Anthropic adds proactive coordinator and parallel workers to Claude Code Projects

Anthropic has rolled out a suite of new tools for Claude Code Projects that let developers run several AI workers in parallel without constant supervision. The upgrade matters because it turns a single‑threaded coding assistant into a coordinated team of autonomous agents, potentially shaving hours off large‑scale refactoring or performance‑tuning tasks.

What new capabilities does Claude Code Projects now offer?

The update introduces five headline features, according to AlphaSignal. A proactive coordinator now chats directly in the project’s main window, posting status updates and answering questions as threads progress. Individual threads can be shared via read‑only links that work for any signed‑in Claude user, while threads that have accessed a local computer remain private to protect local files and shell output. Reminders let a thread hide itself until a scheduled time, after which it resurfaces in the thread list and notifies the coordinator. Two slash commands – /btw and /fork – open a side thread for tangential exploration or branch the current thread into a new one, respectively. Finally, the threads panel now surfaces recent, updated, and pinned threads at the top and can display thumbnail previews of generated artifacts such as diagrams or documents.

How does the proactive coordinator change the development workflow?

Before the upgrade, developers had to constantly scan the thread list for sessions awaiting input, which created a “watch‑the‑workers” bottleneck. The coordinator now pushes pending decisions, completed tasks and blocked work into the main chat, turning the project into a single conversational hub. When a thread finishes a step, the coordinator reports the result and may ask follow‑up questions, allowing the developer to focus on higher‑level decisions rather than low‑level monitoring. This hands‑free approach reduces context‑switching and keeps the overall project narrative coherent.

What limits and pricing apply to the parallel workers?

The features are in public beta for Pro and Max subscribers who joined the waitlist on September 17; Team and Enterprise plans are excluded. There is no extra fee for creating projects, but each worker consumes the subscriber’s existing Claude Code allowance. Accounts are capped at 200 new threads per day across all projects. When a thread hits its usage limit, it automatically pauses and resumes once capacity is available. Usage is shown separately in the usage controls, and developers can pick the model and effort level for both the coordinator and the workers. New projects default to the Opus model for both roles, with a low‑effort setting for the coordinator and a high‑effort setting for workers. Running several high‑effort workers simultaneously can exhaust plan limits faster than a single Claude Code session.

Which development tasks benefit most from parallel AI workers?

Anthropic’s own examples focus on work that can be split into independent units and later reconciled through Git branches and pull requests. Performance profiling can assign separate threads to benchmark different endpoints, test optimizations, and open pull requests for the fastest results. Cross‑repository migrations can spin up a thread per codebase (API, web, mobile) and have each report the required merge order. Batch changes such as large‑scale refactors, dependency upgrades, test repairs, or bug‑triage can be distributed across many parallel sessions, letting the coordinator aggregate progress and surface merge conflicts when branches converge. Tasks with strict dependencies still need manual sequencing, but the parallel model shines when many small, independent changes are needed.

How can Grafy users experiment with similar parallel coding workflows?

While Grafy does not yet offer a built‑in multi‑worker coordinator like Claude Code Projects, the Grafy Code studio lets you launch separate AI‑driven coding sessions and keep them organized with the Graph view. You can mimic the /fork command by creating a new branch in the Graph, and you can use the Chat app to ask a “coordinator” bot to summarize status across multiple sessions. For developers who want to test the concept without a dedicated Anthropic account, start a project in Grafy Code, open several tabs with different prompts (e.g., one for profiling, another for refactoring), and use the Graph panel to link their outputs. The same read‑only sharing principle applies: Grafy’s public links let collaborators view a specific session without editing rights. This hands‑free, parallel approach can be trialled today without waiting for a beta rollout.

---

Source: https://alphasignal.ai/news/anthropic-upgrades-claude-code-projects-to-run-parallel-ai-workers-hands-free

Frequently asked questions

Does the proactive coordinator eliminate the need to monitor individual threads?

Yes. The coordinator now posts updates and answers questions directly in the main project chat, so developers no longer have to scan a thread list for waiting sessions.

Are there any extra costs for using parallel workers in Claude Code Projects?

No separate fee is charged, but each worker draws from the subscriber’s existing Claude Code allowance, and the platform caps accounts at 200 new threads per day.

Can Grafy Code run multiple AI workers in the same way Claude Code Projects does?

Grafy Code does not have an automatic coordinator, but you can run several independent coding sessions, organize them with the Graph view, and use Chat to consolidate status, achieving a similar parallel workflow.

---

Source: Hacker News — front page — Anthropic Upgrades Claude Code Projects to Run Parallel AI Workers Hands-Free