The Pulse
Anthropic Turns Claude Code Projects Into Multi-Agent Workspaces
Reviewed the announcement for a verbatim statement attributed to a named person and role; it contains no attributable quotation.
AI.info Team ·
Claude Projects Get a Coordinator
Anthropic is changing Claude Projects from a place to collect conversations into a workspace that can assign and manage software tasks across multiple Claude Code sessions. In a product announcement dated September 17, the company says users can describe a goal while Claude scopes the work, delegates tasks, coordinates parallel threads, reviews the results, and assembles the outcome.
The redesigned Projects experience enters beta for selected Claude Pro and Max subscribers who use cloud sessions in Claude Code and do not already have Projects on the web or desktop. Anthropic says access will expand to more Claude Code users on those plans during the following week, with updated Projects for all Claude users, as well as Team and Enterprise plans, coming later.
Anthropic’s examples focus on work that would normally require a developer to divide a large assignment manually. A team could ask Claude to reduce an application’s checkout p75 latency, for instance, and have it profile endpoints, test optimizations, and open pull requests in parallel threads. A separate project could connect API, web, and mobile repositories, then create a thread for each repository to migrate callers away from a deprecated version-one endpoint.
Each Thread Runs Its Own Claude Code Session
The new structure separates management from execution. A coordinator handles the project conversation, while individual threads perform the assigned work. Users can monitor progress in the main chat, inspect individual threads, and redirect them when the work moves off course.
When a Project begins, users select a goal and provide a repository or other context. Anthropic says they can also configure the cloud environment, connectors, plugins, instructions, and model. Claude can suggest work immediately, route new requests to existing or newly created threads, and check back as tasks progress.
Each thread runs as a Claude Code cloud session on its own branch and copy of the repository. That arrangement gives concurrent tasks separate working environments, but it does not remove ordinary software-development conflicts: when threads touch the same code, overlapping changes can produce a merge conflict that must be resolved like any other pull request.
Threads can also divide their own assignments through subagents, loops, and workflows. Anthropic presents that layered structure as a way to complete larger jobs faster, although the company does not publish performance measurements for the new Projects design.
Shared Memory Replaces Repeated Briefings
Anthropic is also adding shared memory across a Project. Every thread can contribute to and draw from the same record of project details, decisions, and results, reducing the need to repeat background information in each new conversation.
The announcement gives practical examples of what Claude may retain: a release date moving to Friday, the reason an export was dropped, or the person who should be consulted before changes reach a billing service. Users can ask Claude to change how often it checks in, how frequently it starts new threads, and how detailed its updates should be.
A new library stores files added by the user along with artifacts produced by Claude. That gives later threads a shared collection of project materials rather than forcing each task to begin with a fresh upload or a separate handoff.
More Parallel Sessions Also Mean Faster Usage Limits
The product’s operating model carries a direct trade-off. Anthropic says each thread is a full Claude Code session, so Projects that run several threads at once can reach usage limits more quickly than a single conversation.
Users can check usage for an individual Project and select the model and effort level for both the coordinator chat and worker threads. The controls matter because the new system turns one request into a group of active sessions, each consuming resources while the coordinator continues to manage the overall assignment.
For now, threads run in Anthropic’s cloud. The company says support for running them on a user’s own machine, alongside local tools and code and behind the user’s network, is coming soon. Existing Pro and Max Projects will continue operating as before while Anthropic expands the redesigned experience to Claude chat and Cowork.
From Parallel Coding to Persistent Work
The release places Claude Projects closer to a persistent work-management system than a conventional chat folder. A user can state a multi-part goal, let Claude open separate workstreams, review progress from a central conversation, and return to the accumulated decisions and artifacts later.
That design is aimed at work that extends beyond one response: repository migrations, performance investigations, document-heavy assignments, and other projects that need repeated execution and review. Anthropic’s announcement does not claim that the system eliminates developer oversight. Pull requests, tests, branch conflicts, usage controls, and user guidance remain part of the workflow.
The beta begins with selected Pro and Max subscribers using cloud-based Claude Code sessions on September 17, 2026. Anthropic’s next stated step is wider access on those plans, followed by expansion across Claude, Team, and Enterprise.