Anthropic Relaunches Claude Code Projects to Run Multiple AI Agents in Parallel

Anthropic redesigned the Projects feature inside Claude Code on September 17, turning what used to be a shared folder of chats and reference files into a single, persistent conversation that acts as a coordinator directing several AI agents working at the same time. Anthropic’s own announcement, titled “Projects redesigned: from folder to conversation,” frames the change as a shift from static organization to active delegation: instead of opening a new Claude Code session and re-explaining the state of an ongoing engineering effort every time, a developer now briefs one project chat, and Claude splits the work across separate cloud sessions that keep running on their own.

The Verge and VentureBeat both reported on the launch the same day. Anthropic is treating the feature as a beta, not as a finished product available to everyone at once.

A Coordinator That Delegates to Cloud Sessions

The redesigned Projects has two layers. The project’s main chat is the coordinator: a user briefs it on a goal roughly the way they would “brief a chief of staff,” and the coordinator decides whether a new request belongs in an existing thread or needs a new one. Under the coordinator sit the threads themselves, and each thread is a full Claude Code cloud session running on its own branch and copy of the repository. As Anthropic puts it, “threads do the work, Claude directs it.” A thread can further split its own assignment using subagents, loops, and workflows so that large jobs finish faster, and a user can drop into any individual thread to check on it or step in directly, or manage everything from the coordinator’s project-level chat instead.

If two threads happen to touch the same code, Anthropic and The Verge both describe the resolution as ordinary and unglamorous: the overlap becomes a merge conflict, handled the same way any other pull request conflict would be. Every thread also reads from and writes to one shared project memory, which Anthropic says can carry forward specifics such as a release date that moved, why a feature was dropped, or who to check in with before touching a sensitive part of the codebase, “reducing the need for complex prompt engineering” each time a new thread starts. Separately, Claude also tracks a user’s working and communication style, including how often it should check in or how frequently it should start new threads. A project library holds files a user has added and artifacts Claude has generated, available to every thread in the project.

How This Differs From the Original Claude Projects

Anthropic’s Claude Projects has existed since June 2024, when it launched for Claude.ai Pro and Team users as a way to group chats around a shared set of uploaded files and instructions inside a roughly 200,000-token context window. That feature still exists and, per Anthropic, continues to work unchanged during this rollout. What launched today reuses the Projects name but solves a different problem: coordinating agentic software development across several concurrent Claude Code sessions, rather than giving a single conversation a larger pile of reference material to draw on.

VentureBeat frames the distinction against rival tools: where OpenAI’s Codex offers a multi-agent workspace and GitHub’s approach is repository-native, Anthropic is positioning a persistent conversational coordinator as the primary interface a developer supervises everything through. The Verge separately compares the underlying idea, one operator directing a roster of agents, to xAI’s Grok Bot and other tools built around managing groups of AI agents rather than a single assistant.

Cloud-Only for Now, and a Real Usage Tradeoff

At launch, every thread runs in Anthropic’s cloud rather than on a developer’s own machine or behind a private network. Anthropic says support for local tools and code is “coming very soon,” though it has not given a firm date. Because a project can spin up several threads simultaneously and each one is a complete Claude Code session, Anthropic itself acknowledges that “projects can reach usage limits faster” than a single conversation would, a caveat VentureBeat also flagged in its own coverage. The feature does not carry its own separate price; it draws directly from a user’s existing Pro or Max plan. Anthropic says users can track project-specific usage and choose the model and effort level for both the coordinator and the individual worker threads, which is the main lever available for managing that cost.

A Narrow Beta, for Now

The redesigned Projects is available today only to “select Claude Pro and Max subscribers” using cloud sessions in Claude Code who do not already have the older web or desktop Projects feature enabled, according to Anthropic. Access is set to expand to more Pro and Max users over the coming week, with Team and Enterprise plans, along with Anthropic’s Cowork product and regular Claude chats, gaining access at a later, unspecified date. Pro and Max subscribers who do not yet have access can join a waitlist, and Anthropic says existing projects keep working exactly as they do today until the rollout reaches them.

The launch lands one day after Anthropic merged the separate front ends for Claude chat and Cowork into a single window. Taken together, the two changes point toward the same goal: making Claude’s various surfaces (chat, Cowork, and now Claude Code’s Projects) feel like one continuous product rather than a set of separate tools a developer has to choose between.