Anthropic Adds a Coordinator to Claude Projects for Running AI Work in Parallel

Anthropic has introduced a new coordinator feature for Claude Projects, aimed at simplifying the process of running AI work in parallel. Previously, users had to manually split tasks, track progress, and combine results when working on multiple Claude Code sessions simultaneously. The coordinator thread in the redesigned project manager breaks down a stated goal into parallel threads, assigns work to worker threads, and combines the results.

Each thread runs as a full Claude Code cloud session on its own branch, with the capability to spawn subagents, loops, or workflows for complex tasks. Optimizing checkout latency or retiring a deprecated API are examples of the kind of workload this feature targets. These tasks may seem simple in concept but can be logistically challenging without proper coordination.

Merges are handled through standard pull requests, which allows for review and approval, similar to the process for work from a junior engineer. This approach ensures that users maintain control over the merging process rather than relying on an automated system.

A notable improvement in Anthropic's redesign is the persistence of context across threads within a project. This means that deadlines, technical decisions, communication preferences, and release dates are retained, reducing the need for re-explaining context to new sessions. A project library centralizes uploaded files and Claude-generated artifacts, making it easier for subsequent threads to build upon each other's output.

Users can customize the coordinator's autonomy through controls that dictate how often it checks in, how aggressively it creates new threads, and the level of detail in updates. Additionally, projects are accessible through a phone interface, providing a convenient way for users to review and approve changes remotely.

The rollout of this new feature is staged, starting with beta access for Claude Pro and Max subscribers who haven't previously set up projects on web or desktop platforms. Anthropic plans to expand access to chat, Cowork, Team, and Enterprise plans in the coming weeks. This phased approach allows the company to identify any potential issues before a wider release.

Mitch Ashley, vice president and practice lead for CIO & Technology Buyers and Software Lifecycle Engineering at The Futurum Group, sees the shift to PR-based merging as a crucial factor in the adoption of this feature. While running work in parallel is relatively easy to implement, the trustworthiness of the system relies on the quality of the coordination logic that decides when to merge and what constitutes completion.

This operational question, rather than a purely modeling one, will be key for DevOps teams evaluating the feature.