Anthropic restricts Claude Code third-party API access

Anthropic sent notices to Claude Pro and Max subscribers telling them third-party harnesses like OpenClaw will no longer be allowed to use their subscription credits. The justification is that these tools put 'outsized strain' on systems, which immediately prompted the obvious objection: if we're paying for token consumption and consuming tokens, what exactly is the problem? Commenters called it a money grab and noted it hits startups disproportionately, since they tend to build lightweight wrappers rather than paying for direct API access at scale.

The pattern here is familiar. A platform grows a developer ecosystem, that ecosystem starts consuming more than expected, and the platform draws a line. This happened with Twitter's API, Reddit's API, and now Anthropic is doing a version of it with inference subscriptions. The difference is that AI inference is genuinely expensive in a way that Twitter API calls were not, so the economic argument has more basis.

Commenters note that 'agent-agnostic' tooling is coming fast, meaning the market will route around this. But in the short term, anyone who built a product on top of Claude's subscription tier rather than the direct API is getting squeezed.