Anthropic relaunched the Projects feature within Claude Code on September 17, 2026, transitioning the tool from a standard AI assistant into an orchestration hub capable of managing multiple autonomous agents. The updated system introduces a central “Coordinator” agent designed to scope complex developer requests and delegate them to parallel worker threads that can operate simultaneously without manual oversight.
This relaunch marks a shift toward “always-on” development environments. Unlike previous iterations that focused on single-thread conversations, the new Claude Code Projects architecture allows the Coordinator to spin up multiple cloud sessions. Each worker thread operates as a full Claude Code instance on its own git branch and repository copy, enabling the AI to tackle different parts of a codebase at the same time. The agents maintain synchronization through a unified memory, shared project goals, and a common library of artifacts.
The Coordinator and Worker Logic
The core of the system is the delegation logic. When a developer submits a broad prompt—such as refactoring a module while simultaneously writing unit tests—the Coordinator analyzes the scope and spawns dedicated threads for each task. Because these threads run on separate git branches, overlapping edits to the same files are managed through standard git merge conflicts, which the agents are designed to resolve.
The orchestration is powered by Anthropic’s Claude Fable 5.1 and Mythos 5.1 models, which were released on September 1, 2026. These models provide the underlying reasoning required for agents to maintain context across long-running workstreams. Anthropic has reported that this agentic approach is already heavily utilized internally, with Claude agents currently leading 26% of the company’s own research and development work.
Cloud-Only Execution and Beta Limits
While the update expands the scale of AI-assisted coding, it currently operates with specific technical boundaries. All worker threads run exclusively in the cloud, meaning they cannot access a developer’s local file system or private internal networks at this time. Anthropic has stated that local support is expected to arrive in a future update. Additionally, because each parallel thread initiates a full cloud session, developers should note that running multiple agents simultaneously can rapidly consume the usage limits associated with their subscription tiers.
The updated Projects feature is currently in beta. Access is restricted to select subscribers on the Claude Pro and Max plans. For those with access, the system aims to provide a persistent environment where the AI continues to execute background tasks even after the developer has logged off.
