The argument that AI will replace engineers misunderstands what engineering work actually is.
AI tools accelerate the mechanical parts of software construction. Code generation, boilerplate, test scaffolding, documentation. These are real and meaningful gains. But the hard parts of engineering — understanding what to build, deciding how the system should behave, navigating uncertainty and trade-offs — those parts are not being automated. They are being made more consequential.
When the cost of producing code goes down, the cost of building the wrong thing does not. It goes up. You arrive at the wrong destination faster.
What this means for teams
A small team in the AI era has leverage that would have required three times the headcount five years ago. One strong engineer who understands the domain, reasons clearly about trade-offs, and can use these tools deliberately will outproduce an average team of five.
But leverage is asymmetric. A weak team with AI tools produces wrong outputs faster. The quality of the thinking that guides the tools matters more, not less.
The implications for team design are direct:
- Smaller, higher-quality teams over larger average-quality teams
- More time on problem definition before any code
- Deeper investment in the people who know what good looks like
- Much less tolerance for unclear ownership — because the velocity of execution means that unclear ownership compounds quickly into expensive rework
The compounding advantage
Strong teams were always better. But the gap was capped by execution speed. You could only ship so fast, build so many features, maintain so much surface area.
That cap is lifting.
A team that reasons clearly and uses AI tools well will compound on its own quality in a way that was structurally unavailable before. The leverage is real. So is the risk on the other side.
The question for any engineering organization right now is not whether to use AI tools. It is whether the team quality, the decision-making culture, and the clarity of scope are good enough to make the leverage net positive.
For most teams, that is a harder problem than the tooling itself.