Most AI coding tools still put the coordination burden on you. You break the work into chunks, run sessions sequentially, and reassemble the results yourself. Anthropic is taking a different approach. The company has announced a redesigned Projects experience inside Claude Code, currently in beta, where Claude doesn’t just execute tasks. It manages them.
What’s actually changed
The old Projects feature was essentially a folder with memory. The new version is closer to a project manager. You set a goal, connect your repos, and Claude scopes the request, spins up parallel threads, monitors outputs, and assembles a finished result. Each thread is a full Claude Code cloud session working on its own branch. The coordinator keeps everything organized, and overlapping changes resolve as standard merge conflicts.
The practical examples Anthropic gives are specific and worth paying attention to. Connect your API, web, and mobile repos, set a goal to retire a deprecated v1 endpoint, and Claude opens a thread per repo, migrates callers, runs tests, and tells you which PRs need to merge first. That’s not a chatbot. That’s closer to delegating to a junior engineering team.
Memory that actually persists
Every thread now contributes to a shared project memory. Claude can remember that the release moved to Friday, why a feature was dropped, or which service requires a check-in before you touch it. This is a direct response to one of the biggest complaints about long-running AI workflows: context evaporates between sessions, forcing constant re-prompting.
The model also adapts to your communication style over time. You can tell it how often to check in, how aggressively to open new threads, and how detailed its updates should be. There’s also a project library that collects files and artifacts produced across sessions, so new work builds on past results rather than starting cold.
Who gets access and what to watch
Beta access is rolling out first to Claude Pro and Max subscribers using cloud sessions in Claude Code without existing web or desktop projects. Team and Enterprise plan support, plus the full chat and Cowork integration, comes later. Local execution, so threads can run on your machine against your local tools and network, is listed as coming soon.
The usage implications are real. Running several full Claude Code sessions in parallel burns through limits faster than single-session work. Anthropic says you can monitor project-specific usage and tune the model and effort levels per thread, which is a necessary control given the cost exposure.
- Parallel threads per repo for multi-repo tasks
- Shared memory across all threads in a project
- Artifact and file library across sessions
- Usage controls per thread and coordinator
- Local execution support coming soon
The competitive framing here matters. GitHub Copilot Workspace and Cursor’s background agents are pushing in a similar direction, but neither has shipped a coordinator-plus-memory architecture quite like this at the project level. So the real question is execution: does the coordination actually hold up on complex, multi-repo work, or does it fragment the way most multi-agent pipelines still do? That’s what the beta will tell us.
