Post #2097456

2025-09-13 17:22 UTC

Do any of the AI code assistants help significantly with *reading* code, instead of writing it? Not sure why I care, since I have no interest in either AI or coding at this point, but it seems like a bigger net win than letting it write code that then needs *deep* scrutiny before it can be deemed safe. Just curious I guess.

Replies (2)

@samstokes@hachyderm.io 2025-09-13 19:06

@Obdurodon@hachyderm.io in my experience, yes. I've used Claude Code and Codex CLI to ask questions in large complex codebases (e.g. "where does this field get populated" when the answer requires cross-referencing across several layers or even services) and had a high success rate of the answer being both useful and correct (which of course I independently verify, don't *trust* the nondeterministic oracle). Material time saver vs doing the spelunking myself, or messaging the people who might remember the answer. Interestingly these are the same large codebases where the same tools struggle to *write* code (vs toy/greenfield where they do well).

Open ##2226527
@samstokes@hachyderm.io 2025-09-13 19:08

@Obdurodon@hachyderm.io I've also used them to orient myself in a new codebase ("how is the code structured", "how does this system work", "how do I run the tests").

Open ##2226528