Anthropic has restarted its projects using Claude code and parallel agents.

Anthropic redesigned Projects in Claude Code. It’s using a coordinator to assign tasks to separate threads, which then make their own pull request and run tests in a cloud session. This is currently in beta to a few Pro & Max users; sign up here.

Without effective coordination, parallel agents are likely to waste tokens without much gain in quality. In this case, Anthropic is doing the coordination itself, rather than leaving that up to you. I can’t wait to try this in beta.