Anthropic relaunches Claude Code Projects to run parallel cloud agents with shared memory

Anthropic rebuilds Projects around parallel agents

Anthropic has relaunched Projects inside Claude Code, rebuilding the feature as a workspace where several AI agents can work on the same effort at the same time. According to The Verge, each project can hold multiple "threads" that run different tasks in parallel, all drawing on a single shared memory, a common set of goals, and a shared collection of files and artifacts.

The Verge places the release alongside Grok Bot and other tools built to manage groups of AI agents rather than one assistant at a time, an increasingly crowded corner of the market as vendors race to make agents useful in real workflows.

How threads and the coordinator work

The mechanics lean on familiar developer plumbing rather than exotic AI architecture. The Verge reports that each thread is a Claude Code cloud session with its own branch and its own copy of the repository, while a "coordinator" directs the threads and keeps the work organized.

When two threads touch the same code, nothing unusual happens: the overlap surfaces as an ordinary merge conflict, resolved the same way overlaps in any other pull request are. Threads can also divide their delegated work further using subagents, loops, and workflows when a large assignment needs to finish faster, according to The Verge.

Interaction stays flexible. Users can talk to each thread individually, or monitor and adjust the whole effort through the main project chat, which serves as the hub for the project.

Cloud today, local soon, staged rollout

At launch, the threads run in Anthropic's cloud rather than on a developer's machine. The company says support for local tools and local code is coming "very soon," The Verge reports, which frames the current release as the hosted half of a broader roadmap.

Availability is phased. The Verge notes the updated Projects feature entered beta on September 17 for what Anthropic calls "select Claude Pro and Max subscribers." Access is set to expand later to all Pro, Max, Team, and Enterprise users, and the feature is also slated to reach Cowork and ordinary Claude chats, an indication that multi-agent projects may not stay confined to the developer tool where they debuted.

Why it matters

The relaunch changes the basic unit of Claude Code from a single coding assistant into something closer to a small team. If parallel threads with a coordinator and shared memory prove reliable, the bottleneck in agentic coding shifts from what one model can do in one session to how well a fleet of sessions can split up, execute, and reconcile work.

The reconciliation detail matters as much as the parallelism. By resolving overlapping edits through standard merge conflicts rather than a bespoke consensus mechanism, Anthropic is betting that existing Git workflows can absorb multi-agent output without developers learning new tooling, and that the hard questions stay inside code review, where engineers already spend their attention.

The rollout plan is also telling. Bringing Projects to Cowork and regular Claude chats would move multi-agent coordination out of the terminal and into the mainstream product, putting Anthropic in direct competition with the wider wave of agent-orchestration tools rather than only rival coding assistants.