Anthropic updated Claude Code on September 17, 2026, with a redesigned Projects experience that lets one coordinator direct multiple cloud-based AI agents working in parallel. Instead of a single chat session, developers get a shared workspace where agents draw on common memory, goals, files, and artifacts.
TL;DR: Anthropic relaunched Projects in Claude Code on September 17, 2026, turning the feature into a shared workspace where a coordinator directs parallel cloud-based AI agents. Agents draw on shared memory, goals, files, and artifacts, with completed work merged as pull requests — and sessions keep running after you close your laptop.
What Is the New Projects Feature in Claude Code?
Projects in Claude Code are now shared workspaces where multiple AI agents coordinate coding tasks in parallel, according to The Verge. The feature launched in beta on September 17, 2026, rolling out in stages and starting with Claude Code users on Pro and Max plans.
The redesign changes the basic model of working with an AI assistant. Previously, a developer worked with one agent in one conversation. Now a Project acts as a container that holds context, goals, files, and artifacts, and lets several agents operate inside it at the same time.
Existing projects keep working as they do today, Anthropic notes in its Help Center documentation. That matters for teams who already use Projects to organize documents and conversations — nothing breaks during the staged rollout.
The local version is the catch. ZDNET reports that developers working locally will have to wait, since the parallel orchestration initially runs on Anthropic’s cloud infrastructure rather than on your machine. The workspace model also fits how enterprises actually build software: as VentureBeat observes, a business application rarely has a discrete endpoint — work runs long, spans many tasks, and needs continuity that a single chat thread cannot provide.
How Does the Coordinator Manage Parallel Agent Threads?
A coordinator sits at the center of the new Projects experience and directs parallel agent threads, Unite.AI reports. When you describe a large task — refactoring a module, building a feature, cleaning up tests — the coordinator splits that work into separate streams and assigns each stream to an agent running in the cloud.
The division happens dynamically. The coordinator decides how many threads a task needs, what each thread should focus on, and in what order results should come back. Developers don’t manually spawn agents or copy context between windows; the coordinator handles delegation inside the Project.
Each agent thread works on its own slice of the problem. Because the threads run in the cloud rather than on your laptop, they can execute simultaneously without competing for your local resources. Anthropic’s documentation describes the coordinator as the component that keeps these threads aligned with the project’s shared goals.
What does this look like in practice? You write one instruction, and behind the scenes several agents make progress on different parts of it. The coordinator monitors each thread, keeps agents from duplicating work, and collects outcomes as they finish. In my opinion, this is the most interesting shift: the developer moves from babysitting a single agent to supervising a small team of them.
Why Does Shared Memory Matter for Agent Teams?
Shared memory is what keeps parallel agents from becoming parallel strangers. UA.NEWS summarizes the update as enabling multiple cloud-based AI agents to coordinate tasks through shared memory, goals, files, and artifacts — four shared resources that every thread in a Project can read.
Without shared context, each agent would need the same background information repeated in its own prompt. With it, an agent that discovers a breaking change in one thread can leave that knowledge where other threads will find it. The coordinator also relies on this shared layer when it decides how to split and sequence work.
The shared resources in a Project include:
- Memory — accumulated knowledge from previous sessions and completed tasks
- Goals — the project-level objectives every thread works toward
- Files — the codebase and documents agents operate on
- Artifacts — outputs produced by agents, visible to the whole team
- Conversation history — the “always-on” thread VentureBeat describes, which remembers and delegates long-running dev work
This design addresses a real weakness of single-agent workflows: context loss between sessions. Because the Project remembers, a task you start on Monday doesn’t evaporate when you reopen Claude Code on Wednesday. The workspace persists, and so does everything the agents learned while working in it.
Do Sessions Keep Running When You Close Your Laptop?
Yes. MarkTechPost’s headline captures the key operational detail: Anthropic launched Claude Code Projects in beta with parallel cloud sessions that keep running after you close your laptop. Because agent threads execute on Anthropic’s infrastructure, your machine is just a window into the work — not the engine behind it.
This is a meaningful departure from the classic local CLI experience, where closing the terminal ends the session and any long-running task dies with it. With Projects, a migration job or a lengthy test suite keeps churning in the cloud while you’re offline, in a meeting, or asleep.
The always-on quality changes how developers schedule work. Long tasks no longer need to be babysat or split artificially into chunks that fit a workday. You can hand the Project something substantial, close the lid, and review the results later.
There’s a trade-off, though. ZDNET notes that local developers must wait — meaning developers who rely on running Claude Code against local environments won’t get the same parallel, persistent behavior until Anthropic extends the feature. For now, the always-on sessions are a property of the cloud-based beta rollout, available to Pro and Max subscribers in stages.
How Do Parallel Code Streams Get Merged Back Together?
Finished work from parallel agent threads is merged as pull requests, according to wwwWhatsNew, which describes the update as coordinating multiple agents working simultaneously through code streams, shared memory, and pull-request merges.
The pull-request model is familiar ground for any team using Git. Each agent stream produces its changes as a discrete, reviewable unit. That gives developers a checkpoint: before code lands in the main branch, it passes through the standard review and integration machinery teams already trust.
The flow works roughly like this:
- You give the coordinator a task.
- The coordinator splits it into parallel streams.
- Each stream’s agent works in the cloud against the project.
- Completed streams come back as pull requests.
- You review, approve, or request changes before merging.
This structure keeps parallel agents from stomping on each other’s edits. Instead of several agents writing to the same working tree at once — a recipe for conflicts — each stream’s output is packaged separately and reconciled at merge time through the normal version-control process.
For enterprise teams, the reviewability is the point. VentureBeat frames the feature as suited to long-running business applications precisely because the work stream never really ends; discrete pull requests make an open-ended, multi-agent effort auditable one merge at a time.
Who Can Access the Beta, and When Will It Roll Out Fully?
The redesigned Projects experience is available today in beta, and Anthropic is rolling it out in stages — starting with Claude Code users. According to the Claude Help Center, the rollout begins with Claude Code before expanding to other Claude surfaces. Reports from MarkTechPost indicate the beta covers Pro and Max plans, meaning individual developers and power users get first access. Enterprise teams will follow as the staged rollout continues. How long will the beta phase last? Anthropic has not published a fixed end date, which suggests the company wants to validate the coordinator’s behavior on real workloads before a general release. Existing users don’t need to migrate anything: per the Help Center, existing projects keep working exactly as they do today, and the new experience appears alongside them rather than replacing them. For developers on lower-tier plans, the wait may be the price of a stable rollout — Anthropic appears to be prioritizing its most intensive Claude Code users first, since they generate the parallel-agent workloads the new architecture was designed to handle. Check your account’s feature announcements if you want early access.
Why Must Local Developers Wait for the Redesign?
Developers who run Claude Code entirely on their local machines won’t get the new Projects experience in the first wave — and there’s a structural reason for that. As ZDNET reports, the revised Projects adds AI orchestration, but local developers must wait. The coordinator model depends on cloud infrastructure: parallel agent threads, shared memory, and always-on sessions that continue after you close your laptop all run on Anthropic’s servers. A purely local setup can’t guarantee those properties. Does that mean local workflows are second-class? Not exactly. The current local Projects behavior remains fully supported, so nothing breaks while developers wait. But the headline features — a coordinator splitting work into parallel streams, agents drawing on shared goals, files, and artifacts — are architecturally cloud-native. The Verge’s coverage frames the relaunch explicitly as managing multiple AI agents in the cloud. For developers with strict data-residency requirements or limited connectivity, this is a genuine constraint to evaluate before adopting the new model. Anthropic hasn’t announced a timeline for local parity, so local-first teams should plan around the current behavior for now.
How Does This Fit Into Anthropic’s Broader Agent Strategy?
The Projects relaunch is a clear signal of where Anthropic is heading: from single-turn assistants toward orchestrated, long-running agent systems. The update, announced September 17, 2026, per Unite.AI, introduces a coordinator that directs parallel agent threads drawing on shared memory. That’s not an incremental feature — it’s an architecture. Instead of one agent executing one task, Claude Code now operates as a small engineering team with a supervisor. Why does this matter for Anthropic’s position in the market? Because coding agents are the most commercially proven AI use case today, and coordination is the next battleground. Competitors are racing toward multi-agent workflows, and Anthropic’s answer is to make the orchestration layer invisible: developers describe a goal, and the coordinator handles decomposition, delegation, and merging. VentureBeat describes the result as an ‘always-on’ conversation that remembers and delegates long-running dev work. Shared memory across agents also creates stickiness — the longer a project lives in Claude, the more context accumulates, and the harder it becomes to move elsewhere.
What Do Enterprises Gain From Always-On Development Work?
Enterprise software rarely has a discrete endpoint, and that’s exactly why always-on agent sessions resonate with larger organizations. As VentureBeat notes, a company’s digital storefront, website, content management system, or procurement platform is never ‘finished’ — it requires continuous iteration. Claude Code Projects matches that rhythm: cloud sessions keep running after you close your laptop, per MarkTechPost, so long refactors or migrations can progress overnight. What else do enterprises get? A concrete list:
- Parallel agent threads working on the same codebase simultaneously
- Shared memory, goals, files, and artifacts across all agents (UA.NEWS)
- A coordinator that delegates tasks and merges results
- Persistent project context that survives across sessions and team members
- Cloud execution that doesn’t depend on any single developer’s machine
- Reduced coordination overhead compared to manually managing separate agent sessions
The trade-off is real, though. Enterprises with compliance or data-residency policies must confirm that cloud-hosted agent sessions meet their requirements before relying on them for proprietary code.
Is Cloud-Based Agent Orchestration the Future of Coding Tools?
The industry trajectory points strongly toward yes — cloud-orchestrated agent teams are becoming the default model for serious development work. The evidence is converging from multiple directions. Anthropic’s September 2026 relaunch puts a coordinator over parallel cloud threads. Coverage from iTechPost and WERSM describes shared workspaces where multiple agents coordinate coding tasks in parallel, which was unthinkable in mainstream tools two years ago. Why is the cloud the natural home for this? Because orchestration needs persistent state, heavy compute, and availability that individual laptops can’t provide. Polish tech coverage captured the model concisely: one boss, many AI agents at once (SukcesAI). That’s the shape of things — developers becoming reviewers and directors rather than line-by-line coders. In my opinion, the local-first crowd will retain a niche for privacy-sensitive and offline work, but the productivity math of parallel agents with shared memory is hard to argue against. The open questions are governance and cost: who reviews agent merges, and what does running five agents overnight do to your bill?
Frequently Asked Questions
When did Anthropic launch the redesigned Claude Code Projects?
Anthropic announced the redesigned Projects experience on September 17, 2026, releasing it in beta with a coordinator that directs parallel agent threads drawing on shared memory, according to Unite.AI. The rollout began with Claude Code and is proceeding in stages.
How does the new Projects coordinator actually work?
The coordinator acts as a supervisor that splits a development goal into parallel cloud threads, delegates them to multiple agents, and merges their output. Agents coordinate through shared memory, goals, files, and artifacts, and results are fused like pull requests, per wwwhatnew’s report on the update.
Which plans get access to the Projects beta?
The beta is rolling out in stages starting with Claude Code, and MarkTechPost reports it is available on Pro and Max plans. Anthropic has not announced a firm general-availability date, so access depends on the staged rollout reaching your account.
Do existing Projects stop working after the update?
No. The Claude Help Center states explicitly that existing projects keep working as they do today. The new version of Projects is available in beta alongside them, so no migration is required.
Summary
- Anthropic relaunched Claude Code Projects on September 17, 2026, adding a coordinator that manages multiple AI agents working in parallel in the cloud.
- Agents share memory, goals, files, and artifacts, with merged results handled like pull requests — one supervisor, many workers.
- Cloud sessions are ‘always-on’ and continue after you close your laptop, a strong fit for enterprises running never-finished business applications.
- The beta rolls out in stages starting with Claude Code on Pro and Max plans; local developers must wait, and existing projects keep working unchanged.
- The main open questions are data governance for cloud-hosted sessions and the compute cost of running parallel agents.
If the Projects beta hasn’t reached your account yet, watch Anthropic’s announcements and evaluate how your current workflow would map onto parallel agent threads — the teams that plan their orchestration strategy early will adapt fastest.