Alex has been trying to hire an engineer for three months. Not just any engineer. A specific person who works at a company that ships features in days while his forty-seven person team takes weeks.
He has watched them launch three features in the time his team shipped one. Same complexity. Same domain. They just move faster.
He knows why. They do not wait. They use AI agents to generate complete specifications, implementation, tests, and deployment configurations in a single pass. They ship with one other person in two to three days.
His team? Each engineer needs eight other people to ship one feature. Product Managers write specifications. Designers create mocks. Quality Assurance tests separately. Security reviews separately. Development Operations deploys separately. An engineering manager coordinates all of this.
Eight people. Seven weeks. One feature.
He wants to hire engineers who eliminate that entire coordination chain. Look. Human Resources said no. Three times.
In his first attempt, he focused on the market rate. He identified the engineer in May. They were interested. Compensation expectations were two hundred fifty thousand dollars base plus a twenty-five percent bonus.
His Head of Talent pulled benchmark data. A Senior Software Engineer in our market averages one hundred forty-five thousand dollars total compensation. Your budget is one hundred thousand dollars base plus bonus. That is already above market for our region.
He tried explaining this was not a normal senior engineer role. This person eliminates the entire Software Development Life Cycle coordination overhead. They ship complete features with one other person, not eight.
She looked confused. What do you mean eliminates the Software Development Life Cycle? Everyone goes through our process.
That is the point. They do not need the process.
We need to maintain internal equity, she said. If we pay one engineer three hundred twelve thousand dollars, everyone will expect the same.
He tried the math. We have forty-seven engineers at one hundred thousand dollars, that is four point seven million dollars. Add Product Managers, Quality Assurance, security, Development Operations, managers, and the total delivery cost is about seven point seven million dollars annually. We shipped twenty-four features last year. That is three hundred twenty thousand dollars per feature.
With twelve of these engineers at three hundred twelve thousand dollars, total delivery cost would be four point two five million dollars. They would ship two hundred forty features. That is seventeen thousand dollars per feature. We would save three point five million dollars and ship ten times more.
She wrote it down. I will take this to the compensation committee.
Two weeks later he presented to the committee. They said the role does not exist in market data. How do we know this is real?
Because startups are paying it. The company this engineer works at just raised a Series B. Twelve people. Twenty-three million dollars in annual recurring revenue. That is one point nine million dollars revenue per employee. We are at seven hundred fifty thousand dollars per employee with eight hundred people.
We are not a startup. We need to maintain internal equity and sustainable cost structures.
Decision. Denied. Counter-proposal. Hire three engineers at one hundred thousand dollars each. He took it.
So, what happened next? Six months later those three engineers have shipped six features total.
Each one needs a Product Manager to write requirements for two to three weeks, a designer for mockups taking one to two weeks, code review with a two to three day turnaround, a Quality Assurance cycle for one to two weeks, security review for a week, and Development Operations deployment with a three to five day wait for scheduled windows.
The coordination overhead consumes forty percent of their time. They are productive. They are good engineers. They are just trapped in a system that requires eight people and seven weeks to ship anything.
The engineer he could not hire? Joined a competitor.
Last week that competitor launched a feature that has been on his roadmap for four months. They built it in nineteen days.
In August, he went back to the compensation committee for a second try with better data. He brought numbers he had been tracking all quarter.
His three new engineers cost three hundred thousand dollars in salary. Their allocated share of Product Managers, design, Quality Assurance, security, Development Operations, and management added another four hundred thousand dollars. Total cost seven hundred thousand dollars to ship six features. That is one hundred sixteen thousand six hundred sixty-six dollars per feature with an average cycle time of seven point two weeks.
For agent-native engineers he calculated three hundred twelve thousand dollars per engineer with minimal supporting costs. Three engineers would be about one million dollars total to ship a projected sixty features based on competitor velocity. That is sixteen thousand six hundred sixty-six dollars per feature with a three to four day cycle time.
This is still theoretical, the committee chair said. We do not have proof these engineers would perform this way in our environment.
Let me run an experiment, he said. Hire one. Give me ninety days. Track features shipped, cycle time, cost per feature, and people required. If it does not work we spent seventy-eight thousand dollars on a definitive answer.
What if it works and then everyone wants the same compensation?
Then we have a model that ships features at sixteen thousand dollars instead of one hundred sixteen thousand dollars. That is a problem I want to have.
They escalated it to the Chief Financial Officer.
Two weeks later, he sat down with the Chief Financial Officer. She was concerned about precedent. If we pay one engineer three hundred twelve thousand dollars we will have retention issues with the others.
We have retention issues now, he said. I have lost two senior engineers in the past quarter to companies paying two hundred fifty thousand dollars and up. We are losing people because our delivery process is frustrating. Engineers do not want to spend forty percent of their time waiting for approvals.
She asked what happens if we do not do this.
We continue shipping twenty-four features per year at three hundred twenty thousand dollars per feature while competitors ship over two hundred features at twenty thousand dollars per feature. Our cost structure becomes unsustainable and our product falls behind. In two years we are explaining to the board why we are losing market share despite having ten times the engineering headcount of our competitors.
What is the ask?
Authorize one hire at two hundred fifty thousand dollars to three hundred fifty thousand dollars base plus a twenty-five percent bonus. Bypass the normal compensation process. Give me ninety days to prove this works with data. If it does not, I will own it.
She looked at the quarterly numbers. They had missed three delivery commitments in the second quarter.
One hire. Ninety days. You report results to me weekly. If this does not work we do not do it again.
Approved.
Now, it was time to get it done. He did not go through recruiting. He called an engineer he had identified at a fifteen person startup. They have been shipping AI-powered features consistently for eighteen months.
How are you shipping complete features in two to three days?
We use agents to generate complete specifications with all the edge cases, security requirements, and test scenarios up front. Then the agent implements everything in one pass. Code, tests, documentation, deployment configurations. I validate the specification is complete, run the tests, and deploy. Most features take twelve to sixteen hours of actual work across two to three days.
How many people do you need to ship a feature?
Me and one other person to review the specification before implementation. That is it. No separate Quality Assurance, no security review meetings, no deployment tickets. It is all in the specification.
What would it take to do that here?
You would need to let me work directly with product on specifications. No user stories, no sprint planning. We generate the complete specification together with the agent in two to four hours. Then I ship it. If your process requires eight handoffs this will not work.
What if we create a two-person team that bypasses the normal process?
Then I am interested.
He offered three hundred twenty-five thousand dollars base plus a twenty-five percent bonus. They started in September.
We are now sixty days in. The agent-native engineer has shipped fourteen features. Average cycle time is two point nine days. Two people involved per feature. Cost per feature is around nineteen thousand dollars.
His traditional team of forty-six engineers shipped four features. Average cycle time is seven point eight weeks. Eight to ten people involved per feature. Cost per feature is around three hundred forty thousand dollars.
Same feature complexity. Same production standards.
But here is what he did not expect. After the agent-native engineer shipped their seventh feature, one of his senior engineers came to his office.
How is this person shipping so fast?
They are using agents to generate complete specifications that include everything. Requirements, edge cases, security, tests, and deployment. Then the agent implements all of it at once. No handoffs. No waiting.
Can I try that?
Yes, but you will need to work differently. Instead of waiting for product to write user stories you will work with product to generate complete specifications with the agent. Two to four hours of specification work instead of two to three weeks of story refinement.
What about code review?
The agent generates tests with the code. If the tests pass, ship it. Code review is for the specification before implementation, not the code after.
What about Quality Assurance?
That is what the tests are for. If you need Quality Assurance your specification was not complete enough.
Two weeks later that senior engineer shipped their first feature this way. Three days, one other person involved. They came back and said this is how we should be working.
Now he has three engineers working this way. Four more want to try.
Here is what he is learning. The compensation committee keeps asking for market data for this role. There is not any yet. The role is emerging right now.
It is like twenty oh eight when Development Operations Engineer did not exist in Human Resources databases. Systems Administrators were benchmarked at eighty-five thousand dollars. Early Development Operations engineers were getting one hundred forty thousand dollars to one hundred eighty thousand dollars. Human Resources said that is sixty-five percent above market. Companies that paid it deployed one hundred times more frequently and won their markets. Companies that waited for Human Resources benchmarks to catch up were eighteen months behind.
He is not waiting eighteen months.
The agent-native engineer is not ten times more productive at coding. They are productive because they eliminated the waiting. His traditional engineers spend two to three weeks waiting for product to write requirements, two to three days waiting for code review, one to two weeks waiting for Quality Assurance, and three to five days waiting for deployment windows. That is forty to fifty percent of their time waiting for other people.
The agent-native engineer generates complete specifications with product in four hours using Claude. The agent writes code and tests simultaneously. They deploy continuously. Zero waiting time. That is the difference.
The compensation committee sees three hundred twenty-five thousand dollars versus one hundred thousand dollars and thinks it is two hundred twenty-five percent more expensive. The actual cost comparison is different.
A traditional engineer gets one hundred thousand dollars salary plus their share of Product Managers, design, Quality Assurance, security, Development Operations, and management which is about one hundred thirty thousand dollars. Total system cost is two hundred thirty thousand dollars per engineer. They ship about four features per year. Cost per feature is fifty-seven thousand five hundred dollars.
An agent-native engineer gets three hundred twenty-five thousand dollars salary plus minimal support costs of about twenty thousand dollars. Total system cost is three hundred forty-five thousand dollars per engineer. They ship about sixty features per year. Cost per feature is five thousand seven hundred fifty dollars.
The agent-native engineer costs fifty percent more per person but ninety percent less per feature. That is what the compensation committee cannot see.
He cannot hire twenty of these engineers through the normal process. The compensation committee will never approve it. But he can transition existing engineers to this model. Four of his current team are already working this way. They are shipping faster, they are less frustrated, and they are proving this is not about special people. It is about working in a different system.
He is not asking to hire twenty engineers at three hundred twenty-five thousand dollars. He is asking to let his existing team work without the eight-person coordination chain. That costs nothing.
Next week, he is presenting to the board. One slide.
We currently spend seven point seven million dollars annually to ship twenty-four features per year. Cost per feature is three hundred twenty thousand dollars. Time to market is seven to eight weeks. For four point two five million dollars annually we can ship two hundred forty features per year. Cost per feature is seventeen thousand seven hundred dollars. Time to market is two to five days. The constraint is not capital. It is our delivery process and the compensation framework that prevents us from hiring people who can work outside it.
His request. Authority to transition twelve more engineers to the agent-native model and hire three more at market rate, which is two hundred fifty thousand dollars to three hundred fifty thousand dollars base plus a twenty-five percent bonus, without compensation committee approval.
If they say yes, he is scaling this. If they say no, he is updating his résumé. Because he cannot keep explaining to customers why they missed another deadline while startups with fifteen people are shipping their roadmap faster than they can.
Right. Here is what this means for you. If you are trying to hire engineers who can ship faster using AI agents, here is what matters.
First. Stop calling them AI engineers. Human Resources hears that and thinks senior engineer who uses a tool. They will benchmark it at one hundred forty-five thousand dollars. That is not the role. Call them engineers who eliminate delivery dependencies or engineers who ship complete features with minimal coordination. Make Human Resources see the system cost difference, not the tool difference.
Second. Calculate the total delivery system cost. Do not compare salaries. Compare cost per feature shipped. Include Product Managers, designers, Quality Assurance, security, Development Operations, and engineering managers in your calculations. Show what it actually costs to ship one feature through your current system. Then show what it costs with someone who eliminates six of those eight dependencies.
Third. Run a ninety-day experiment, not a hiring initiative. Do not ask to hire ten agent-native engineers at three hundred thousand dollars each. Ask to hire one engineer at three hundred thousand dollars for a ninety-day experiment to prove we can ship features at twenty thousand dollars instead of three hundred twenty thousand dollars. Make it an experiment with clear metrics and a personal commitment to track results. That is approvable.
Fourth. Work with your Chief Financial Officer, not your compensation committee. Compensation committees optimize for internal equity. Chief Financial Officers optimize for business outcomes. Show your Chief Financial Officer that we are spending more to ship less, and here is a way to spend less and ship more. That is a capital allocation conversation, not a Human Resources policy conversation.
Fifth. Build the proof with one engineer, then scale. Alex is not asking to transform his entire organization on day one. He hired one engineer, proved it works, and now four more are doing it. In ninety days he will have ten to fifteen working this way. In six months it will be thirty. He is not asking for permission to scale anymore. He is showing the results and scaling into them.
Finally. Prepare to bypass the process. The traditional hiring process will not work for this. Job descriptions optimized for keyword matching will not find these people. Compensation bands built for twenty twenty-three roles will not approve twenty twenty-five compensation. You will need to identify specific people, call them directly, and get executive approval to bypass the normal process. That is uncomfortable, but it is faster than waiting for Human Resources systems to catch up to a role transformation that is happening right now.
Look. Here is the real problem. Alex does not blame Human Resources. They are doing their job. Maintain internal equity. Control costs. Use benchmark data to make defensible decisions.
The problem is their job was designed for a world where roles changed slowly, productivity scaled linearly with headcount, and market data reflected current roles.
That world does not exist anymore. Roles are transforming faster than salary surveys can capture. Productivity is non-linear when one person can do what used to take eight. Market data is twelve to eighteen months behind role transformations.
The frameworks of Human Resources worked perfectly for twenty years. They do not work now. And Alex cannot wait for them to catch up.
So, where is he now? He is sixty days into his ninety-day experiment. The agent-native engineer has shipped fourteen features. His traditional team of forty-six has shipped four. Four of his existing engineers are now working the same way. They are shipping two to three features per week instead of one feature every eight weeks.
The compound effect is starting to show. They will ship more features this quarter than they did in all of last year.
But he is still frustrated. Because he knows exactly who he wants to hire. He knows what they would deliver. He knows the return on investment. And he still cannot get a compensation committee to approve it without months of benchmark analysis and internal equity studies.
Meanwhile startups are hiring these engineers at three hundred thousand dollars to four hundred thousand dollars and shipping products that make his look slow.
So he is bypassing the process entirely. Going directly to his Chief Executive Officer and the board with data. We can spend less and ship more if we hire people who eliminate coordination overhead.
If they approve it, he will scale fast. If they do not, he will know this company is not capable of adapting to how software gets built now.
Either way, he will know in two weeks. And he will not spend another three months in compensation committee meetings debating whether three hundred twelve thousand dollars is above market while competitors pay four hundred thousand dollars and capture his market.
The constraint is not capital. The constraint is not talent availability. The constraint is that Human Resources frameworks built for twenty oh five cannot evaluate roles that emerged in twenty twenty-three. And he cannot let that constraint determine whether his company is competitive in twenty twenty-five.
So he is going around it. You should too.