The pressure is real. The math isn’t.
Executives are hearing the same message from boards, investors, and peers: adopt AI fast and show the savings. That pressure rolls downhill. Managers are asked to forecast productivity gains, and in many organizations the forecast arrives as a headcount number before anyone has measured what AI actually changed.
The logic sounds simple. If AI makes each person 30% faster, you need 30% fewer people. But that assumes work moves through an organization the way it moves through one person’s keyboard, and it doesn’t.
Organizations acting on that assumption are making permanent decisions based on a model of productivity they haven’t tested. And they’re often cutting the very people who make AI worth the investment.
Speed in one place becomes a queue in another
A recent article from IT Revolution, AI-Driven Organizational Value, puts data behind what many engineering leaders have seen firsthand. Drawing on Faros AI research spanning 22,000 developers, the authors describe a bank team that accelerated its pull request pipeline with AI. Review time rose roughly 441%. Incidents per pull request rose roughly 243%.
The team didn’t get slower. Its speed showed up downstream as a queue. Clear code review, and the constraint moves to validation. Clear validation, and it moves to operations, then to the business’s ability to decide what’s worth building next.
The paper’s core point is that freed-up capacity is not the same as captured value. That’s why AI adoption so often fails to show up in the P&L. Most organizations measure hours saved in one function, not outcomes delivered by the whole system.
Where AI actually creates value: fewer handoffs, more learning
In our experience, the real gain from AI looks different from the headcount story. AI is most powerful in the hands of strong, opinionated experts. An experienced engineer or product lead who knows what good looks like can now cover more breadth: shaping a design, scaffolding the code, writing tests, and checking the result, without handing each step to someone else.
That produces two benefits of very different size. Fewer handoffs save some time, since every handoff carries wait time and lost context, but the savings are smaller than most forecasts assume.
The bigger gain comes from iterating more. When one expert can move from idea to working prototype quickly, the team can test more ideas and learn faster in the same window.
AI speeds up someone who can judge the output. Without that judgment, it speeds up rework. Cutting experts to fund AI removes the people who make it valuable.
The same finish line, a better-informed outcome
The productivity pitch usually skips this part: AI often doesn’t move the end date much. What changes is what you know when you get there.
Picture two teams with the same 12-week window. The first builds one version of the solution and validates it near the end. The second, working with AI, builds and tests three or four prototypes, drops the weak ideas early, and puts the strongest one in front of users by week six. Both teams finish in week 12. Only one had the chance to learn its way to the answer.
That’s the return leaders should be looking for: fewer wrong bets and decisions grounded in evidence. In the P&L, it shows up as better products and less rework. It rarely shows up as a smaller org chart.
Where AI does reduce headcount
AI can reduce headcount, but mostly in large teams. In a small, focused team, each expert holds knowledge the work depends on. AI can widen what each of them covers, but it can’t replace that expertise. Cut one person, and their judgment leaves with them.
Large teams are different. When many people have overlapping skills, AI can absorb some of that overlap, and a smaller team can deliver the same work. That productivity gain is real.
If the goal is to move faster and lower costs, reducing headcount can do both. Amazon’s well-known “two-pizza teams,” small enough to feed with two pizzas, were designed to keep communication overhead low without giving up the skills a team needs to own its work. Teams adopting AI need a similar balance: reducing overlap, improving end-to-end automation, and optimizing the whole value stream rather than a single step in it. Less overlap also means more risk when someone leaves, so a deliberate, measured approach matters for long-term success.
A leaner team also has to work differently, and the product owner role is a good example. On a team running two-week sprints, a product owner who reviews work at the sprint demo used to be enough. Once the team can build and test a prototype in a couple of days, that same rhythm leaves most of those prototypes waiting on a decision, and the product owner becomes the bottleneck. They need to give feedback far more often and stay closely connected to the team’s day-to-day work. Without that, the team just builds faster in the wrong direction.
The fundamentals haven’t changed
From a development perspective, AI doesn’t overturn what we’ve learned over the past two decades. Better outcomes come from tighter iterations. Tighter iterations depend on fast feedback loops. And fast feedback loops only exist where strong agile practices and DevOps principles are already in place.
AI raises the stakes on all of it. If a team can now produce five prototypes in the time it used to produce one, every downstream step has to keep pace: automated testing, continuous integration and delivery, observability, and a product owner ready to make decisions quickly. Some teams are moving toward a hyper-agile rhythm, with iteration cycles measured in days rather than sprints.
Where those foundations are weak, AI exposes them. The speed piles up in review queues, test environments, and approval steps, exactly as the IT Revolution research found.
Five questions to answer before you cut
If you’re under pressure to show AI returns, work through these before making headcount decisions.
- Where is your real bottleneck? Map the full path from idea to customer value. If AI speeds up one stage, find where the work will queue next.
- Are you measuring learning or activity? Track iterations, experiments, and validated decisions, not just hours saved or tools adopted. The IT Revolution paper offers a useful filter: how many AI-built tools are used regularly by someone other than the person who built them?
- Who are your opinionated experts? Identify the people whose judgment makes AI output trustworthy. Those are the people to equip, not the people to cut.
- Can your delivery system absorb more iterations? Check your testing, deployment, and feedback practices before you add speed upstream.
- Are you making cost decisions on purpose? AI usage carries real, ongoing cost. Decide where it earns its keep rather than discovering the bill at quarter end.
Navigate your AI journey
Find where you are on the AI journey
Every organization is at a different point. AIM’s AI Navigation Map lays out the common challenges at each stage, how we help, and examples of the outcomes. Start with the challenge that sounds most like yours.
Explore the AI Navigation MapInsights By
