[drankinatty]

I've got no love for AI, don't use it, but also after writing code for more than 40 years, keeping things in perspective helps. Whether it's you pecking away or some coding assistant helping, there will always be the potential for a regression or two to fly under the radar. (not like I've never done that before... nope.)

The issue the coding tools like Claude present is the sheer size and scope of changes and commits they generate that would take mere mortals months of careful coding to do.

That's an issue everyone using those tools will have to confront. I don't know Andrew personally, from a "let's go have a beer" standpoint, but I've known him from the samba list and his work with rsync for a very long time.

My take on the issue is less about the regressions and Claude screw-ups and more the lesson to all about the reliability of the coding tools and the diligence required to validate what they spit out.

It's an unfortunate black-eye, no doubt, but it's not a unique one. The takeaway is if something like this can slip by somebody like Andrew, then we all need to redouble the validation effort, lest we too are destined to share an unfortunate black-eye or two.

Never forget, "to err is human, but to really foul things up requires a computer."

AI just applies that adage at industrial-scale.