3 ms·
Added to my reading list, thanks. I was not really thinking in terms of reusing all knowledge at once. It is useful to be able to see what was the reasoning b
by rufasterisco 2mo ago
Added to my reading list, thanks.
I was not really thinking in terms of reusing all knowledge at once.
It is useful to be able to see what was the reasoning behind a single PR to discriminate for effort. The difference between copy+paste+yes+yes+yes+open-pr and asking n relevant questions about the code is what tends to matter.
Also, suppose after a few months you need to look at that code again. Having the possibility to go back and look at what was the human/agent conversation about it helps clarifying which tradeoffs where being considered.
Sticking to PRs (and meat proxy), a discussion over which reasoning led to that code should help a team pick and choose reliable collaborative patterns.
Overall, I don’t think this is a panacea, simply that not storing it feels wrong, to me.
- ltbarcly3 2mo agoI think it might be more useful to store a summary of the key decisions, but with mistakes, AI fumbles, iterating on test failures, and all the noise boiled out. If you kept, maybe as part of the commit log a summary of the AI discussion leading to that commit I could set that as very valuable.
- rufasterisco 2mo agoI went there, and walked back to full sessions, mostly to be able to resume chats. As a general principle, why would i want to lose precision? A senior might care about seeing a junior conversation to provide advice. A colleague might spot another one making as different assumptions about X, and have a chat about it. AI grepped file XX instead of XY and made a wrong decision based on that. An engineering manager might care about completely different metrics. But to your point, think it's fair to say that there isn't a single correct answer to this: different people like different stuff and at eventually standards will emerge (or not).