A case study in testing with 100+ Claude agents in parallel (imbue.com) AI

Imbue describes how it uses its mngr tool to test and improve its own demo workflow by turning a bash tutorial script into pytest end-to-end tests, then running more than 100 Claude agents in parallel to debug failures, expand coverage, and generate artifacts. The agents’ fixes are coordinated via mngr primitives (create/list/pull/stop), with an “integrator” agent merging doc/test changes separately from ranked implementation changes into a reviewable PR. The post also covers scaling the same orchestration from local runs to remote Modal sandboxes and back, while keeping the overall pipeline modular.