Opportunity assessment: high. Anthropic just relaunched Projects inside Claude Code, and the new version is built for one job: running several AI agents at once under a single roof. According to The Verge AI, each project now bundles shared memory, shared goals, and a common library of files and artifacts, then splits the actual work across parallel “threads” that a “coordinator” agent directs.
That’s a meaningful shift. Claude Code started life as a single agent you talked to in a terminal. This update turns it into something closer to a small engineering team you manage from one chat window.
Situation Report
Here’s what The Verge AI reports is actually running under the hood:
- Each project holds shared memory, goals, and a file library that every thread can see.
- Each thread is its own Claude Code cloud session, working on its own branch and its own copy of the repo.
- A coordinator agent keeps the threads organized and tracks overall progress.
- If two threads touch the same code, the overlap gets resolved as a merge conflict, “just like any other PR.”
- Threads can split their own delegated work further using subagents, loops, and workflows, so big assignments finish faster.
- You can talk to any thread directly, or monitor and steer everything from the main project chat.
The Verge AI compares the setup to Grok Bot and other tools built to manage groups of AI agents. The pattern is becoming familiar: one orchestrator, many workers, a shared context layer holding it together.
Why the Branch-Per-Thread Design Matters
What stands out here is that Anthropic didn’t invent a new coordination system. It leaned on git. Every thread gets its own branch and repo copy, and conflicts surface as merge conflicts, not as some proprietary “agent disagreement” state you’d need to learn.
That’s the pragmatic choice. Engineering teams already know how to review branches and resolve conflicts. Dropping AI agents into that existing workflow means less retraining and fewer surprises when two agents step on each other’s code.
It also makes the output auditable. Each thread’s work lands as a discrete branch you can inspect, test, or throw away. You aren’t trusting one giant agent to keep a massive change coherent in its head.
Tactical Use Cases
Based on the capabilities described, the obvious applications look like this:
- Large refactors. Hand the coordinator a goal, let threads take separate modules in parallel, then review the merges.
- Feature plus tests plus docs. Three threads, three jobs, one project chat to watch them all.
- Bug sweeps. Assign each thread a cluster of issues and let the coordinator track what’s closed.
- Exploration. Spin up threads to prototype competing approaches on separate branches, then keep the winner.
Constraints and Caveats
Two limitations worth flagging from the original report:
- Cloud only, for now. At launch, threads run in the cloud. Anthropic says support for local tools and code is coming “very soon,” but if your workflow depends on local dependencies or private infra, you’re waiting.
- Gated beta. Access starts today for “select Claude Pro and Max subscribers.” Everyone else, including Team and Enterprise, gets it later.
There’s also the unspoken cost question. Running four or five cloud sessions in parallel, each potentially spawning subagents, burns through usage far faster than a single session. The Verge AI doesn’t cover pricing details, so budget accordingly if you get early access.
Availability
- Today: Beta for select Claude Pro and Max subscribers.
- Later: All Pro, Max, Team, and Enterprise users.
- Also coming: The feature extends to Cowork and regular Claude chats, not just Claude Code.
That last point is easy to skip past, but it’s significant. Multi-agent projects won’t stay a developer-only tool. Anthropic is planning to bring the same coordinator-and-threads model to everyday Claude conversations, which puts agent orchestration in front of a much broader audience.
Assessment
This is Anthropic staking out the “agent manager” layer before competitors lock it down. The hard part of multi-agent work was never spinning up more agents. It was keeping them coherent, keeping their work reviewable, and giving a human one place to steer. Projects addresses all three by reusing tools developers already trust.
The real test comes when local support lands and the beta widens. Until then, this is a preview of how Anthropic thinks you’ll be running AI teams, not just AI assistants. The full breakdown is in The Verge AI’s original coverage.