TIL, via passing mention in a video about a different bit of AI business-model fuckery, that Claude Code users recently started being billed for extra “API usage” – in the reported case, about $200 extra – if the project they were working on had a file named HERMES.md.
> “I sincerely apologize for the disruption you experienced with the billing routing issue. We take service reliability very seriously. However, I need to let you know that we are unable to issue compensation for degraded service or technical errors that result in incorrect billing routing.”
Translation: “Yes, our system charged you $200 through the wrong billing path because of a string in your git history. No, we won’t give it back. But thanks for the QA work!”
I’m not an AI enjoyer and haven’t been forced to use it for work, so if this had been me, I’d have reached the “fuck this, I’m out” stage just looking at the $200 a month price on a Claude Max subscription. These extra couple layers of fuckery would have driven me somewhere dark and full of fire.
- First, if it was a bug, the right thing to do is to fucking correct it and undo the harm it caused, as much as possible. And for bugs that accidentally siphon something like money or monthly token allowance from a user’s account, undoing the harm is trivial: return the money or the tokens. Anthropic chose not to do this.
- Second, the name “HERMES.md” was weirdly specific enough to trigger my search-engine reflexes, and a minute later, I’d found a locally hosted AI agent named Hermes, as well as widespread speculation about how Anthropic would really really like its customers to be as locked into cloud services as possible and not be in the habit of using locally hosted AI.
Quoth an Anthropic dev:
That subscription structure is maybe three years old. Has it been disrupted already? And are we now suddenly acknowledging that “disruption” is only supposed to be an inherently good thing when techbros and slopbros make bank off of it?
Ray of Infamy 
Leave a comment