Anthropic has redesigned its Claude Code projects feature, now in beta, to let users orchestrate multiple AI agents running in parallel cloud sessions under a single project with shared memory, goals, and file libraries3,4.
The revamped architecture introduces "threads," each an independent Claude Code cloud session operating on its own branch and copy of a code repository. A "coordinator" layer directs the threads, assigning and organizing tasks across them. When threads modify overlapping code, the system resolves conflicts as standard merge conflicts, mirroring a typical pull-request workflow. Each thread can further subdivide its delegated work, enabling recursive task decomposition.
The workflow begins with a single conversation in which a user describes the work to be done; Claude then manages execution across the parallel threads. Sessions continue running in the cloud even after a user closes their laptop2.
The feature is available in Claude Code's cloud environment, but local developers will have to wait for support1. No timeline for local availability was provided in the accessible reporting.
ANALYSIS The design pattern, a coordinator dispatching work to branched cloud sessions that reconcile via merge conflicts, maps AI-agent orchestration onto version-control primitives already familiar to software teams. That choice lowers the conceptual overhead for developers adopting multi-agent workflows inside existing Git-based pipelines.
By keeping each thread on its own branch and treating overlaps as merge conflicts rather than attempting automatic resolution, Anthropic shifts final arbitration to the developer, preserving human control at the integration point.