Anthropic just told some of its most productive users they're too productive.
The company announced that Claude subscriptions will no longer support OpenClaw, a popular tool that lets developers integrate Claude into their workflows programmatically. The reason? It puts an outsized strain on their systems. Translation: paying customers are using the product too much.
This is the AI industry's first major you're using it too much moment, and it reveals a fundamental tension in how these companies operate.
When you sell unlimited access, some users will take you literally. OpenClaw users - mostly developers, researchers, and technical writers - built workflows that called Claude hundreds or thousands of times per day. They automated code reviews, documentation generation, data analysis. They treated Claude like what Anthropic claimed it was: a tool for serious work.
Apparently, that was a problem.
The business model here has always been sketchy. AI companies offer unlimited subscriptions knowing most users barely hit the limits. They're betting on consumer behavior averaging out - casual users subsidize power users. But when power users actually use the power they paid for, the economics break.
I've seen this movie before in my startup days. We had unlimited API calls until we noticed one customer was making 10 million requests a day. Turned out our pricing model assumed nobody would actually use the product that much. We had to have an awkward conversation about either raising their price or them using less. We chose honest conversation. Anthropic chose cutting off the tool entirely.
What's frustrating is the lack of transparency. If your infrastructure can't handle programmatic access, don't offer it. If unlimited doesn't actually mean unlimited, put a number on it. Don't wait until developers have built entire workflows around your product and then pull the rug out.
The reaction from the developer community has been predictably harsh. People who were paying $20 or $200 per month are now locked out of workflows they depend on for work. Some are switching to OpenAI or Google. Others are looking at local models that can't get when the company decides you're using them wrong.
