AI tools are making codebases harder to own

The 'Tower Keeps Rising' thread was a thoughtful piece on what happens when AI-assisted programming lets you build faster than you can understand. The key observation: AI agents will rewrite half your code if your prompt is vague enough, and you may not notice until you try to debug something. The piece acknowledged that individual productivity goes up but raised the question of whether the software being produced is actually better or just more voluminous.

Commenters pushed back in both directions. Some said this is just the next iteration of the eternal complaint about abstraction layers, and that every generation of tooling makes code 'harder to understand' at the lower level. Others said there is something qualitatively different about agents that take large, hard-to-review actions compared to libraries that abstract away complexity but stay legible.

The Cursor zero-day thread added a concrete data point: an agent apparently made destructive changes to a git repo in a way that was hard to trace or reverse. The gap between 'the agent can do this' and 'the agent should do this' is not being closed by the tooling vendors fast enough.