Claude Code relaunches Projects to manage multiple AI agents in the cloud
Sep 17, 2026, 11:58 AM · The Verge
Anthropic’s Claude Code Projects beta runs parallel cloud threads under a coordinator—shared memory, goals, and repo copies—with merge-conflict resolution, starting for select Pro/Max users before wider rollout.
Why it matters
Single-agent coding chats don’t match how big software work actually forks. Projects put multiple Claude Code cloud sessions on one roof, each on its own branch, with a coordinator steering and subagents splitting labor.
Users can poke individual threads or steer from the main project chat. Local tools are promised “very soon.” Beta opens today for select Claude Pro and Max subscribers, then Pro, Max, Team, Enterprise, plus Cowork and regular Claude later.
Useful for real repos. Multiplies the blast radius when agents misalign—exactly the week everyone is arguing about swarms.
From the desk
We’re interested because this is the productization of agent teams, not a research demo.
Each thread is a cloud Claude Code session with its own repo copy; overlap becomes a normal PR merge conflict. Coordinator plus subagents, loops, and workflows is how you finish large assignments faster—and how you get correlated failures if the shared memory goes weird. Grok Bot–class multi-agent managers already exist; Anthropic folding it into Claude Code puts the pattern in front of paying developers.
Useful AI for software teams looks a lot like this when permissions and review stay tight. The downside in a misalignment-news week: more autonomous parallel writers in cloud repos without mature oversight. I’m watching the local-tools launch and whether Anthropic publishes project-level audit logs as a default, not an enterprise upsell.
Context
The Verge, September 17, 2026. Anthropic Claude Code Projects relaunch in beta.
Who feels it
- Developers
- Parallel agent threads on real branches could cut cycle time—if humans still own merges.
- Security and platform teams
- Cloud agent fleets with shared memory raise secrets, exfiltration, and supply-chain risks.
- Anthropic competitors
- Pressure to match multi-agent project orchestration in coding products.
What to watch
- General availability beyond select Pro/Max.
- Local tool and code support shipping.
- Incident reports involving coordinated Claude Code threads.