Claude Code Projects turns one conversation into a coordinator for parallel coding agents

Long-running software work often outgrows a single coding-agent chat. Requirements change, tasks branch, tests fail, and decisions made days earlier must remain available to the next session.

Anthropic’s redesigned Claude Code Projects addresses that problem by making a project one persistent conversation with a coordinating role. Instead of manually managing a folder and a series of isolated chats, a developer can describe evolving work in one place while Claude breaks suitable tasks into parallel cloud-based threads. The beta started on September 17 for select Claude Pro and Max subscribers using Claude Code cloud sessions. 810

The central change: one coordinator, many coding threads

A Claude Code Project is designed around a continuing project conversation. The coordinator receives new requests, answers simple questions when appropriate, and creates threads for work that should run separately. Those threads report their results back to the main conversation, giving the developer a single place to follow the broader objective. 18

Each thread is a complete Claude Code cloud session with its own copy of the repository and its own branch. That isolation allows several tasks to run at once rather than forcing a developer to wait for one agent session to finish before starting the next. Threads can continue running after the developer closes their laptop. 158

In practical terms, a project can divide a larger goal into discrete implementation or investigation tasks. Individual threads can work with the repository, run tests, and open pull requests when the work calls for it, while the main conversation remains the place to direct follow-up work and assess what is complete or blocked. 2813

What stays persistent across the project

The feature’s value is not simply parallel execution. It is the context shared across those executions.

Every new thread begins with the project’s repositories, files, instructions, and memory. Project memory can carry forward decisions, priorities, constraints, and other context that would otherwise need to be restated in each new coding session. 8

Projects also include a Library that collects files supplied by the user and artifacts produced by Claude. That creates a persistent source of reference material for later work rather than leaving useful outputs buried in a past conversation. 58

This makes the redesigned experience meaningfully different from using separate chats for related tasks: the system is intended to preserve the evolving working context as well as the initial project information.

Parallel work does not eliminate Git conflicts

Separate branches and repository copies reduce direct interference between threads, but they do not make concurrent edits automatically compatible. If two threads modify the same code, the overlap is handled as a conventional Git merge conflict, just as it would be with two human-authored pull requests. 5

That distinction matters. Claude Code Projects can coordinate and isolate work, but it does not replace engineering judgment around code review, integration, test coverage, or release approval.

Usage and cloud-only constraints

Parallelism is not free capacity. Claude’s support documentation says usage limits determine how long a person can work with Claude Code before a reset, and usage across Claude product surfaces counts toward the same plan limit. 30

Developers should therefore treat multiple active threads as a way to organize and accelerate independent work—not as a way to bypass plan allowances. The supplied reporting describes a beta cap of 200 newly created threads per day and notes that background threads can affect usage, but Anthropic’s available help material does not establish a detailed public accounting formula for the Projects beta. 1

The redesigned Projects workflow is cloud-based at launch. That means it is best suited to repositories and files available to the cloud sessions; it does not bring the persistent multi-thread model to local-only resources or internal networks. Claude Code itself can run locally in a terminal, but local execution for this particular Projects workflow had not been released at the time of the beta rollout. 2614

Who can use it now?

The staged beta began with select Pro and Max subscribers using cloud sessions in Claude Code. Anthropic said the rollout would expand to more eligible Pro and Max users, followed later by broader Claude access and Team and Enterprise availability. Existing Projects users may remain on the earlier experience until their accounts are upgraded. 6810

No firm date for local support or the wider Team and Enterprise rollout is established in the available sources.

Why the redesign matters

The product shift is from a coding assistant that handles one session at a time to an orchestration layer for ongoing development. The coordinator maintains the project-level thread of work; cloud threads perform isolated tasks in parallel; shared memory and the Library carry context forward.

That positions Claude Code Projects around a difficult part of agentic software development: not merely generating code, but keeping multiple pieces of work organized over time. The limits remain important: developers still need to supervise access, review pull requests, resolve conflicting changes, and manage usage. But for work that naturally separates into parallel tasks, the redesigned Projects model offers a more durable workflow than repeatedly starting from scratch in disconnected chats. 58