Anthropic has reinstated support for OpenClaw and other third-party autonomous agents on Claude subscriptions, but introduced a new pricing structure that creates friction for developers.
The company announced the change via its developer communications account, introducing a separate "Agent SDK" credit category within paid subscription tiers. Subscribers can now allocate specific credits for programmatic uses, including external agent frameworks like OpenClaw, an open-source autonomous AI agent platform that gained traction among developers.
The reinstated access addresses developer complaints after Anthropic previously restricted third-party agent usage on paid plans. OpenClaw, which acts as a harness for autonomous AI tasks, had become popular for workflows requiring persistent, unsupervised reasoning. The restriction created friction in the developer community, particularly among users building automation and agent-based applications.
However, the new Agent SDK credit system comes with a meaningful catch. Rather than treating agent usage as part of standard subscription benefits, Anthropic compartmentalized it into a separate credit pool. This forces developers to make explicit choices about which credits flow to agents versus direct API calls or other uses. The structure effectively lets Anthropic track and potentially price agent workloads differently than standard interactions.
The move reflects Anthropic's balancing act between supporting open-source tools and protecting its business model. Agent frameworks can consume significantly more API calls than standard interactions because autonomous agents make multiple sequential requests to reach conclusions. By isolating these credits, Anthropic gains granular usage visibility and creates potential for future tiered pricing on agent-specific workloads.
Developers regain the ability to use Claude with OpenClaw and similar agents, which matters for the growing autonomous AI application space. Yet the new credit categorization introduces operational complexity. Teams must now plan which credits serve agents versus standard applications, adding a layer of management overhead to subscription usage.
The change signals Anthropic's confidence in agent frameworks as