Story

I've seen a few anecdotes lately that say that they use Claude Code on legacy codebase and with relatively little supervision it can work on complex problems. Then the claim that Claude Code writes most of its own code, and that they no longer mentor their newcomers - instead, AI answers their questions and they can start making meaningful changes within the first few days. To me it sounds almost too good to be true, so I'd love to have some reality check.

I've spent most of my career in legacy codebases, reading, tracing behavior, making careful changes, and writing tests to protect them. I've taken a sabbatical though, which ends soon, and I'm quite worried and excited to what has happened during this time.

For those working on legacy codebases:

- Has the workflow really shifted to prompting AI, reviewing output, and maintaining .md instructions?

- Does your company allow Claude Code, Codex or similar tools? If not, what do you use?

- Do companies worry about costs and code privacy?

- Where do you think this is headed, a year from now?

Concrete examples, good or bad, would be especially helpful. Thanks.