Anthropic Rebuilds Claude Projects to Run Parallel AI Coding Agents

Anthropic has redesigned Claude Projects around a coordinator that dispatches and supervises multiple Claude Code cloud sessions in parallel. A user describes the desired outcome; the coordinator scopes the job, assigns pieces to worker threads, reviews their output, and returns a consolidated result. Because those workers run remotely, they can continue after the user’s laptop closes.

The beta debuts in Claude Code, where projects previously centered on a file collection and one conversation. Multi-session builds required developers to divide the work, manage handoffs, and combine the results. In its launch post, Anthropic says the coordinator now handles those orchestration tasks.

One coordinator, many branches

The redesigned architecture gives each project two layers: a coordinator that receives instructions and worker threads that execute them. The coordinator can route each request to a new or existing thread, monitor progress, review outputs, and assemble the final result.

Each worker thread runs as a Claude Code cloud session with its own branch and isolated copy of the repository. Normal Git merge rules still apply. When threads edit the same lines, developers may need to resolve a merge conflict before combining their work. Threads can also divide larger assignments among subagents, allowing a migration or investigation to expand without requiring the user to design every step.

Parallel work in practice

• Checkout performance: Set a goal to reduce p75 latency, the response-time threshold met by 75% of requests. Claude can profile each endpoint, test optimizations, and open separate pull requests in parallel.
• API retirement: Connect API, web, and mobile repositories, then ask Claude to retire a deprecated v1 endpoint. It can create a thread for each repository, migrate callers, run tests, open pull requests, and report the required merge order.

Cloud execution keeps those threads running after the laptop closes. An Overview panel identifies work awaiting input, while mobile access lets users answer questions or redirect a thread during a long-running build.

Memory that follows the project

Shared project memory allows every thread to contribute context and retrieve it later. Claude can retain details such as a delayed release date, the reason an export was removed, or the team that owns a billing service, reducing the need to paste the same background into each conversation.

A project library stores uploaded files and artifacts produced by Claude, giving later threads access to earlier outputs. Projects can also retain working preferences, including how often Claude should request input and how much detail its updates should contain.

Who gets the beta

• Initial access: Selected Claude Pro and Max subscribers who use Claude Code cloud sessions and have no existing projects on the web or desktop.
• Next phase: More Claude Code users on Pro and Max during the week after the announcement.
• Later rollout: Updated projects across Claude, followed by Team and Enterprise plans. Anthropic has not provided dates for those stages.

Pro and Max subscribers without access can join the waitlist. Existing projects on those plans will continue working unchanged until the rollout reaches their accounts.

Usage and cloud boundaries

1. Parallel sessions consume more usage. Every worker is a full Claude Code session, so a project running several threads can reach plan limits faster. Users can inspect project-specific usage and choose the model and effort level for both the coordinator and worker threads.
2. Workers currently run only in the cloud. Projects cannot access files that remain on a laptop or services available only through a private network or VPN. Anthropic says support for local code, tools, and network resources is coming “very soon,” although it has not announced a date.

Orchestration moves into Claude

Agent frameworks often require developers to define subagents, decide when to launch them, track their state, and integrate their results. Claude Projects puts those controls inside a persistent workspace with shared memory, stored artifacts, cloud execution, and branch-based isolation.

The design’s usefulness will depend on how reliably the coordinator scopes work, preserves constraints, catches failures, and combines results. Developers still need to review pull requests and resolve overlapping edits; delegation, status tracking, and handoffs can continue in the service. The approach is best suited to long-running tasks that divide cleanly, including test expansion, endpoint migrations, performance investigations, and coordinated changes across repositories.