Anthropic Revamps Claude Code Projects to Orchestrate Multiple AI Agents in the Cloud

Anthropic has redesigned Claude Code’s Projects feature to let users run several AI agents inside a single workspace, with each agent handling a separate task while sharing goals, files and memory. The upgrade, available in beta starting today for a limited group of subscribers, matters because it pushes Claude Code closer to a coordinated team of agents rather than a single coding assistant.

The new version is aimed at developers who want to split a large software assignment into smaller pieces without losing oversight. Anthropic says each project can now manage multiple cloud-based threads at once, with a coordinator overseeing the work and resolving overlaps much like pull-request conflicts in a normal development workflow.

What Anthropic changed in Claude Code Projects

The biggest change is that Projects can now supervise multiple Claude Code agents at the same time instead of treating each task more like a one-off conversation. Anthropic’s updated setup gives every project a shared context layer that includes memory, objectives and a common library of files and outputs.

In practical terms, that means one user can ask a project to tackle a broad coding effort, then let separate agents work on different branches in parallel. Instead of waiting for one model to finish before starting the next, the project can distribute work across several cloud sessions and bring the results back together under one umbrella.

How the multi-agent system works

Anthropic says each thread inside a project runs as its own Claude Code cloud session, with a dedicated branch and copy of the repository. The project coordinator keeps those threads aligned, while any changes that touch the same code are handled through normal merge-conflict resolution.

That structure mirrors a common software team workflow: parallel work streams, shared direction, and a final integration step. Anthropic also says that each thread can go further and break assigned work into subagents, loops and workflows if the task is too large for a single pass.

Anthropic describes the arrangement as a set of cloud sessions operating on separate branches, with coordination in place to keep tasks organized and merge conflicts resolved in the same way developers already handle overlapping changes.

Why this matters for developers

The update is important because coding assistants are increasingly being judged not just on code generation, but on whether they can operate like a useful part of an engineering team. A single chatbot can draft snippets, explain bugs or refactor functions, but a multi-agent project system can divide and conquer larger tasks with less manual hand-holding.

For software teams, that could reduce the friction of assigning one model to investigate a bug, another to draft a fix and a third to verify how the changes affect surrounding code. It also creates a more natural fit for complex repositories, where many small dependencies make isolated prompting inefficient.

How does this compare with other agent tools?

Claude Code’s Projects feature now resembles a broader wave of software tools that are trying to coordinate fleets of AI agents rather than isolate one assistant per prompt. Similar approaches have emerged in products that manage groups of agents under a supervisor model, where each agent tackles a defined slice of work and reports back into a shared workspace.

The direction of travel is easy to see: coding assistants are evolving from reactive helpers into orchestrated systems that can plan, delegate and reconcile output. That shift matters because the real bottleneck in software work is often not writing the first draft of code, but organizing the many supporting steps needed to produce something reliable and maintainable.

Why the coordinator model is gaining traction

The coordinator model solves a practical problem. Large engineering tasks often involve research, implementation, testing, documentation and cleanup. If one model has to do everything sequentially, it can become slow or lose context. If a project can assign parts of the assignment to multiple agents, the workflow becomes more parallel and potentially more efficient.

It also makes AI easier to supervise. Rather than asking a model to improvise through a sprawling task in one long exchange, a project coordinator can keep a tighter grip on scope, track what each thread is doing and pull the results together into a single deliverable.

Who can use the new Claude Code Projects beta?

The beta is initially limited to select Claude Pro and Max subscribers. Anthropic says the rollout will broaden later to additional paid tiers, including Pro, Max, Team and Enterprise users, and then expand further to include Cowork and regular Claude chats.

At launch, the new threads operate in the cloud only. Anthropic says support for local tools and code integration is coming very soon, which suggests the company wants the system to work more seamlessly across both hosted and local development environments.

What users can do right now

• Launch a project with multiple AI threads.
• Assign separate coding tasks in parallel.
• Monitor each thread individually.
• Use the main project chat to oversee and update the work.
• Resolve overlapping code changes through standard merge-conflict handling.

Why Anthropic is pushing Claude Code in this direction

Anthropic has been steadily positioning Claude as a serious tool for work, especially for software development. The company’s latest move suggests it sees AI coding as more than autocomplete or chat-based assistance. It is aiming at a fuller workflow layer where the model can help organize entire tasks, not just produce fragments of code.

This is also part of a larger competitive race across the AI industry. Coding products are under pressure to differentiate themselves with better project management, stronger context handling and more reliable task execution. A tool that can coordinate a team of agents has a clear pitch: less prompt juggling, more structured delivery.

That said, parallel AI work also introduces complexity. More threads can mean more chances for inconsistent output, conflicting edits or workflow sprawl if the coordinator is not effective. The usefulness of the feature will likely depend on how well Anthropic balances automation with visibility and control.

Potential benefits for teams

For engineering teams, the feature could be especially appealing if it reduces time spent on routine delegation. A lead developer could imagine assigning one agent to investigate a crash, another to update a related component and a third to draft internal documentation, all within the same project structure.

For solo developers, the same system could serve as a fast way to simulate a small team. Instead of acting as the only coordinator, the user can let Claude Code do more of the task splitting while keeping a central view of progress.

What the launch means for the broader AI coding race

The release shows how quickly coding assistants are becoming more operational and less conversational. The next phase of AI coding tools is not just about answering questions about code; it is about creating systems that can manage context, split work and produce integrated outcomes.

That evolution could change the expectations developers have of AI. If agents can reliably coordinate across branches and preserve shared project memory, they may become useful for work that previously required too much oversight to delegate. But if they struggle with consistency or conflict management, the added complexity may become a burden instead of a benefit.

For now, Anthropic’s answer is to give Claude Code a more structured backbone and let users test whether multiple agents can work together more productively than one assistant operating alone. The beta launch is an early look at how that vision might function in a real development environment.

Timeline of the Claude Code Projects update

What to watch next

The key question now is whether developers will trust multi-agent orchestration enough to let it handle meaningful chunks of work. If the system performs well, Anthropic could turn Claude Code into a stronger all-in-one environment for planning, coding and review. If not, users may treat it as a niche feature better suited to experimentation than production.

Either way, the upgrade signals that the future of AI coding tools is becoming more collaborative and more automated. Anthropic is betting that users want an assistant that can act less like a single helper and more like a coordinated team.