Anthropic Relaunches Claude Code Projects to Coordinate Multiple Cloud Agents

Anthropic has relaunched Projects inside Claude Code, a feature that lets users manage multiple AI agents working simultaneously in the cloud, according to a report from The Verge. The update introduces a system where a "coordinator" agent directs several parallel "threads," each handling separate tasks while sharing memory, goals, and a common library of files and artifacts.

How it works

Each thread operates as its own Claude Code cloud session, working on a separate branch and copy of the same repository. If two threads modify the same code, the system resolves the overlap as a merge conflict — the same process developers already use for pull requests. Threads can further break down their assigned work using subagents, loops, and workflows, allowing large tasks to complete faster by splitting them into smaller pieces.

Users can interact with individual threads directly or manage the entire operation through a main project chat that monitors and updates all agents at once. Anthropic compares the setup to tools like Grok Bot, which similarly let users direct groups of coordinated agents rather than a single assistant.

Availability and rollout

At launch, all threads run exclusively in the cloud. Anthropic says support for local tools and local codebases is coming "very soon," though the company has not specified a date. The updated Projects feature is live in beta starting today for "select Claude Pro and Max subscribers." Anthropic plans to expand access to all Pro, Max, Team, and Enterprise users, and eventually bring the feature to Cowork and standard Claude chat interfaces.

No pricing details specific to Projects have been disclosed. Anthropic has not published benchmark data or technical specifications quantifying performance gains from running multiple coordinated threads versus a single Claude Code session.

What this means

This is a product and workflow change, not a new model release — the underlying Claude models powering each thread are unchanged. The move reflects a broader industry shift toward multi-agent orchestration, where the value proposition isn't a smarter single model but better management of many agent instances working concurrently on different parts of the same problem.

The merge-conflict resolution approach is notable because it borrows directly from existing software engineering practice rather than inventing a new coordination primitive, which likely makes the system easier for developers to reason about and debug. The bigger open question is reliability at scale: coordinating multiple autonomous agents editing a shared codebase introduces failure modes — race conditions, redundant work, conflicting assumptions across threads — that a single-agent workflow doesn't have to manage.

Anthropic's staged rollout, starting with a narrow beta before expanding to Team and Enterprise tiers, suggests the company is treating this as an experimental capability rather than a finished product. The promised addition of local tool support will matter most for developers who can't or won't run code exclusively in Anthropic's cloud environment, and its timeline remains unconfirmed.