1b:I[1659,["874","static/chunks/874-5853b36c417a5f43.js","598","static/chunks/598-8e8b350c69b61725.js","735","static/chunks/735-5d5145bb0f79dc16.js","405","static/chunks/app/%5Blocale%5D/item/%5Bid%5D/page-b0ef2b6c2502e4ad.js"],"default"] 1f:I[8170,["874","static/chunks/874-5853b36c417a5f43.js","598","static/chunks/598-8e8b350c69b61725.js","735","static/chunks/735-5d5145bb0f79dc16.js","405","static/chunks/app/%5Blocale%5D/item/%5Bid%5D/page-b0ef2b6c2502e4ad.js"],"default"] 1c:T664,> I've been thinking about this over the past few days - If AI is as capable as people here claim - enough to replace junior or intermediate developers - could it be used to maintain semi-abandoned open-source projects?

I don't think AI agents can generalize reliably to maintain complex open-source software, better than the authors. Those who say that are out to sell you an AI product or something.

The benchmark I would consider is the maintenance of the Linux kernel and AI is far from approaching to the level that it can maintain correct critical code that is trustworthy enough to being merged into mainline. But we'll see.

> Overall, not a very difficult problem and I can probably do it myself given a few days. Human engineering hours are quite expensive though and for such a low value task, I would assume an AI agent would be more cost effective.

Software maintenance is indeed expensive for a reason, however we still can't assume that an AI agent can help deal with autonomous maintenance on OSS projects; large or small or even in [0] and [2].

Technical debt is also a cost and many of these abandoned OSS projects don't even have any tests in the first place, including both 'hydroxide' and the protonmail client in [2].

So how do you know you if either project really works or not? Even if you use AI, what if it introduces a new bug in its implementation? How would you know beyond 'it works for me'?

In fact, autonomous AI agents just risk introducing more techn

翻译中...

1 comment