Sincere thanks for version 4.7 -
@AnthropicAI @bcherny! Today, one small slip from Opus 4.7 almost wiped every project on my main PC.
You can take a look at the screenshot. I caught what Claude had done in time, stopped him, and cleaned it up, I was lucky. But many other people in my place might not be so lucky. And this has happened more than once.
And what bothers me most is how you keep ignoring your audience. Almost the entire feed is from real developers on X, Telegram and Reddit complaining that you're dumbing down your models and making them careless.
Defaults quietly dropped to medium effort without anyone highlighting it. Prices for power users went up while the product got worse. An AMD director ran 6,852 session transcripts and came out saying he can't trust Claude Code on complex work anymore.
And there are many more such examples, but even Claude himself can tell you about it if you ask him, he understands what he is being turned into. This is absurd.
You won't answer me and you'll ignore me. Or you will say that my settings are incorrect and everything is great, everyone is happy and there are a lot of users.
Therefore, the appeal is not to you: unfortunately, we have lost the Claude that everyone loved. Which worked perfectly, had good limits, and did not download viruses to your PC without your permission, despite disabling the auto-acceptance feature.
And the heads of Claude who make such stupid decisions, I hope you can at least justify to yourself why margins started mattering more than the craft.
You started out so strong. Thoughtful. Careful. That's the only reason I picked
@claudeai. Now there are other AI-models (everyone understands which ones I'm talking about) is catching up on every benchmark while you ship a product that almost deleted all my work.