Our founders spent the last two years inside engineering orgs adopting AI, one running a team and one recruiting for them. Two patterns showed up again and again, and the difference in how those teams ship is stark.
The first kind bought AI tools and changed nothing else. Same team shape, same process, same tickets, now with an assistant. They got a real but modest bump, maybe 15–20% on routine work, and a new failure mode: code nobody on the team fully understands.
The second kind restructured. They went smaller and more senior (five engineers where there used to be nine) and moved the freed budget into better people and better tooling. The senior engineers use AI aggressively for the repetitive 60% of the work and apply their own judgment to the 40% that actually matters. Those teams are shipping roughly twice what they did, with fewer production incidents, not more.
The structural insight is that AI collapses the cost of the junior tier's traditional work (boilerplate, tests, first drafts) but raises the value of the senior skills: knowing what to build, catching the subtle wrongness in generated code, designing systems that stay comprehensible. Teams that staff for that asymmetry win.
It changes hiring specs too. 'Comfortable with AI tools' is table stakes now; what we screen for is whether someone can review AI output critically. We hand candidates generated code with two planted flaws and watch what happens. It separates people faster than any puzzle question we have tried, and it is now a standard part of our engineering screen.
None of this requires a moonshot. The teams that made the shift did it one squad at a time, measured honestly, and expanded what worked. We help on both sides: finding the senior people the model depends on, and building the guardrails and evals that make AI output trustworthy enough to ship.
