Anthropic adds parallel AI workers and proactive coordination to Claude Code
Developers can now run multiple independent Claude Code sessions simultaneously to handle complex engineering tasks like migrations and batch refactors.
The short version
- A proactive coordinator now pushes status updates and answers questions directly in the project chat.
- Developers can share specific threads as read-only links, provided they do not have local computer access.
- New slash commands like /btw and /fork allow for branching work and handling tangents in coordinated projects.
- Parallel threads draw from the same Claude Code allowance, with a limit of 200 new threads per day.
- The update is currently available in public beta for Pro and Max subscribers on web, desktop, and mobile.
Why it matters
This shift moves AI coding from a linear chat to a managed workflow where one 'manager' AI oversees multiple 'worker' sessions. It allows for faster execution of large-scale projects by parallelizing tasks that would otherwise have to be done one by one.
From chat to project management
Anthropic has introduced significant updates to Claude Code Projects, transforming how developers interact with the AI for complex engineering. The core of this update is a proactive coordinator that manages multiple independent threads of work. Instead of a developer manually checking every session, the coordinator now pushes status updates and answers questions directly in the main project chat.
This system allows a project to function as a long-running conversation where the coordinator assigns specific tasks to different threads. Each thread operates as an independent Claude Code cloud session, complete with its own repository copy and unique Git branch. This structure allows for a more organized approach to software development than a single, linear chat window.
By automating the reporting of pending decisions and completed tasks, Anthropic aims to reduce the monitoring burden on the user. Developers no longer have to hunt through a list of active sessions to see what is blocked or finished. The coordinator brings the progress of all parallel workers into one central conversation for easier oversight.
Parallelism and branching logic
The new system is designed specifically for engineering work that can be broken down into independent units. For example, a developer can assign separate threads to profile different endpoints, test various optimizations, or handle bug triage across multiple files. When these parallel threads complete their work, the overlapping code changes appear as standard Git merge conflicts that the developer can then resolve.
To facilitate this, Anthropic added slash commands that were previously only in regular sessions. The /fork command allows a developer to branch a new thread from an existing one, while /btw opens a side thread for tangents. These tools help maintain a clean primary workflow while allowing the AI to explore different paths or side tasks without cluttering the main project goal.
This architecture is particularly useful for large-scale operations like cross-repository migrations. A developer can connect multiple repositories—such as API, web, and mobile codebases—and create a dedicated thread for each one. The coordinator then reports the required merge order, allowing the team to move through a complex migration with structured, parallel execution.
Sharing and navigation tools
Anthropic also updated the user interface to make navigating these large projects more intuitive. The threads panel now features a recents row at the top and includes thumbnails for generated artifacts like diagrams and documents. These previews can be controlled through specific view settings, making it easier to see the output of various parallel workers at a glance.
Collaboration features have been expanded through the introduction of read-only links. A developer can share a specific thread with any signed-in Claude user via a generated URL without exposing the entire project. This allows for targeted feedback or review of a specific task while keeping the rest of the project's context private.
There is a specific security restriction on these shared links regarding local data. Threads that have accessed a local computer are excluded from being shared as read-only links. This prevents the shared transcripts from accidentally exposing local file contents, shell outputs, or other sensitive data gathered during a session with local machine access.
Usage limits and availability
The new features are currently available in a public beta for Claude Pro and Max subscribers who joined the waitlist. These users can access the updates across the web, desktop clients, and the latest mobile app. However, the update is currently excluded from Team and Enterprise plans, which means those users will need to wait for a future rollout.
While there is no separate fee for using the Projects feature, every worker thread consumes the subscriber's existing Claude Code allowance. Anthropic limits accounts to 200 new threads per day across all projects to manage capacity. If a thread hits its usage limit, it will pause and automatically resume once more capacity becomes available.
Developers have some control over how these resources are spent by choosing the model and effort level for both the coordinator and worker threads. By default, new projects use Opus for both roles, with the coordinator set to low effort and the workers set to high effort. Because high-effort sessions run simultaneously, users should be aware that parallel work can exhaust plan limits faster than a single session.
What to watch
- Whether Team and Enterprise plans will receive the parallel worker updates.
- How developers handle frequent merge conflicts when many threads work on the same files.
- Updates on the 200 thread-per-day limit as demand for parallel AI work grows.
Try it
If you have a Pro or Max subscription, check your Claude Code Projects to see if you have access to the new coordinator and /fork commands.
Sources and links
- Officialclaude.com/form/projects claude.com
- Reportingalphasignal.ai: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by alphasignal.ai · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
