained a thousand people to move faster inside a system that makes speed rude,” she said. “You know that, right?”

There are moments when a person asks for honesty and you can feel the whole building waiting to see how expensive the answer will be.

“I know the system is the constraint,” I said.

“Then let us change one piece of it.”

I wanted to say yes. That is the part I need you to understand. I liked Priya. I respected her. I liked the clean anger in her work. I liked anyone who could point at the place time disappeared and refuse to call it culture.

But it was August 14, 2029. Forecast lock was two weeks out. Components had a plant cutover in September. There was a rumor Victor would announce his retirement sometime after the annual meeting, which meant every business-unit president had started behaving like weather. My own equity vesting schedule sat in a folder at home that I opened more often than I admitted.

And there was no world in which I was going to spend my last clean year at Kedran forcing Celia into an operating-model fight over warranty triage.

“The timing is wrong,” I said.

Priya did not argue immediately, which was worse.

“Because of quarter close?”

“Quarter close. The plant cutover. The CEO transition rumor. Celia’s distributor commitments. Dana would need to embed someone. Finance would need to agree on funding treatment. Architecture would need to pre-approve exceptions. Field service would have to own the adoption path.”

She pointed at the seven red circles. “Those are the delays.”

“I know.”

“And you are going to call them dependencies.”

I folded my hands on the table. “They are dependencies.”

“They are decisions.”

She was right. I could have said that. Instead I said, “Both can be true.”

Chip found me later outside the cafeteria with a folder under one arm and a bright look that made me brace for expense.

“I’ve been thinking,” he said.

“That has historically been your process.”

He laughed because he thought I was joking. I mostly was.

“Dojo 2.0,” he said. “Executive Immersion. Two days. Small cohorts. Presidents, finance leaders, security, HR, architecture. We bring them into the work.”

He opened the folder. The first page had a title in large type: From Sponsors To Practitioners.

The second page had three prerequisites. Executive attendance. Real value-stream sponsorship. Provisional release authority for successful dojo slices.

Chip tapped the third line. “This is where the program either grows up or admits it is training.”

No executive at Kedran would attend twice. Half would send delegates. Celia would send someone excellent and powerless. The Controls president would have a customer escalation. Victor would come for twenty minutes, give a warm quote, and leave before the hard exercise. Chip would call it traction because Chip was not dangerous enough to be stopped quickly.

The release-authority line changed that, but only in the administrative sense. It gave me the phrase I would need later. Chip remained valuable as an external advisor, but his recommendations had moved ahead of Kedran’s governance readiness. That sentence would let me thank him, reduce his role, and make the reduction look like maturity instead of removal.

“Interesting,” I said.

Chip brightened.

“Let’s socialize it carefully.”

That was one of my cleaner phrases. It meant no, but not today. It meant give the idea a place to sit until urgency found another room.

That evening, before I left, I stood alone in my office and looked at the board deck again. The 312 slide had been revised.

312 repositories retired in first low-risk wave.

Better. More honest. Still useful.

Outside my window, the city was turning gold in the late sun. I could see the tops of buildings, the flat roofs, the traffic folding itself around downtown. Somewhere down there, the Doverline Youth Program was closing for the day. Children were being dragged toward parking lots by tired parents. Somebody was wiping marker off a table. Somebody was rebooting a loaner laptop.

I did not know yet that I would end up there, taking meetings in smaller rooms where a broken projector was still allowed to be the whole problem. I only knew that three hundred and twelve looked strong if the denominator stayed in the speaker notes, so that was where I left it.

Chapter 12: The Number

Late 2029

Executive Play: Retire At Peak Narrative Value

My financial advisor did not show me a dashboard.

That surprised me. Wealth managers loved dashboards the way technology executives loved roadmaps: stacked bars, probability fans, polite colors, and numbers arranged so fear appeared to have accepted professional supervision. He had three sheets of plain white paper, a conference room with new carpet, and a pen he kept aligned with the edge of the table.

“If Kedran lands where we expect,” he said, turning the first page toward me, “and if the retirement package is handled cleanly, this is the figure.”

He tapped the bottom line.

I had spent most of my career telling people that numbers did not explain themselves. Revenue needed backlog. Headcount needed skill mix. Delivery metrics needed scope, cycle time, release authority, and a denominator no one had sanded smooth for the board. This number needed none of that. It sat on the page, black and quiet, and separated the life I had been defending from the life I could afford to choose.

Private-island money, no. People outside my income bracket assume executive compensation becomes private-island money if you leave it alone long enough, but most of it becomes taxes, tuition, insurance, family help, market timing, the old cabin you cannot price like an asset, the club you joined because the city did business there, and the quiet cost of being the person everyone expects to be able to help.

The number was still enough.

Enough for Hannah’s remaining medical school years. Enough to keep the Ludington cabin my grandfather built, with its uneven porch and fish-cleaning table, without turning it into a memorial to my own inability to let go. Enough to help Lauren and the baby without making the help feel like a board appropriation. Enough for Claire to choose comfort without asking whether I needed to stay one more vesting cycle.

Enough to stop waking at 3:40 in the morning with a sentence from an Audit Committee pre-read already forming in my mouth.

The advisor gave me the practiced silence of a man who made his living around other people’s relief. He had a calm face, expensive cuffs, and the moral flexibility required to help anxious rich people describe anxiety as allocation.

“The key,” he said after the silence had done its work, “is getting through the vesting date and preserving the transition narrative.”

“Preserving the narrative.”

He smiled, not hearing the phrase the way I heard it. “You have an unusually strong story right now. The AI program, the business press, the board visibility, Victor’s confidence. It positions retirement as a natural handoff after a major foundation has been built.”

He slid the second sheet across the table. It was a printout from the Midland Business Ledger, folded with the care of a man who still believed paper made press coverage more real.

Kedran Builds Responsible Path For AI-Enabled Software Delivery

I had read it twice before breakfast the day it ran and then pretended, even to myself, that I had not. Russell had sent it with a note that said Great external validation, which meant Communications had successfully fed the reporter three phrases, two names, and one CEO quote polished until it reflected nobody in particular.

There I was in the third paragraph, speaking about responsible acceleration, enterprise guardrails, and modern software productivity. Chip got credit for the scaled dojo model. Mara got a sentence about disciplined investment. Victor got the final quote because CEOs get the final quote unless they are under indictment or dead.

The advisor had highlighted my paragraph.

“This helps,” he said.

I wanted to tell him that the words made me tired. The scaled dojo model had trained good engineers faster than the system deserved and then returned most of them to approval queues that outlived the people who created them. Platform friction reduction meant we had removed two gates from low-risk applications while the revenue systems stayed behind glass. The responsible AI path had become a corridor of policies, dashboards, tool licenses, review boards, and executive language broad enough to shelter every hard thing we had not changed.

Instead I said, “Good.”

He nodded. “Very good. The clean story matters for board goodwill, advisory opportunities, nonprofit roles, future compensation if you want it.”

“I do not think I want another large-company job.”

He paused. For once that morning, he looked genuinely surprised. “You may feel differently after six months.”

“Maybe.”

“People at your level often underestimate the loss of platform.”

Platform. Even retirement had an architecture diagram now.

I looked at the number again. No green tile. No survey score. No footnote explaining why the benefit would arrive in the next phase if the organization stayed committed. Just a hard figure and the tax assumptions beneath it.

“What happens if I leave before the next fiscal-year metrics mature?” I asked.

“From a financial perspective?”

“From a story perspective.”

He did not need long. “That may be ideal. You leave after visible progress and before the harder comparisons. That is not cynical. It is timing.”

Most cynical things in my world had professional synonyms.

Timing. Sequencing. Maturity. Absorption.

I folded the paper once and put it inside my jacket. The advisor pretended not to notice that I handled it like a medical result.

On the way back to Kedran, construction pushed the car north, and the Doverline Youth Program came into view with its winter banners snapping against the light poles. I had a two o’clock with Mara, a four o’clock with Russell, and an inbox full of congratulations for an article that had made our work sound cleaner than it was. I asked the driver to pull over for ten minutes.

The air outside had that hard November brightness Central Indiana gets when the cold has not yet become routine. A yellow school bus idled at the curb. Two teachers in puffy coats counted children with the grave concentration of people responsible for treasure in mittens. Inside the lobby, a sign-in tablet chirped at each badge, and the line moved in uneven little surges as parents found phones, gloves, permission slips, and children who had slipped toward the robotics table.

One of the volunteers at the desk saw me and raised a hand.

“Tom V.,” she said. “Here to inspect the robots?”

“Only if they have formed a committee.”

She laughed. She had retired from the public schools and could silence a room by looking disappointed. She had once asked me for a donation increase in the same tone she must have used on eighth graders who believed a fire drill suspended homework.

I stayed near the entrance, close enough to smell wet coats and floor polish, long enough to watch a boy with glasses press both hands against the glass around a model train. He whispered to his father, and the father bent down as if the sentence might contain a secret worth hearing.

The confession people wanted would have been easier if I did not believe in Kedran. I believed in factories, repair networks, invoices that cleared on time, service promises kept to a farmer, a plant manager, or a hospital maintenance lead who needed a part to arrive Tuesday because Tuesday was not a metaphor to them. I believed in the old machines we kept alive and the people who knew which vibration mattered.

I also believed in this.

A sign-in tablet working. A child getting in. A teacher not losing a permission slip. A donor database accurate enough that the next semester got funded. There was no analyst call around it. Nobody gave guidance on wonder.

My phone buzzed.

Mara: Where are you?

I typed: On my way. Traffic.

That was mostly true. I was always in traffic. Sometimes it was on a road.

Victor wanted the retirement announcement to feel planned without feeling final. Mara wanted the timing tied to the completion of the three-year AI foundation, but not so tightly that anyone could ask whether the foundation held weight. Russell wanted a video, an internal Q&A, a photo package, and a farewell message in which I looked reflective without sounding terminal. I asked him to keep the video under four minutes, which he received as a personal injury.

By six, I was home. Claire sat at the kitchen table with a stack of envelopes, her laptop, a yellow legal pad, and a glass of wine placed far enough from the papers to show she had already assessed the risk. She had the expression she used when a problem had numbers in it.

“Your program sent the gala pledge reminder,” she said.

“Our program.”

“I am not the one who buys tables because a development director says your name in a disappointed voice.”

“That is an unfair description.”

“It is an exact description.”

I hung my coat in the hall and kissed the top of her head. She let me, then pointed to the chair across from her.

“Sit.”

Claire had run budgets for a hospital system before our children were old enough to remember her having an office. When she left paid work, people called it staying home, which was a phrase invented by people who had never watched a household, three aging parents, two children, school boards, medical forms, and a spouse’s executive calendar route through one person. She read spreadsheets the way other people read weather.

I gave her the folded paper.

She read it once, then again. Her face changed only around the eyes.

“This is the number?”

“That is the number.”

“If the retirement goes the way he thinks?”

“We are safe.”

She set the page down. “Safe is not a small word.”

“No.”

The refrigerator clicked on. Upstairs, the old house made one of its winter noises, a pipe or joist complaining in language older than any of my systems. Claire looked at the paper and then at me.

“What about Kedran?”

“What about it?”

“Is the work done?”

I rubbed my thumb along the fold in the page. “Work like this is never done.”

She gave me the look she reserved for sentences that hid behind their own reasonableness.

“That is convenient.”

I smiled because habit reached my face before judgment did. She did not smile back.

“It is also true,” I said.

“Both can be true.”

She had been listening to me too long.

I leaned back. My knees hurt, which had become one of the body’s small insults, a running commentary on conference rooms, flights, stadium stairs, and years spent standing at receptions while pretending the third shrimp on a toothpick counted as dinner.

There was a sentence I could have said then. I could have said I was never going to fix Kedran. I could have said I had been trying to get through it, not transform it, and that the difference had been clear to me long before the program had a name.

I did not give Claire that sentence.

Once a sentence like that leaves your mouth, it stops being a thought and becomes a thing another person has to carry. Claire had carried enough of my careful language. She did not need the naked version just because retirement had made me safe enough to say it.

“Nobody fixes a company that size in three years,” I said.

“That is not what I asked.”

“I know.”

She waited. That was one of Claire’s better weapons. Silence, honestly used.

“My job was to keep it steady while the next version of the work became possible,” I said. “Keep people employed where I could. Keep Victor from making a promise he would regret. Keep Mara from booking savings no one could deliver. Keep Dana from being rolled by people who discovered risk after the damage had already escaped. Keep engineers from being pushed into another program that blamed them when the operating model refused to move.”

Every word was true.

It was also the version with the family math removed.

Claire did not let me stop with Kedran. She named the life around it: the cabin, the club, Hannah’s tuition, and the version of me Central Indiana knew.

“Those mattered too,” she said.

I looked toward the window, where the kitchen and my reflection occupied the same dark glass. “Yes.”

It was the cleanest answer I could give without handing her the uglier one. The life had mattered more than I had allowed myself to say in any room with fluorescent lights, and the folded number on the table had made that private accounting visible in black ink.

Claire took a sip of wine. “That is the closest thing to a human answer you have given me in months.”

“I do not know if it improves the picture.”

“It makes the picture look like you.”

I breathed out.

“Nobody was going to fix Kedran in three years,” I said. “Not me, not Lena, not Chip with a wall full of markers, not a board mandate written after two dinners and a benchmarking packet. Not with tens of thousands of repositories, a run of acquisitions, plant systems nobody wanted to touch, and a public-company calendar that could turn every hard thing into next quarter’s problem.”

“You believe that.”

“I do.”

“And it helps you.”

“It does.”

She nodded. “That is the hard part.”

“Maybe ten years from now it gets bad enough,” I said. “Maybe the cost of delay becomes larger than the cost of the fight. Maybe a customer leaves, a plant misses hard enough, a board member loses patience, or the tools become so capable that pretending the bottleneck is code finally sounds ridiculous even in a pre-read.”

Claire waited.

“Or maybe by then AI can read the board book and tell directors what the company has been avoiding.”

“Would they listen?”

I pictured it for longer than the joke deserved: an AI board member with perfect recall, no golf calendar, no favorite plant president, no patience for appendix language, asking for the denominator every time Russell made a number large enough to stand alone. It would be terrible company at dinner and probably the first director Evelyn ever liked.

“They would ask it to make the answer less disruptive,” I said.

Claire smiled then, not because it was funny enough, but because it was sad enough.

I had watched Kedran inherit the future twice before AI arrived. Agile was real. DevOps was real. The people who brought them in were serious, often right, and occasionally brave. My predecessor watched both get absorbed into ceremonies, coaches, dashboards, tooling standards, governance language, and maturity scores while funding, architecture, release authority, and business-unit power stayed protected in the same rooms. AI was faster and more visible, but the company had learned the digestion process.

“AI was the same problem with better demos,” I said.

Claire turned the folded page with one finger. “And one better exit.”

I did not answer quickly because the answer was yes.

When the kids were young, we used to take the ferry out of Ludington, not far from the cabin my grandfather put up before anyone in our family used words like portfolio. The place was part fishing shack, part stubborn carpentry lesson, with soft pine floors, a screen door that slapped too hard, and a smell of lake water and old tackle that returned every time we opened it in May. I had always liked the ferry for reasons that became more convenient as my career got older. It was outdated, expensive, beloved, useful, and too heavy to be confused with an idea. It crossed because people still bought tickets and freight still needed water under it, and because replacing the old power plant would have looked simple only to someone who did not have to make the numbers work. I used that image so often in my head that it became permission. Old systems keep moving. They matter because they move. Tending them is not nothing.

But caretaking can become a beautiful word for staying put.

My phone rang on the table. The screen showed the executive director of the Doverline Youth Program. Claire glanced at it.

“Our program,” she said.

I answered.

The program director skipped the ceremonial part. She was good at that. “Tom, do you have five minutes?”

“For you, yes.”

“I heard about Kedran. Off the record, at least for now. Congratulations, if they are allowed before announcements.”

“Careful congratulations are accepted.”

She laughed once. Then the lobby tone left her voice.

“I need advice,” she said. “Maybe more than advice.”

The program’s technology situation had crossed an invisible line. Student check-in was brittle. Family records lived in three places, each one confident and each one wrong in a different direction. Classroom Wi-Fi was handled by a vendor that treated Saturday outages like Monday problems. Education staff were using consumer AI tools because worksheets, language variations, and activity prompts did not wait for a governance model. Facilities owned too many passwords. The part-time consultant who knew the server closet had taken a job in Cincinnati and left behind a diagram that looked like a treasure map drawn during turbulence.

“We need a CIO,” she said. “A real one. Not a large-company version. Someone who can make technology boring where it should be boring and useful where it should be useful.”

I looked at Claire.

“That sounds smaller than what I do now.”

“It is,” she said. “Radically. That is why I hoped you might tell me who would be right for it.”

Claire watched me over the rim of her glass. The number lay between us on the table, still folded, still enough.

Outside, a car moved down the wet street. Inside, I could hear the furnace, the refrigerator, and the program director waiting through the line.

Chapter 13: The Handoff

2030

Executive Play: Hand Off The Unfixed System

My retirement announcement went out on a Tuesday morning in January.

Russell had wanted Thursday. Mara wanted Tuesday because markets dislike drama, and Tuesday gives people two working days to ask questions before inventing better ones. Victor agreed with Mara, which meant Russell discovered the communications wisdom of Tuesday by 8:15 and sent a revised calendar invite with no visible injury.

The email was warm, brief, and built from phrases that could not be sued by history. It said I had modernized Kedran’s technology foundation, strengthened cybersecurity and resilience, integrated acquisitions, and led the company’s responsible path into AI-enabled software delivery. The words did what they were supposed to do. They made a complicated life sound linear.

At 9:04, my phone started vibrating and did not stop. Former employees. Current employees. Board members. A plant manager from Ohio who once called me at midnight because a label printer had become a revenue issue. A network operations manager. Someone from the program lobby, who texted, Retired means more robot inspections, yes?

That one made me smile.

At ten, Victor called from his car. I could hear the road noise under his voice and the light impatience of a man being driven between obligations.

“You put us on a responsible AI path, Tom,” he said. “I mean that.”

I looked through my office window at the winter plaza below. People in dark coats moved fast between doors, heads down, badges swinging. Bare trees made the headquarters look more honest than it did in summer.

“We put a path in place,” I said.

“You always make it smaller when someone thanks you.”

For a second I almost gave him the version that belonged in a room without phones: that we had spent roughly one hundred ten million dollars over three years and produced trained engineers, better policies, cleaner tooling, useful local automation, several hundred retired repositories, a press story, and no macro company benefit I could defend if Evelyn kept asking follow-up questions. I could describe pockets of improvement. I could not prove Kedran was faster in the way the company needed the word to mean.

That was the final safe zone: enough true progress to thank people for, enough unresolved truth to leave for the next phase, no sentence sharp enough to force the company to choose in public.

Instead I said, “Thank you, Victor.”

He accepted the thanks and moved on. That was another executive skill.

Russell’s tribute video arrived for review that afternoon. It opened with old photos of me in Kedran hard hats, then cut to engineers, plant managers, and executives saying kind things under lighting that made every room look like a deposition. Lena appeared in a short clip from 2027. Chip stood in front of a dojo wall covered in sticky notes and looked as if retirement were a framework he had helped invent. Mara spoke for eleven seconds and made restraint sound affectionate.

Then came Priya.

She sat in the platform room with a closed notebook in front of her and the pipeline monitors glowing behind her.

“Tom listens,” she said. “That matters in a place this big.”

That was all. Priya had given Russell a sentence no one could object to and no one could mistake for surrender.

I approved the video and asked him to cut forty seconds of Chip. He asked which forty. I said, “Your choice.”

Priya came to my office two days later.

She did not sit at first. She stood near the door with a notebook under one arm, as if taking the chair might make the conversation softer than she wanted.

“I am not here to talk you out of leaving,” she said.

“I appreciate that.”

“I think you should leave.”

“I appreciate that less, but go on.”

Priya did not smile. She sat then, crossing one ankle over the other, notebook still closed.

“The transition memo matters.”

“I know.”

“Name it plainly.”

I waited, though I already knew where she was going and what it would cost to let her arrive there without interruption.

“Tools and training are not enough,” she said. “The next CIO needs to know that. Not between the lines. Not as maturity language. Say the operating model has to change. Funding. Architecture. Security review. Release governance. Ownership. Metrics. Authority. Say the program did not change those things at enterprise scale.”

The sentence sat on my desk like a bill.

I had always liked practical people because they saved time. I had also spent a career surviving them by turning their nouns into safer ones.

“I will put it in the memo,” I said.

She looked at me.

“Will you?”

“Yes.”

“Thomas.”

People used my full first name when they wanted me to stop performing. Claire did it. Mara did it when she was tired. Priya had learned.

“I will put it in,” I said.

The transition memo took three days because every honest sentence had to pass through the part of my mind that knew how companies used honest sentences against the people still employed inside them. A memo can be a map, but it can also be a weapon. A bored board member could turn precision into indictment. An ambitious successor could use it to humiliate the people whose help he needed. Finance could ask why we had spent what we spent. Legal could ask why the risk had been so clearly understood and so politely carried.

My first version said:

AI-assisted development improved local team capability, but Kedran has not materially changed end-to-end delivery throughput because funding, architecture, security review, release governance, and business ownership remain fragmented across divisions.

I printed that sentence and left it on my desk while I took a call from the program director about check-in tablets and family-record exports. When I came back, the page looked too honest for the people who would have to live under it after I left.

The final memo said:

The next phase should continue maturing operating model integration across funding, architecture, security, and delivery governance.

Chip read it and called it “a strong maturity bridge.”

Of course he did.

Priya read it and did not answer my email for six hours. When she did, she wrote: I understand.

That was worse than anger.

The next morning she came by before my first meeting. No notebook this time.

“I will stay through the transition,” she said.

“I am glad.”

“Do not be. I am staying because if I leave now, the pattern library turns into a shared drive folder and the next group gets blamed for not making the miracle repeat.”

There was nothing in that sentence I could improve.

“What do you need from me?” I asked.

“Do not leave them thinking effort is supposed to replace authority.”

I looked down at the memo on my desk. It was printed on good paper. Kedran had always respected good paper.

“I tried to say enough.”

“I know,” she said. “That is the problem.”

My final board meeting was held on a wet Thursday in February. Rain moved down the windows in thin lines. The room smelled like coffee, wool coats, and the lemon cleaner facilities used on the big table. Victor thanked me first and spoke about resilience, cybersecurity, acquisition integration, pandemic response, and the AI program. He was generous. Victor could be selfish with decisions and generous with praise, a combination that gets a person further than it should.

Mara spoke next. “Tom leaves us with a mature foundation,” she said. “More importantly, he leaves the next team with discipline.”

I looked at her. She did not look back, which meant she knew exactly what she had done. Discipline was true. Foundation was convenient.

Evelyn waited until the end.

“Tom,” she said, “one question.”

Several people shifted in their chairs. Even Victor looked down at his notes.

“Is Kedran faster?”

No adjective. No safe noun. No phrase I could improve by adding context.

I folded my hands on the table. A dozen answers were available. I could talk about pockets of acceleration, developer confidence, test generation, tool consolidation, policy maturity, and the difficulty of measuring throughput across acquired systems with incompatible release paths. I could tell the electricity story again, which had become so useful to me that I trusted it less. I thought of Lena asking whether I wanted the pilot to work. I thought of Priya saying they can write the code now; they still cannot merge it. I thought of Dana carrying every unsolved risk until the rest of us called caution a bottleneck. I thought of the advisor’s number. I thought of Claire saying both can be true.

“In places,” I said.

The room did not know what to do with that.

Evelyn nodded once.

“Thank you,” she said.

My last day at Kedran had cake at three.

I shook hands. I hugged people who wanted hugs. I let Russell play the video. I accepted a framed photo of the headquarters building taken at sunrise, which was kind and impossible to store. A network operations manager pressed a folded note into my hand. Dana gave me a bottle of bourbon and said, “Do not make me the bad guy in your retirement stories.” Mara told me not to take a board seat for at least six months. Victor put both hands on my shoulders and said, “You earned this.”

Maybe I had. Earning and escaping can look similar from the outside.

Claire drove me home because she said retired men should not operate heavy machinery on the first day. I told her the car weighed less than my last budget spreadsheet.

For the first time in three years, the next Monday on my calendar did not belong to Kedran.

It belonged to the Doverline Youth Program.

The title was still CIO, because institutions love familiar boxes even when they cannot afford the furniture that usually comes with them. The office was a converted storage room behind education, with one window, two filing cabinets, a chair that leaned slightly left, and a whiteboard someone had used months earlier to plan a summer camp supply order. I had eight people in technology if I counted the part-time help desk technician who also knew how to restart three robotics carts by touch, plus two vendors I did not yet trust and one volunteer who had written the original family-record export and was now considered infrastructure.

At Kedran, my first hour on a Monday could involve a board pre-read, a cyber insurance question, a budget bridge, two acquisition systems, three meeting links, and a chat channel created by people who did not want to wait for governance. At the program, my first hour involved a sign-in tablet that worked if you held it at an angle, a classroom printer that had become the enemy of fourth grade, and a family report that counted one student twice because his last name had an apostrophe.

No one asked for a transformation roadmap.

The director walked me through the building before opening. The place had the strange peace of an after-school center before children arrive: polished floors, sleeping classrooms, lights warming up, staff moving with keys and radios and coffee. A facilities manager showed me the network closet with the expression of a man presenting an injured relative.

“It is not as bad as it looks,” he said.

I looked at the switches, the unlabeled cables, the battery backup with a warning light, and the handwritten note taped to the wall that said DO NOT UNPLUG BLUE.

“That sentence has done a lot of work in my life,” I said.

By three-thirty, children were coming through the doors in bright bursts, and the sign-in tablet failed in front of a robotics group from the east side. No steering committee formed. No one asked whether the check-in failure should be treated as a digital family-experience modernization opportunity. The program manager looked at me and said, “Can you help?”

I could.

I got on the floor in my good pants, found the loose connection, swapped a cable, and asked the help desk technician to order three backups before the end of the day. Then I opened the AI tool with the education lead and showed her how to turn one science activity into three reading levels and a Spanish version she could hand to a bilingual staff member for review instead of asking a tired teacher to start from a blank page at nine at night.

“Is that allowed?” she asked.

“It is if we write down what we are doing and keep a person in the loop.”

“Do we need a committee?”

I thought about Kedran, about the millions of dollars, the green dashboards, the dojos, the charters, Chip’s two smartwatches, Lena’s blue pen, Priya’s closed notebook, Evelyn asking for the denominator, and all the ways a company can turn a tool into a mirror and then argue about the reflection.

“Not for this,” I said.

That afternoon, I sat with the whole technology team around a table that had glitter in its seams. We made a list of the work that mattered most: stable check-in, clean family records, safer classroom networks, better device inventory, fewer shared passwords, a simple AI policy for education staff, and a way for weekend support calls to reach a human who knew the building.

Eight people. One table. Work close enough to a child at the door that nobody could hide the outcome in an appendix.

It was not easy. Small places have politics too, and budgets that make every purchase feel personal. The program had donors, grants, committees, old vendors, favorite classes, and staff who had learned to route around systems because the systems had routed around them first. I was not walking into purity. I was walking into a place where the distance between a decision and its consequence was short enough that my old habits had less room to breathe.

Near five, the director stopped by my office.

“How was your first day?” she asked.

I looked at the whiteboard. The summer camp supply list was still there beneath our network priorities because I had not wanted to erase it yet. Glue sticks, batteries, gallon bags, markers. Useful things, named plainly.

“Boring,” I said.

She smiled. “Good?”

I thought of Claire, Hannah, Lauren’s baby, the cabin near Ludington, the ferry still crossing because the economics of old machinery are rarely improved by lectures, and the fifteen hundred people I had tried to keep steady inside a company too large to turn just because the future had arrived ahead of schedule.

“Very good,” I said.

Then I shut down my laptop, put the backup cable order in my bag so I would not forget it, and went home while there was still daylight.