On Friday evening, Anthropic announced a sudden policy change banning the use of third-party harnesses, specifically naming OpenClaw, with Claude subscriptions. Users were given less than 24 hours’ notice before enforcement began on April 4th at noon. This move means that OpenClaw users will no longer have their usage count against their Claude subscription limits and will need to pay extra for continued use. Anthropic justified this by citing capacity constraints and the unsustainable usage patterns driven by third-party tools, offering refunds for those dissatisfied with the change. The company is clearly facing a GPU crunch and is trying to manage demand through both incentives, like doubling usage outside peak hours, and restrictions, such as limiting session times during peak periods.
Despite the abruptness of the ban, switching from Claude models to alternatives like GPT 5.4 within OpenClaw is straightforward and involves minimal effort, especially for users who have optimized prompt files tailored to different models. The video creator demonstrated how quickly they switched all their Claude models to GPT 5.4, highlighting that the transition is seamless and supported by the OpenClaw community. This flexibility is crucial given the uncertainty around Anthropic’s policies and the ongoing GPU capacity issues. The creator also recommended Digital Ocean’s Agentic inference cloud as a reliable alternative infrastructure for running AI models, emphasizing its simplicity and cost-effectiveness compared to hyperscalers or bare-metal GPU setups.
Anthropic’s communication around these changes has been confusing and inconsistent. Questions remain about whether the agents SDK, which some users rely on to connect OpenClaw with Claude, is still permitted. Anthropic representatives have stated no changes to the SDK at this time but have not provided clear guidance, leaving users uncertain about compliance risks. Meanwhile, users report sudden quota reductions and usage spikes that deplete their subscriptions quickly, exacerbating frustration. The company’s status page shows frequent downtime and performance issues, reflecting the strain on their infrastructure amid rapid growth.
Anthropic’s revenue is skyrocketing, reportedly reaching a $30 billion run rate and surpassing OpenAI publicly. They have secured major deals, including one with Google for TPU usage, fueling their expansion but also intensifying resource constraints. This growth underscores the importance of adopting a multimodel strategy for developers building AI agents. The video stresses the value of combining frontier models like Claude or GPT with local open-source models for tasks such as classification and summarization, which can help manage costs and reduce reliance on any single provider. This approach is increasingly necessary given the volatility in access and pricing from major AI platforms.
Finally, the video contrasts Anthropic’s restrictive and unclear policies with OpenAI’s more liberal quota management and ongoing improvements to GPT 5.4’s personality within OpenClaw. The creator notes that OpenAI appears more supportive of third-party tools like OpenClaw, especially following the involvement of Peter Steinberger, OpenClaw’s creator, who recently joined OpenAI. While Anthropic focuses on maximizing revenue and aggressively managing capacity, OpenAI seems to foster a more open ecosystem. The video concludes by encouraging viewers to adopt a multimodel approach and stay adaptable amid the rapidly evolving AI landscape.
