My Claude weekly limit lasted longer once I made Opus a project manager

Users of Anthropic's Claude Pro service face weekly usage limits and a rolling five-hour window, which can be quickly exhausted by intensive models such as Opus. To complete a substantial coding project, one user adopted a strategy where Opus, the most capable model, was explicitly tasked with only orchestrating the development, rather than performing the actual coding or testing.

This approach involved Opus spawning multiple sub-agents, specifically Sonnet and Haiku models, to handle the implementation and testing work. The user observed that Opus, when limited to supervisory functions, "sat back and waited," which reportedly did not contribute significantly to usage consumption. Most of the usage was attributed to the sub-agents performing the core development tasks.

The reported outcome was the successful completion of a large application, a replacement for Google Sheets, with usage capacity remaining before the weekly reset. This suggests that leveraging Opus for its planning and oversight capabilities, while offloading execution to models that consume fewer tokens, could be a practical method for users to maximize their available AI capacity on complex projects.