Anthropic Tightening Claude Code Access for Power Users
Anthropic sent notices to Claude Code subscribers using third-party harnesses like OpenClaw, effectively banning the practice and citing 'outsized strain on systems.' The thread is frustrated: users paid for a subscription, the pricing model is token-based, and the strain argument does not hold up if consumption stays within paid limits. One commenter put it plainly: it is like buying a gym membership and being told you are working out too much.
The pattern this connects to is the broader problem of AI companies struggling to price usage correctly. Flat subscriptions break down when a small fraction of users are running heavy agent loops. But the response of restricting access rather than introducing tiered pricing or per-seat API keys is alienating exactly the power users who evangelize these tools.
Commenters predict a wave of 'agent-agnostic' tools that work across multiple providers, which would commoditize any single provider's moat and push the market toward whoever offers the best price-to-capability ratio.
So what?
If your product depends on a fixed Claude or GPT subscription for heavy workloads, you are exposed to unilateral access changes with minimal notice. Build multi-provider flexibility into your architecture now. The providers themselves are signaling that the flat subscription model does not work for high-volume use cases, which means pricing models are about to get more complex and more expensive for serious builders.