Anthropic has relaunched Projects in Claude Code, turning the coding tool into a hub where several AI agents can work at once under a single project umbrella. The feature began rolling out in beta on September 17, 2026, starting with select Claude Pro and Max subscribers.
The relaunch places Claude Code closer to a small development team than a single chat assistant. Instead of one agent responding to one request, Projects organizes many parallel workers around shared context and a coordinating layer. The approach echoes a broader push across the AI coding market toward tools that can manage multiple agents rather than just one.
One project, many agents
At the center of the relaunched feature is a shared workspace. Every project carries common memory, goals, and a library of files and artifacts. Those resources are available to the agents working inside it, so separate tasks do not start from a blank slate.
Within a project, work is divided into threads. Each thread runs a different task in parallel, and each one is a Claude Code cloud session. A coordinator sits above the threads and directs them, keeping the overall effort organized as the pieces move forward.
The architecture is meant to prevent parallel work from turning into parallel chaos. Because every thread operates on its own branch and its own copy of the repository, agents can make changes without immediately stepping on one another. When two threads touch the same code, the overlap surfaces as a merge conflict.
Conflicts handled like pull requests
Anthropic is not inventing a new conflict-resolution process for the multi-agent era. According to the company’s description, merge conflicts are handled like standard pull requests. That keeps the resolution step familiar to developers who already review and merge code in a normal workflow.
The branch-and-repo-copy model also gives each thread a clean boundary. A thread can pursue its assignment independently, and the coordinator can track how the separate pieces fit together. When changes collide, the conflict becomes a reviewable event rather than an invisible failure.
Scaling work with subagents, loops, and workflows
Threads are not limited to a single layer of delegation. They can split their assigned work further using subagents, loops, and workflows. That makes the system suited to larger assignments where one agent would otherwise carry the entire load.
The coordinator remains the organizing force across all of it. It directs the threads, while the threads can fan out into smaller units of work. The result is a hierarchy of agents rather than a flat group of independent sessions.
Anthropic has not specified how many agents or threads a project can run. The beta will be the first broad test of how the coordination model holds up in practice.
How users stay in the loop
Developers are not locked out of the process while the agents work. A user can interact with each thread individually, stepping into a specific task when needed. They can also monitor and update the project through a main project chat, which serves as the central point for oversight.
That dual level of interaction matters for a system built on parallel execution. It lets a user drill into one thread’s work without losing sight of the larger project, or stay at the coordinator level and adjust direction across the board.
Cloud first, local support coming
At launch, threads run in the cloud. Anthropic says local tool and code support is coming very soon, but the initial beta is cloud-only. That keeps the first release focused on the hosted session model before extending to local development environments.
The beta is available starting today for select Pro and Max subscribers. Anthropic plans to widen access later to all Pro, Max, Team, and Enterprise users. The company has also said access will expand to Cowork and regular Claude chats at a later point, though it has not given a specific timeline for those steps.
Part of a crowded multi-agent race
The relaunched Projects feature puts Claude Code in the same category as Grok Bot and other tools built to manage multiple AI agents. The comparison is not incidental: multi-agent orchestration has become a competitive front among AI coding products.
For Anthropic, the bet is that developers want more than a faster single assistant. Projects offers a structure for delegating work across many agents while keeping shared context, a coordinator, and a familiar conflict-resolution path. Whether that structure becomes the default way developers work with AI remains an open question, but the beta now puts it in users’ hands.
