On September 17, 2026, Anthropic launched a redesigned Projects experience in beta for Claude Code, turning what was previously a folder-like workspace into a system where a coordinator agent delegates work to parallel threads that run as independent cloud sessions. The update is available starting today to select Claude Pro and Max subscribers who use cloud sessions in Claude Code and have no existing projects on the web or desktop; access will expand to more Pro and Max users over the coming week, with Team, Enterprise, Cowork, and regular chat to follow.
Each project now centers on a goal and a repository or context. The coordinator suggests work it can pick up immediately, spawns threads that operate on their own branches and copies of the repo, and merges overlapping changes through standard pull-request conflict resolution. Threads can further split work using subagents, loops, and workflows. A shared memory persists across threads so that decisions — such as a release date shift or a billing-service ownership note — carry forward without repeated prompting. A project library collects user-uploaded files and Claude-generated artifacts for reuse.
Confirmed
- Redesigned Projects are in beta in Claude Code as of September 17, 2026.
- Initial access: select Pro and Max subscribers using cloud sessions in Claude Code with no existing web/desktop projects.
- Rollout order: more Pro/Max users over the next week, then Team, Enterprise, Cowork, and regular chat.
- Architecture: one coordinator chat plus multiple threads; each thread is a Claude Code cloud session on its own branch and repo copy.
- Merge conflicts between threads resolved as standard PR conflicts.
- Shared memory and a project library (user files + Claude artifacts) persist across threads.
- Threads run in the cloud today; local execution alongside user tools and behind the user's network is coming "very soon."
- Usage draws from the subscriber's plan; running several threads at once consumes quota faster.
- Model and effort levels can be selected separately for the coordinator and worker threads.
Unknown
- Exact criteria for "select" Pro/Max subscribers in the initial beta wave.
- Whether the coordinator and worker threads can use different model families (e.g., Opus for coordinator, Sonnet for workers) or only effort tiers.
- Pricing or quota implications for heavy multi-thread workloads beyond "uses it faster."
- Timeline for local execution support beyond "very soon."
- Whether existing web/desktop projects on Pro/Max will be migrated automatically or require user action when the rollout reaches chat and Cowork.
- Independent benchmarks comparing multi-thread throughput to single-session Claude Code on representative codebase tasks.
Our take
Anthropic is betting that the bottleneck for AI-assisted engineering is no longer single-turn reasoning but coordination across long-running, multi-repo tasks. By making the coordinator a first-class agent that manages branches, memory, and merge conflicts, the redesign shifts Claude Code from a pair-programming tool toward a project-level orchestrator. The real test will be whether shared memory stays coherent as threads diverge and whether the usage economics make sense for teams that spin up dozens of parallel threads on large codebases.
Series: 1. Anthropic to Make Claude Code Auto Mode Default for Pro, Max, and Team on August 14 · 2. Vibe Coding Tools 2026: The Landscape Shifts Toward Production Readiness · 3. Anthropic Adds /design Command to Claude Code for In-Terminal UI Prototyping · 4. Anthropic extends 50% weekly limit boost for Claude Code through August 2026 · 5. OpenAI Open-Sources Codex Harness to Let Developers Embed Agent Loops in Their Own Products · 6. Anthropic redesigns Claude Projects as multi-agent coordinator in Claude Code · In-Terminal Coding
