Anthropic restricting third-party harness access to Claude Code
Anthropic sent notices to subscribers telling them third-party harnesses like OpenClaw can no longer be used with Claude Code subscriptions. The official framing is that these tools put 'outsized strain' on systems. The HN response is immediate skepticism: if you are paying for token consumption and the tokens are being consumed, what exactly is the problem?
Several commenters point out the business logic: Anthropic wants to own the agentic workflow layer, not just supply the model underneath someone else's interface. By blocking third-party harnesses, they push users toward first-party tooling, which gives Anthropic more control over the experience and more visibility into usage patterns. One commenter predicts a wave of 'agent-agnostic' tools coming soon that can run across multiple providers.
The startup impact is direct. Multiple small companies built their product on top of Claude Code plus a custom harness, treating Anthropic as infrastructure. That is now a riskier bet. The broader signal: AI providers are moving up the stack, which means startups building on top of those providers are having their floor raised beneath them.
So what?
If your product sits between an AI provider and an end user, you are in a squeeze. Anthropic, OpenAI, and Google all have incentives to own that layer themselves. Build on top of AI APIs with the assumption that any workflow you create that drives significant usage will eventually be productized by the underlying provider. Either diversify across providers now or build the kind of customer relationship that survives a model switch.