Claude Code Projects now coordinates multiple AI agents without requiring constant supervision

- Reading
- 4 min
- Tables
- 1
Anthropic updated Claude Code Projects so the coordinator agent proactively reports on the progress of each task, reducing the need for developers to keep checking multiple parallel sessions.
Anthropic updated Claude Code Projects with a set of changes that reduce the manual effort of tracking multiple AI sessions running at the same time, according to a report by AlphaSignal based on the company's documentation. The feature has existed since its public beta launch on September 17, for Pro and Max plan subscribers who joined the waitlist.
Each project in Claude Code works as a long-running conversation with a "coordinator" that distributes tasks among threads. Each thread runs as an independent Claude Code session in the cloud, with its own copy of the repository and a separate Git branch. Code conflicts appear as ordinary merge conflicts when the branches are combined.
The coordinator now speaks without being asked
Before, the developer had to manually monitor the thread list to find out which sessions were stuck waiting for a decision. Now the coordinator answers questions directly in the project chat and posts status updates as threads progress, bringing pending decisions, completed tasks, and blocked work into the main conversation.
For those who use Claude Code to automate technical tasks, this change addresses the biggest friction point of the multi-agent model: constant vigilance. Running five or six threads in parallel only pays off if the cost of monitoring drops along with the speed gain.
Sharing with a safety limit
Individual threads can now be shared via read-only links with anyone authenticated on Claude. To do this, simply open the thread menu, select Share, and send the generated URL, without exposing the rest of the project.
Threads that had access to the developer's local computer are excluded from this option. The restriction exists to prevent shared transcripts from exposing the contents of local files, shell output, or other data collected from the machine during the session.
Three commands to organize the workflow
Three new features handle what to do when a task isn't linear:
| Control | Behavior | Best use |
|---|---|---|
| Reminders | Hides a thread until a scheduled time, then returns it to the list and notifies the coordinator | Work that depends on later review, deployment, or external response |
/btw | Opens a separate thread for a tangent, preserving the goal and context of the current thread | Investigating a related issue without diverting ongoing work |
/fork | Creates a new thread from the state of the current thread | Testing an alternative implementation or a different debugging path |
The /btw and /fork commands already existed in regular Claude Code sessions; the news is that they are now available within the coordinated Projects workflow, applying the same branching logic to work across multiple threads.
Thread panel gains visual shortcuts
The thread panel now displays a row of recent, updated, and pinned items at the top, along with thumbnails of generated artifacts, such as documents and diagrams, controlled by the display settings. Anthropic also simplified project settings and access to local folders. The changes are already available on the web, in desktop clients, and in the latest version of the mobile app.
The limit that doesn't change: account quota
Projects has no separate billing, but each worker consumes the same Claude Code usage quota the subscriber already has on the Pro or Max plan. Anthropic limits accounts to 200 new threads per day across all projects combined; a thread that hits the usage limit pauses and resumes automatically once capacity becomes available again.
Projects usage appears separately in the consumption controls, and developers can choose the model and effort level for both the coordinator and the workers. New projects come with Opus by default for both roles, low effort for the coordinator and high effort for the workers. This matters in practice: several high-effort sessions running at the same time deplete the plan's quota faster than a single regular Claude Code session.
Team and Enterprise still don't support the feature, which leaves parallelism restricted to individual Pro and Max accounts. For teams that rely on a corporate plan, this means the multi-agent workflow remains unavailable for now, even if the rest of the team already uses Claude Code day to day.
Where parallelism pays off (and where it doesn't)
Anthropic itself describes three scenarios where splitting work into parallel threads pays off:
- Performance work: separate threads for profiling endpoints, testing optimizations, and opening pull requests, with a target like
p75checkout latency guiding the success criteria. - Cross-repository migrations: connecting API, web, and mobile repositories, creating one thread per codebase, and letting the coordinator report the necessary merge order.
- Batch changes: distributing refactors, dependency upgrades, test fixes, or bug-triage items across parallel sessions.
Tasks with rigid dependencies still require the developer to sequence pull requests and resolve conflicts manually; the coordinator doesn't replace that decision. Projects also remains limited to the Claude Code cloud environment, without support for Team and Enterprise, and small changes that fit within a single branch and context generally consume fewer resources in a regular session than inside a project with multiple workers.
Translated from the Brazilian Portuguese original · Read the original
Fake job offer hides malware in a Git repository inside a ZIP
A Russian developer documented how a supposed recruiter used Git hooks and filters to run malicious code with nothing more than a git status, without requiring the execution of any suspicious file.





