Home / Technology / Anthropic Reverses Agent Ban with New Credit System
Anthropic Reverses Agent Ban with New Credit System
14 May
Summary
- Claude subscriptions now include specific credits for external agents.
- New policy restricts agent usage to fixed, non-rollover monthly credits.
- Community backlash criticizes the reduction in subscription value.
Anthropic has reinstated the ability for Claude subscribers to power external AI agents, such as OpenClaw, through a new 'Agent SDK' credit system. This marks a reversal of a policy enacted in early April 2026, which had prohibited such usage due to capacity and service issues caused by unoptimized third-party agents consuming excessive tokens.
The new system allocates a fixed monthly credit for programmatic use, distinct from standard subscription limits. If these Agent SDK credits are unused, they expire at the end of the month. This approach shifts the cost of inefficiency back to the user, preventing strain on Anthropic's infrastructure and protecting its financial margins.
While Anthropic frames this as a simplification, the developer community has reacted negatively, viewing the metered credits as a significant reduction in value compared to previous usage. Concerns have been raised about the limited utility of the credits and the potential for reduced accessibility for developers relying on external tools for automation. The change effectively ends an era of 'compute arbitrage' where subscriptions offered substantial cost savings for agentic workflows.
