Development with Living Agents

I keep thinking about the future of IT companies and programmers, trying to find a sensible place for developers in the architecture of tomorrow. And I think I may have figured it out.

Programmers should verify all AI-generated code. For serious projects, where you cannot blindly hand development over to Codex or Claude but also cannot afford lengthy manual development, you could build a code-quality, architecture, and documentation team around experienced developers. And to make sure nobody slacks off, there should be a tool that feeds developers modules they must examine and document manually: figure out the code and explain what each method, function, and fragment actually does.