Anthropic is deploying a new capability for Claude Code that allows one instance of the AI tool to coordinate multiple other Claude Code sessions working in parallel. Starting Thursday, the company announced that certain Claude Pro and Max subscribers can access the redesigned Projects feature in beta through cloud sessions, with broader rollout to those plan tiers scheduled for the coming week. The revamped system centers on a coordinator that receives an engineering goal, divides it into smaller tasks, and assigns them to several Claude Code sessions running simultaneously—removing the manual labor developers previously performed to partition work and consolidate results.
The coordinator handles request scoping, work delegation, parallel thread coordination, output review, and final assembly, according to an Anthropic representative. After a developer defines a goal, Claude determines how to distribute the work across threads, which the developer can monitor or open individually for intervention. Each thread operates as its own Claude Code cloud session with a dedicated branch and repository copy, and can deploy subagents, loops, and workflows to manage smaller components of the assignment. Running multiple threads simultaneously consumes tokens at a faster rate, with Anthropic noting that Projects will reach usage limits more quickly when several threads are active because each one counts as a complete Claude Code session. The company currently faces a class-action lawsuit filed last week by developers on its Max plan, who claim that promoted usage increases included weekly caps Anthropic initially failed to disclose.
The report describes a scenario where a team retiring an old API endpoint could link its API, web, and mobile repositories and have Claude establish a thread for each to update callers, execute tests, and open pull requests before determining which changes require merging first. Developers can track progress through the main Project chat or access individual threads for inspection or redirection without manually starting and managing each Claude Code session themselves. Each worker thread can further subdivide its assignment using Claude Code's existing subagents, loops, and workflows when necessary. The coordinator includes a review layer that may carry practical weight: on the Real-SWE benchmark, which evaluates coding agents against private codebases, even Anthropic's own Fable 5.1—the highest scorer—failed more than 60 percent of the time, suggesting that coordination and output review matter as much as raw model capability when agents work with unfamiliar production code.
Each thread works from its own branch, maintaining separation but not preventing two threads from modifying identical code. When that occurs, Anthropic handles it as a merge conflict, identical to any other pull request. The company is adding shared memory so information learned in one thread becomes available to others without developers repeatedly providing the same context—Projects can retain details such as a changed release date, the rationale for dropping a feature, or who must approve changes to a particular service, with that information persisting as work continues over days or weeks. Claude can also remember developer preferences for project management, including check-in frequency, when to start new threads, and update detail level, while a new library stores files added by the user alongside artifacts Claude produces so future work can draw on material already created within the project. Users can monitor Project-specific usage and select the model and effort level for both coordinator and worker threads, while support for running threads locally alongside a developer's tools and code—including resources behind a private network—is arriving soon. The beta will expand to additional Claude Code users on Pro and Max plans over the coming week, with mobile support arriving soon and Team and Enterprise access planned for later; existing subscribers who already use Projects will stay on the current version until Anthropic upgrades them. The redesign forms part of a broader product consolidation: Anthropic recently merged Claude chat and its Cowork interface into a single view, wagering that eliminating the choice of mode would reduce friction rather than increase it. The shift to parallel coordination represents a structural change in how developers might approach complex, multi-repository engineering work—though whether the token economics and error rates make this practical for production use at scale remains an open question that early adopters will need to answer through real-world deployment.