01
Public company, fourteen engineers
Bill manages a software team at a public company. Nine years there. Reads pull requests on the weekend because he cannot help it.
Slide 01
A procurement mistake gave 300 engineers unlimited model access for five business days. Bill's team shipped four-to-six months of normal output. The variance report flagged a $38,000 overage against a $4,000 line. By Monday the rate limits were back on, a steering committee was scheduled, and Bill was not invited.
Slide 02
Bill's team noticed by Tuesday afternoon. By Wednesday morning, the four most senior engineers had Slacked each other and agreed not to file a ticket.
They did not abuse it. They worked. They knew the window would close.
Bill manages a software team at a public company. Nine years there. Reads pull requests on the weekend because he cannot help it.
A new enterprise contract with one of the major model vendors went into effect. The per-seat caps did not.
The four most senior engineers Slacked each other and agreed not to file a ticket. They knew exactly what was about to happen.
Slide 03
The kind of work that had been sitting in the backlog long enough that nobody bothered to re-estimate it.
Not a quick patch. Actually fixed. The kind of work nobody volunteers for.
With tests, docs, and a runbook. The estimate was not wrong for a human team. It was wrong for the new world.
The agent let her draft, validate, and walk the diff across four service repos in a single afternoon.
Slide 04
"Even on the most cynical view, even if half had to be redone, the company traded six months of fourteen engineers' time, roughly two and a half million dollars of loaded labor, for $38,000 in tokens. And then they shut it off."
Bill, doing the math out loud while pushing his daughter on the swing.
Has not written code in nineteen years. Did not open a single pull request from the week. Forwarded the variance report with one line: "We need to start measuring stories in tokens."
Steering committee. Intake process. Dashboard. The org's reflex against an unexpected outcome is to build more org.
This is the part Bill keeps coming back to. The information is not the constraint.
Slide 05
The $38,000 overage is, in their world, the exact thing procurement exists to stop. They are doing their job correctly.
The CISO does not get a bonus for refusing the higher seat tier. What they have not been given is a fast lane to say "yes, with controls." So the default answer is the slow one.
Approving requires due diligence. Denying does not. Denying is how finance hits its SLA on time. The calendar wins.
None of these scorecards roll up to whether the company accomplished the outcome that justified the planning cycle in the first place.
Notice who is not in the room. The CTO who is supposed to be shipping faster than the smaller competitor. The CFO whose quarterly earnings track engineering throughput. The CHRO who is about to absorb the retention hit. The CEO who put "AI-driven productivity" in the last earnings call as a 2026 priority. All four of them are downstream of the no.
Slide 06
Slide 07
Once the frame is named publicly, the no starts to look like what it is.
Chris never got a yes or an apology. He just got the RAM.
A justification deck that took two weeks. A benchmark that took longer to compile than the build itself. Every signature routed to another signature.
Senior engineers, ten years past the point where any of this should have been a conversation. Same fight, different decade.
Bill is still on the asking-politely side of the no. He does not yet know it is the move. The only weapon that has ever worked is to name the frame in public.
Slide 08
They have asked Bill, very precisely, what would constitute a violation. Each of them is building something in the evenings on the same class of AI tooling the company will not buy them.
Six engineers asking lawyer-questions in one quarter is a signal.
They keep their personal accounts off company hardware. They do not want to get fired. The data risk the company feared is not the risk the company has.
More questions about the moonlighting policy this quarter than in any quarter of the last nine years. None has asked Bill for permission. They have asked what would constitute a violation.
The RTX 5090 he is shopping for has nothing to do with work. It is for the side project he has been thinking about for a year. He is the engineering manager. He is not going to file a moonlighting question, because filing one would tell HR he is preparing to leave.
Slide 09
Slide 10
$20,000 cash comp plus benefits, RSUs, payroll tax, laptop, on-call stipend. Fourteen of them. About $4.7M a year, fixed, before a single feature ships.
$500 per engineer per month. $7,000 a month for the team. The defensible claim: 25-40% productivity lift on the slice of an engineer's time that is actually shipping new code. That is what the $84K buys.
Baseline attrition runs 8-15%. On fourteen people, Bill loses one or two seniors a year regardless. The denial does not change the rate. It changes the mix. The ones who leave are the $400,000 ones.
Why does your $28,000-a-month engineer not deserve $500 a month in tools? Is it because $500 a month feels like a lot of money in a budget meeting and $28,000 a month does not, because the $28,000 sits on a different line item that is treated as fixed?
Slide 11
Sponsored line items survive review. Re-litigated line items do not. If the CTO will not sponsor it, that is also information.
Your company has, in its own usage logs, the most expensive A/B test it will ever run. The control group is every other week of the year. The treatment group is five days. The result is in the variance report finance already wrote. Use it.
"Eight Is Not Enough" worked because it became a phrase the org could not unhear. Find the equivalent for your $500 line item. Put it in writing. Put it in meetings you are not running.
You did what you could within the constraints. The consequence is going to show up in Q3. You wanted it on the record. That is what doing the job looks like when the system will not yield.
Slide 12
The thing he was describing was not a productivity gain. It was a permission. Your $28,000-a-month engineers are buying their own tools on the weekend with their own money. That is not loyalty. That is them building the resume they will use to leave.
The data risk you said no to has not gone away. The talent risk you traded for it is compounding every Saturday morning. The earnings call promised AI-driven productivity in 2026. The variance report told you what AI-driven productivity looks like. The denial is on file.