Anthropic relaunched the Projects feature inside Claude Code on September 17, 2026, turning it into a control layer for running multiple AI coding agents in parallel. Each Project holds a shared memory, a common set of goals, and a library of files and artifacts, with individual threads working different tasks simultaneously under a single coordinator agent. The design puts Claude Code directly against multi-agent orchestration tools like Grok Bot, and marks Anthropic's clearest bid yet to move coding agents past the single-chat, single-task pattern.
“The revamped projects feature in Claude Code allows users to run multiple agents under the same roof, with a shared memory, goals, and library of files and artifacts.”— Stevie Bonifield, News Writer
Under the hood, every thread inside a Project is a Claude Code cloud session operating on its own branch and its own copy of the repository. The coordinator keeps the work organized across threads, and when two threads touch the same code, the overlap surfaces as a merge conflict — resolved the same way any developer would resolve a pull request. That decision matters: instead of inventing a new conflict-resolution primitive, Anthropic mapped multi-agent coordination onto the git workflow developers already know.
Threads themselves can subdivide. Each one can spawn subagents, loops, and structured workflows to break a large assignment into smaller pieces that run concurrently, which is how Anthropic intends to close the wall-clock gap on bigger jobs. Users can either drop into a specific thread and steer it directly, or manage the whole Project from a single top-level chat that talks to the coordinator.
Key facts
- 01Anthropic relaunched Projects in Claude Code on Sep 17, 2026, letting users run multiple coding agents with shared memory and files.
- 02Each thread runs as a cloud session on its own branch and repo copy, with overlaps resolved as merge conflicts in a standard PR flow.
- 03A coordinator agent directs threads, and each thread can spawn subagents, loops, and workflows to parallelize larger assignments.
- 04Beta access starts with select Claude Pro and Max subscribers, expanding to Team, Enterprise, Cowork, and regular Claude chats.
- 05Threads run only in the cloud at launch; Anthropic says local tools and code support is coming 'very soon.'
The release is beta-gated. Access starts with a subset of Claude Pro and Claude Max subscribers, and Anthropic says it will widen to all Pro, Max, Team, and Enterprise users, plus Cowork and standard Claude chats, on a rolling basis. There is no public price change tied to the feature — Projects rides on the existing tiers.
At launch, threads only run in the cloud. Anthropic says support for local tools and local code is coming "very soon," which is the caveat that will matter most to teams working inside private repos, air-gapped environments, or codebases with dependencies that can't be lifted into a hosted sandbox. Until that lands, Projects is effectively a cloud-only orchestration surface.
The framing also lines up with where the broader coding-agent market is heading. Cursor, led by Michael Truell, has been pushing developers toward AI-assisted workflows inside the editor, while Anthropic is pushing outward from the model, treating the repo and the coordinator as the primary interface rather than the IDE. Both bets assume that the bottleneck is no longer code generation quality but coordination — how many agents you can run in parallel without them stepping on each other.
There are real open questions. Merge-conflict resolution works for humans because humans reason about intent; a coordinator agent resolving conflicts between two other agents is a newer problem, and how often those conflicts require human intervention will decide whether Projects actually saves time or just relocates the work. The cloud-only constraint at launch also means teams cannot yet test Projects against their real production stacks if those stacks include local dependencies.
For Anthropic, Projects is a wedge into a workflow that pure model access does not capture. Selling API tokens to developers is a commodity race; selling a coordinator that manages a fleet of agents against a shared repo is a stickier product, and one that maps directly onto how engineering teams already think about branches, reviews, and merges. If the beta holds up and local support ships on the promised timeline, this is the feature that turns Claude Code from a coding assistant into an execution layer — and that is a very different product to compete with.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
