From Folder to Conversation: Anthropic Redesigns Claude Code Projects Around a Coordinating Agent and Parallel Cloud Threads

On September 17, 2026, Anthropic shipped one of the more consequential quiet redesigns of the agentic coding era. Claude Code Projects — previously a way to group context and files — has been rebuilt around a new mental model the company summarizes in five words: “from folder to conversation.” Instead of a container you organize, a project is now something you brief. You describe an outcome, and Claude scopes the request, delegates the work, coordinates parallel threads, reviews outputs, and assembles the finished result.

The beta is live today for select Claude Pro and Max subscribers who use cloud sessions in Claude Code and don’t have existing projects on the web or desktop apps. Anthropic says access will widen to more Claude Code users on those plans “over the coming week,” with updated projects rolling out across all of Claude — including Team and Enterprise plans — after that. A waitlist is open for eligible users who don’t yet have access, and existing projects keep working unchanged until the rollout reaches chat and Cowork.

The old problem: humans as the scheduler

The blog post opens with a pain every heavy Claude Code user knows: managing multiple sessions across a build meant dividing the work yourself, juggling handoffs, and stitching results back together. The human was the orchestrator; the model was a fleet of talented but amnesic contractors.

The redesign inverts that. Anthropic’s own example reads like a job posting for a chief of staff: set a goal to reduce your app’s checkout p75 latency, then ask Claude to profile each endpoint, test optimizations, and open pull requests in parallel threads. Or connect your API, web, and mobile repositories and set a goal to retire a deprecated v1 endpoint — Claude creates a thread per repo, migrates the callers, runs the tests, opens the PRs, and then tells you which ones need to merge first.

That last detail is quietly significant. Merge-order reasoning is coordination knowledge that previously lived only in a tech lead’s head. Encoding it into the coordinator’s job description moves agent tooling from “writes the code” toward “owns the delivery.”

Threads do the work; a coordinator directs them

Architecturally, the new projects split into two roles. Threads are the workers, and each one is a full Claude Code cloud session operating on its own branch and its own copy of the repository. When a thread completes a unit of work, it opens pull requests and runs the test suite like any other contributor. If two threads touch the same code, the overlap surfaces as an ordinary merge conflict — no magic reconciliation layer, just the same primitive every engineering team already understands.

The coordinator is the other half. When you start a project, you select a goal and the repo or context, and Claude immediately suggests work it can pick up. You configure the project’s cloud environment, connectors, plugins, instructions, and model. From there you can monitor progress in the main project chat or dive into any individual thread to examine and steer the details. Threads themselves can fan out further, splitting delegated work into subagents, loops, and workflows when an assignment is large enough to benefit from it.

Crucially, the system is asynchronous by design. You can steer progress from your phone, and it keeps working after you step away from your computer. That is the operational promise of cloud sessions — decoupling agent runtime from developer attention — now applied to a multi-thread project rather than a single conversation.

Shared memory: the compounding asset

The most strategically interesting piece may be the memory layer. Every thread in a project now adds to and draws from a shared memory, which Anthropic frames as reducing the need for complex prompt engineering. The examples are mundane in the best way: Claude can remember that the release moved to Friday, why the export feature was dropped, or who to check in with before touching the billing service.

This is the difference between a project as a workspace and a project as an institution. Individual sessions expire; institutional memory accrues. Anthropic also says Claude remembers your working and communication style over time — how often it checks in, how frequently it starts new threads, how detailed each update should be — all of which you can ask it to adjust.

Alongside memory, projects gained a library that collects both the files you add and the artifacts Claude produces, making it easier for new work to build on past efforts rather than starting from zero.

The economics: parallelism burns limits

Anthropic is unusually candid about the cost model: because each thread is a full Claude Code session and projects can run several simultaneously, “projects can reach usage limits faster.” The mitigation is control — project-specific usage dashboards, plus the ability to select the model and effort level independently for the coordinator chat and for worker threads. That last knob matters more than it looks. A coordinator mostly needs to route and review; worker threads need depth. Splitting the two lets a team spend its token budget where the difficulty actually is.

One limitation is worth flagging for enterprise buyers: threads run in the cloud today. Running on your own machine — alongside local tools and code, behind your network — is listed as “coming very soon,” which keeps air-gapped or compliance-constrained shops waiting for now.

What it means

The redesign lands in a market where agentic coding tools are converging on the same thesis: the unit of value is no longer a completion or a chat, but a delegable objective executed by a supervised swarm. Anthropic’s implementation stands out less for any single feature than for its choice of primitives — plain git branches and plain merge conflicts as the coordination substrate, wrapped in a coordinator with durable memory.

For developers, the immediate takeaway is practical: if you’re on Pro or Max and eligible, this beta is worth exercising on a real multi-repo chore. For the industry, it’s another data point in this year’s clearest trend — the terminal is becoming a staffing function, and the question is shifting from “can the model do the task” to “can it run the project.”

Anthropic has not yet said when the redesign exits beta, or what the Enterprise rollout will change in terms of administration and audit controls. Those details will decide how fast this moves from a power-user convenience to the default way teams ship software.