Claude Code Relaunches Projects for Multi-Agent AI Workflows in Beta as Cloud Threads Coordinate Tasks
Updated
Updated · The Verge · Sep 17
Claude Code Relaunches Projects for Multi-Agent AI Workflows in Beta as Cloud Threads Coordinate Tasks
3 articles · Updated · The Verge · Sep 17
Summary
Anthropic has relaunched Claude Code Projects in beta, letting select Claude Pro and Max subscribers run multiple coordinated agents inside one workspace with shared memory, goals, files and artifacts.
Each project uses parallel “threads” — separate Claude Code cloud sessions on their own repo branches — while a coordinator assigns work and users can manage threads individually or through a main project chat.
If threads touch the same code, conflicts are handled like standard pull-request merges, and each thread can further break work into subagents, loops and workflows to speed larger assignments.
Cloud-based threads are available at launch, with Anthropic saying support for local tools and code is coming very soon as access later expands to Pro, Max, Team, Enterprise, Cowork and regular Claude chats.
How will Anthropic's new autonomous AI swarms prevent catastrophic code overwrites when multiple agents edit the same files simultaneously?
Could Claude's context firewalls actually blind the coordinating AI, causing it to miss critical flaws hidden within the subagents' summaries?
As AI agents become independent co-workers, who is ultimately responsible when an autonomous subagent accidentally deploys a critical security vulnerability?