Claude Code Projects Redesigned With Multi-Agent Threads

Introduction

Anthropic redesigned Claude's Projects feature and is launching the new version first in Claude Code, moving a project, in the words of its announcement, "from folder to conversation": a coordinated workspace where multiple agents work in parallel. The change, announced on the Claude blog on September 17, 2026, replaces the earlier model of manually dividing work across sessions with a coordinator that scopes a request, delegates pieces of it, and assembles the finished result. Instead of running one Claude Code session at a time, a project can now run several cloud sessions at once, each working toward a shared goal.

The redesign targets a specific pain point: coordinating multi-part engineering work across sessions. Anthropic's own example is a project configured with the goal of reducing an app's checkout p75 latency, where Claude profiles each endpoint, tests optimizations, and opens pull requests in parallel threads. A second example connects API, web, and mobile repositories with a goal to retire a deprecated v1 endpoint; Claude creates one thread per repository to migrate callers, run tests, open pull requests, and report which ones need to merge first.

Feature Overview

Threads and a coordinator. Every project now has threads that do the work and a coordinator that directs them. When a project starts, the user selects a goal along with a repository or context, and Claude suggests work it can begin immediately. Users can configure the project's cloud environment, connectors, plugins, instructions, and model. Progress can be monitored and steered from the main project chat, or by opening an individual thread to review its details directly, similar to briefing a chief of staff who then routes tasks to new or existing threads.

Isolated execution per thread. Under the hood, each thread is a full Claude Code cloud session running on its own branch and its own copy of the repository. If two threads happen to touch the same code, the overlap surfaces as a standard merge conflict rather than a silent overwrite, the same way any two pull requests would collide. Each thread can also break its assigned work into smaller pieces using subagents, loops, and workflows so larger assignments finish faster.

Shared memory across the project. Every thread now contributes to and reads from a shared project memory, which Anthropic says reduces the need for complex prompt engineering. The company's example: Claude can recall that a release moved to Friday, why a particular export was dropped, or who to check with before touching the billing service, without the user restating that context in a new thread. Claude also adapts to a user's working and communication style, including how often it checks in, how frequently it opens new threads, and how much detail its updates include.

A library for files and artifacts. Alongside memory, projects now include a library that collects files a user adds and artifacts Claude produces, making earlier materials easier to locate when new work builds on past output.

The availability terms are specific rather than open-ended. Updated projects are live in beta starting September 17, 2026, for select Claude Pro and Max subscribers who use cloud sessions in Claude Code and don't already have existing projects on the web or desktop apps. Anthropic says access will expand to more Claude Code users on those plans over the following week, with updated projects coming to the rest of Claude and to Team and Enterprise plans after that. Subscribers on Pro or Max without access yet can join a waitlist. Existing projects on Pro and Max continue working as they do today and will be upgraded once the rollout reaches Claude's chat and Cowork surfaces.

Usability Analysis

For a developer used to running one Claude Code session and manually stitching results together, the shift is largely organizational: instead of opening several terminals or cloud sessions and tracking which one is doing what, a single project chat becomes the entry point, with the coordinator handling delegation. The Verge's coverage frames this as similar in spirit to other tools that manage groups of AI agents, where a coordinator directs several worker agents at once rather than a user juggling them directly.

Because each thread is a complete Claude Code cloud session, running several at once naturally consumes more usage than a single conversation would. Anthropic addressed this directly by adding project-specific usage tracking and letting users select the model and effort levels used by the coordinator chat as well as by the worker threads, so capability can be traded against usage at both levels.

One real constraint at launch: threads run only in the cloud. Anthropic says running threads on a user's own machine, alongside local tools, code, and behind a private network, is "coming very soon" but is not yet available. Teams that depend on local-only tooling or resources behind their own network will need to wait for that capability before adopting the redesigned projects for those workflows.

Pros and Cons

The coordinator model removes a real amount of manual orchestration: rather than deciding how to split a task, opening multiple sessions, and later reconciling their outputs by hand, a single goal description lets Claude propose and route the work itself. The shared memory system is a meaningful improvement for long-running engineering work, since decisions and context persist across a project's lifetime instead of resetting with every new session. Git-style conflict resolution between threads is a sensible design choice, since it reuses a workflow developers already understand rather than inventing a new merge process. The ability to steer progress from a phone, and to have work continue after a user steps away, extends what was previously tied to an active terminal session.

The most immediate limitation is availability: the beta covers only select Pro and Max subscribers who use cloud sessions in Claude Code and have no existing web or desktop projects, meaning most current project users cannot try the redesign yet. The lack of local execution at launch is a second limitation for teams whose code, tools, or compliance requirements keep work off Anthropic's cloud. Finally, since each thread is a full cloud session, parallel work can draw down usage allowances noticeably faster than a single conversation, a tradeoff Anthropic addresses with usage visibility and model and effort controls rather than eliminating.

Outlook

The Decoder frames this release as another step toward code automation, following Anthropic's recent move to make auto mode the default in Claude Code. Anthropic's plan to bring updated projects to the rest of Claude, including chat and Cowork, and to Team and Enterprise plans suggests Anthropic sees multi-agent coordination as a pattern for Claude broadly rather than a coding-only feature. Bringing threads to local machines, as promised, would matter most for teams that need to keep code and tooling on their own infrastructure, and is the clearest signal to watch for as the beta widens over the coming weeks.

Conclusion

The Projects redesign is a structural change to how Claude Code handles multi-part work, replacing manual session management with a coordinator that delegates to parallel, memory-sharing threads. It is most useful today for engineering teams comfortable working entirely in the cloud and already inside the Claude Pro or Max cloud-session beta; teams needing local execution, or outside the initial rollout, will need to wait for the promised expansion.

Editor's Verdict

Claude Code Projects Redesigned With Multi-Agent Threads earns a solid recommendation within the Claude space.

The strongest case for paying attention: a coordinator automatically delegates work across parallel threads, removing the need to manually split tasks and stitch results back together. That alone raises the bar for what readers should expect in this space. Reinforcing that, the shared memory system lets a project retain context across sessions, reducing repetitive prompt engineering for long-running work — practical value rather than just headline appeal. The broader signal worth registering is straightforward: the redesign turns a Claude Code project from a folder of files into a coordinated multi-agent workspace, with a coordinator directing parallel threads instead of requiring manual handoffs. On the other side of the ledger, one constraint is real rather than a marketing footnote: the beta is limited to select Claude Pro and Max subscribers with cloud sessions in Claude Code and no existing web or desktop projects, so most current users must wait. It should factor into any serious decision. Layered on top of that, the local tool and code execution path is not available at launch, since threads currently run only in the cloud until on-machine support arrives — which narrows the set of teams for whom this is an obvious yes.

For Anthropic and Claude users, alignment-focused teams, and developers already invested in the Claude ecosystem, this is a serious evaluation candidate, not just a curiosity to bookmark. For everyone else, the safer posture is to monitor coverage and revisit once the use cases that matter to your team are demonstrated in the wild.