We track AI-generated code across the engineering team at Zendrop. The numbers range from 65% to 95% depending on the developer and the work. That spread is the interesting part.
The 95% developer works on well-scoped tasks with clear patterns — API endpoints, CRUD, test coverage, migrations. Their job is orchestration: break it down, prompt, review, ship. The 65% developer works on cross-service data flows, performance-critical paths, integrations with undocumented behavior, security-sensitive code. AI helps with scaffolding, but the core logic is human. Both are senior. The difference is what kind of problem they're solving.
The mistake is treating 95% as the goal and 65% as a problem. If you push for 95% everywhere, you're implicitly saying the hard problems don't exist. The right question isn't "how do we get everyone to 95%?" — it's "are our 65% developers working on the right problems?" The spread is the signal. Pay attention to it.