Hacker News Reader

Comment by low_tech_love | Hacker News Reader

Isn’t this a kind of “leopards ate my face” situation? I thought we had all “agreed” that letting AI write code and take control of software repositories is good, even if we have no idea what is going on beyond a thin surface layer, because well it’s fast and we can fix it later and lol who needs testing? My customers are my testers.

And now it’s suddenly bad because the developer is the customer?

View on HN

The sneaky commit modification is triggered by very modest usage of AI such as auto-completion.

Look, if an agent writes the code and the commit message then adding a Co-authored-by by default is ok. Not even showing it before the commit is made is not, and adding the message when AI was just completing code is not.

I genuinely think it's not ok even then. Copilot is a tool, one of many I use. That tool has no business polluting commit messages without my knowledge.

The appended message isn't even adding any new information, as in this day and age a vast majority of commits is probably "co-authored" by an LLM.

I should have been clearer, the hidden addition is never ok.

If I ask Claude to write a commit message, it will inserted a co-author line (and an ad), but I can see it and disapprove, add a counter instruction to CLAUDE.md etc

I personally don’t understand the need to treat a tool as an “author” but that’s not important, my comment is mostly regarding the backlash of what happened. A feature was rushed in and does not work as intended, in a kind of disastrous way. Now we feel like our customers do when they have to deal with all the crap that our AI co-authors push forward without the right process.

Glorified autocomplete, syntax reminder and random snippet generator thinks it's co-authoring things.

Reply on news.ycombinator.com