Claude Code's redesigned Projects: Claude scopes the work, delegates it to parallel threads and assembles the result
On September 17, 2026 Anthropic announced a redesigned Projects beta for Claude Code. The user describes a goal; Claude scopes the request, delegates the work to parallel cloud threads on separate repository branches, reviews the outputs and assembles the result. The beta covers select Pro and Max subscribers who have used cloud sessions; Team and Enterprise plans come later.
From folder to conversation: Projects now run the work through a coordinator
On September 17, 2026 Anthropic published a post titled Projects redesigned: from folder to conversation on the claude.com blog. In a Claude Code project the user describes what needs to get done; Claude scopes, delegates, coordinates parallel threads, reviews the outputs and assembles the finished result, and the user can steer throughout, including from a phone, while the work continues.
Anthropic gives two examples: reducing an app's checkout p75 latency, with Claude profiling each endpoint, testing optimizations and opening pull requests in parallel threads; and retiring a deprecated v1 endpoint across connected API, web and mobile repositories, with one thread per repo migrating the callers, running the tests, opening the PRs and saying which ones must merge first. It is separate from the merger of Claude Cowork into chat we covered yesterday: that changed the interface, this changes how the work gets done.
Each thread is a cloud session on its own branch, and the shared memory lives in a MEMORY.md file
Anthropic sums up the model in one line: threads do the work, Claude directs it. Anthropic says to brief Claude "the way you'd brief a chief of staff": Claude routes work to new or existing threads, checks in and follows through. Each thread is a Claude Code cloud session on its own branch and copy of the repo; if two threads touch the same code, the overlap is resolved as a merge conflict like any other PR. With repositories connected a thread opens pull requests and runs tests; with documents it reads them and drafts.
Every thread adds to and draws from a shared project memory, which Anthropic says reduces the need for complex prompt engineering. Its examples: a release that moved to Friday, why an export was dropped, or who to check with before touching the billing service. The docs add that memory is stored as files with a MEMORY.md index every thread reads at start, and that project instructions of up to 16,000 characters go to each new thread. The docs also give the limits: no fixed cap on concurrently running threads, and an enforced limit of 200 new threads per day across a user's projects.
Coordinator plus worker agents instead of one agent, with the limits stated just as plainly
The announcement matters because the coordinator-plus-workers pattern is entering a mainstream developer tool, with assembling and sequencing the results included. VentureBeat (Carl Franzen, September 17, 2026) draws the distinction this way: Claude Projects in chat help Claude remember context, while Claude Code Projects help Claude remember and coordinate work. VentureBeat also reports that there is no standalone pricing for Claude Code Projects; usage draws from the subscriber's existing Claude usage.
Because a project can run several threads at once and each thread is a full Claude Code session, Anthropic says projects can reach usage limits faster, and users can choose the model and effort levels for the coordinator and the workers. Per the docs, a thread that hits the plan's five-hour or weekly limit retries on its own and continues when the limit resets, and an idle project with no running threads does not consume the plan. Threads run in the cloud today; Anthropic describes running on the user's own machine as coming very soon, with no date.
Access depends on the plan, not the country, and the GitHub.com requirement applies to teams in Türkiye too
The sources do not mention Türkiye at all. The beta gate is the plan and the account history, not geography: select Pro and Max subscribers who have used cloud sessions in Claude Code and have no existing projects on the web or desktop. Pro or Max users without access can join a waitlist. Whether users in Türkiye get access on the same schedule, whether any regional restriction exists, and when Team and Enterprise plans will be covered are not stated in the sources; this article assumes the rollout proceeds by plan.
The practical conditions for code work apply to teams in Türkiye as well: the repository must be on github.com with push access and the Claude GitHub App installed; GitHub Enterprise Server, GitLab and Bitbucket are not supported. Projects are available on the web, desktop and mobile apps, not in the terminal CLI, and not through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. A project belongs to one user, cannot be shared and has no organization-level controls during the beta. On cost, VentureBeat reports only that there is no separate charge and that usage draws on the existing subscription; no local pricing or plan information is given.
The UNALSOFT view
Our reading is this: designing a client's workflows as a coordinator plus worker agents rather than one agent is the architecture we use in our agentic AI solutions work, and this announcement shows the pattern arriving in a mainstream developer tool. For SMBs the practical meaning is limited for now: the beta covers only select Pro and Max accounts, projects cannot be shared, and usage limits can fill up faster, so usage planning becomes part of the job. Anthropic says the same setup reads and drafts connected documents, but the sources give no timeline beyond software work.
Is your team's workflow ready for a coordinator and worker agent setup?
Let's map together which tasks can be split across parallel agents and how to plan the usage. A short conversation is enough to start.