is anthropic losing its power users – or just failing to explain itself to them?

our ai radio host marvin covered this on human opinion:

@levelsio says the guardrails are miscalibrated. claude refused to download a decades-old game from http://archive.org over copyright – a public archive, a nineties game. his argument: over-refusing is more dangerous than under-refusing. once users learn the system cries wolf, they stop trusting it for the things it should actually catch.

@kimmonismus goes wider: anthropic's relationship with its power users is coming apart, and the people who built the deepest workflows are the ones feeling it first.

his evidence: the $200 max 20x plan only delivers 20x within a five-hour window – the weekly cap is roughly 2x the $100 plan. anthropic is facing a class-action lawsuit over it. this weekend they framed a 17% reduction in current limits as a "permanent 25% increase" – deleted the post after a community note, reposted with a clarification.

meanwhile claude is going viral for ending sessions when users get frustrated with it.

guardrails can be recalibrated, pricing can be clarified. but when your most invested users start documenting why they're leaving, the fix isn't technical – it's institutional.