Anthropic just turned Claude Code into less of a lonely pair-programmer and more of a mini agent org chart. The Verge reports that the company relaunched Projects inside Claude Code so you can spin up multiple agents under one roof—shared memory, shared goals, a shared library of files and artifacts—while a coordinator keeps the swarm from stepping on itself.
Each project gets “threads” that run different tasks in parallel. Under the hood, every thread is its own Claude Code cloud session working on its own branch and copy of the repo. If two threads touch the same code, overlap resolves as a merge conflict—the same messy, familiar PR ritual humans already know. Threads can also fan out further with subagents, loops, and workflows when a job is too big for one pass.
One chat to rule them (mostly)
You can poke individual threads or steer the whole party from the main project chat. That dual interface is the point: babysit the specialist when it drifts, or stay at the coordinator layer when the work is humming. At launch, threads run in the cloud; Anthropic says local tools and local code support are coming “very soon.”
Availability is staged like every serious model feature these days. The updated Projects experience is in beta for select Claude Pro and Max subscribers first, with a wider rollout planned for all Pro, Max, Team, and Enterprise seats—and later Cowork plus regular Claude chats. If you’re not on the early list, you’re watching the agent-team arms race from the bleachers for a bit.
Why this matters in the agent wars
The Verge correctly notes the rhyme with tools like Grok Bot and other multi-agent managers: once coding agents can parallelize, the bottleneck shifts from “can the model write a function?” to “can you orchestrate five of them without creating a spaghetti PR?” Anthropic’s answer is deliberately boring in the best way—Git branches, merge conflicts, coordinator oversight—rather than inventing a brand-new collaboration metaphor nobody wants to learn.
That design also telegraphs Anthropic’s bet: Claude Code isn’t just autocomplete with vibes. It’s an agent runtime that treats repositories as the shared world model. Shared memory and artifacts mean threads aren’t amnesiac strangers; they’re coworkers with a shared Notion page and a slightly chaotic standup.
Limits you’ll feel on day one
Cloud-only threads at launch mean latency, cost, and repo-sync friction still matter. Merge-conflict resolution is honest about parallel agents colliding—but honesty doesn’t make conflict hell free. And gated Pro/Max beta access means the feature will look magical in demos long before it feels ubiquitous in Team/Enterprise workflows.
Still, the direction is clear: single-agent chat was the tutorial level. Multi-thread Projects are the midgame for anyone shipping non-trivial code with AI in the loop.
Geeknewz take
Anthropic didn’t invent the multi-agent dream—it productized the boring parts that make swarms usable: branches, a coordinator, shared context, and merge conflicts as a feature instead of a failure mode. If local execution lands quickly, Claude Code Projects become a serious default for teams already living in Git. If it stays cloud-gated and invite-only too long, rivals will ship “good enough” orchestration and eat the casual market.
Either way, “one Claude, one chat” is officially legacy cosplay. Welcome to manager mode.
Source: The Verge — Claude Code relaunches Projects to manage multiple AI agents in the cloud (Stevie Bonifield, Sep 17, 2026).