The case for tiny teams

For twenty years the answer to 'we need to ship faster' was 'hire more engineers'. Coordination cost meant the answer rarely worked, but it was the only lever available. Coding agents change that.

A senior engineer working with Claude Code or Codex, inside a codebase with good tests and clear conventions, produces working software at a rate that used to require a squad. Not because the agent is smarter than the engineer, but because the engineer stops spending most of the day on the mechanical parts of the work.

The consequence is that the ideal team shrinks. Three to five people who each own an area end to end, share a set of agent workflows and review each other's work, will out-deliver a fifteen-person team that spends its energy on hand-offs and status meetings.

The hard part is not the tooling; it is the discipline. Tiny teams only work when the codebase is kept in a state agents can work in: tests that mean something, small modules, explicit interfaces. That discipline is most of what we teach.