Anthropic squeezes third-party Claude Code users

A thread resurfaced around Anthropic's decision to stop allowing Claude Code subscriptions to work through OpenClaw, a third-party harness. Anthropic's stated reason was that these tools put 'an outsized strain on our systems,' which landed badly with users who pointed out they were paying for a fixed subscription and staying within stated limits. The response was a one-time credit equal to one month's subscription price and a push toward paying API rates instead.

This is the classic platform tension: a company offers a flat subscription, power users find ways to maximize it, and the company redraws the lines. The HN thread was pointed about this, with several people noting that the move unfairly hits startups who built workflows around the subscription model. One commenter raised the obvious alternative: let users buy API keys with usage caps, which would align incentives better.

The deeper issue is that agentic use cases consume tokens at a rate that makes flat subscriptions economically unsustainable for providers. This is not a Claude-specific problem. It's a structural tension that every AI subscription product will hit as agent usage grows.