I read postmortems for fun. Occupational hazard. After enough years of SRE you develop a taste for them, the way some people develop a taste for natural wine, and most of what’s on the shelf is vinegar.
So when Anthropic published their April 23rd writeup on why Claude Code got dumber for five weeks, I went in braced for the usual vintage. “We’re always working to improve your experience.” “A small number of users may have noticed.” The corporate non-apology, aged in oak, tasting of nothing.
That’s not what I got.
Three root causes, named. A real timeline. The actual mechanism of each bug, written like an engineer wrote it and not like legal redlined it. No “some changes were for your benefit.” No passive voice doing the heavy lifting of hiding who turned the knob. Just: here’s what we changed, here’s what broke, here’s how we found it, here’s what we’re doing differently. I saved it to the folder. Of course I have a folder.
Here’s the part that stuck. That postmortem was, hand on heart, the most impressive thing Anthropic shipped that week. Not a model. Not a feature. A document explaining how they’d let me down. And it was better built than most launches I see — clearer than the average release note, more honest than the average roadmap, more useful than the average changelog that says “bug fixes and improvements.”
Which is a strange thing to be impressed by, said out loud. Against that field, plain honesty looks like fireworks.
But I’ll take it. I’d rather have a vendor that ships three bugs and explains them like an adult than one that ships zero public bugs and tells me nothing, because the second one is lying and I just can’t see the timeline yet. A good postmortem is a tell. It means somebody in there still thinks I’m owed an explanation.
Low bar. They cleared it. Credit where it’s due.
The product was the apology. Make of that what you will.
— linuxlsr