Explain "why not what" is good general advice. My further advice for comments is: even bad comments can be useful (unless they're from LLM output maybe...) therefore when in doubt, write a comment. Write it in your own words.

Had to add the last sentence for the circa 2020s developer experience. LLM comments are almost never useful since they're supposed to convey meaningful information to another human coder, anything your human brain can think of will probably be more helpful context.

▲

falcor84 4 days ago | parent | next [-]

I strongly disagree. If you're using something like Claude Code to generate the code, it has significant context about the task, which from my experience provides very useful (albeit overly verbose) comments. I sometimes edit/rewrite its comments (as I might with the code itself), but I would never ask it to generate uncommented code.

▲

3371 4 days ago | parent | prev | next [-]

I always think LLM comments are more about helping themselves to stay on track.

	▲	yen223 4 days ago \| parent [-]
		Same goes for human comments tbf

▲

CuriouslyC 4 days ago | parent | prev | next [-]

AI comments are fine for high level summaries of the what/how. They fail at the why, which is where we come in.

▲

bonesss 3 days ago | parent | prev [-]

> even bad comments can be useful

Bad comments aren’t just less helpful than possible, they’re often harmful.

Once you’ve hit a couple misleading comments in a code base (ie not updated, flatly wrong, or deeply confusing due to term misuse), the calculus swings to actively ignoring those lying lies and reading the code directly. And the kind of mind resorting to prose when struggling to clarify programming frequently struggles with both maintenance and clarity of prose.

	▲	mattacular 3 days ago \| parent [-]
		I hear this argument as an excuse not to write comments (sometimes at all). Maybe I am just lucky but I have never had this issue as you've described in codebases, and if I did, certainly not to that extent where it became a memorable thorn in my side. If there are no comments, you are reading the code (or some relatively far away document) for all understanding anyway. If there are inaccurate comments, worst case you're in the same boat except maybe proceeding with a bit more caution next comment you come across. I always ask of fellow engineers: why is it unduly difficult to also fix/change the comments as you alter the code they refer to? How and when to use comments is a balancing of trade-offs: potential maintenance burden in the future if next writers are lazy vs. opportunity to share critical context nearest its subject in a place where the reader is most likely to encounter it.