On November 13–14, Anthropic updated the coordinator in Claude Code Projects to proactively report status, pending decisions, and blockers. This was not a new model launch, but it brought multi-task collaboration closer to a real product for the first time. The trade-off: every parallel thread may continue consuming subscription quota.
What this is
In September, Claude Code Projects changed projects from file directories into continuously running sessions: after users provide a goal and code repository, the coordinator breaks the work into tasks, with each thread modifying code on a separate branch, submitting pull requests, and running tests. After the November update, the coordinator no longer waits for users to check each task one by one. It directly pushes progress updates, completed work, and items requiring decisions.
Also new to projects are /btw side questions, /fork thread forking, thread reminders, and read-only sharing. Users can monitor multiple AI tasks as they would a group chat, without having to serve as the “foreman” themselves.
Industry view
We believe this update matters more than simply adding commands: product competition is shifting from “who can generate answers” to “who can manage a group of continuously working AIs.” Cursor’s background tasks, OpenAI’s Codex pricing model, and Claude Code Projects are moving scheduling, tracking, and review work—previously handled by developers—into the platform.
The risks are equally clear. Each worker—an AI executing a subtask—is a full session and defaults to a higher reasoning setting, so token costs are likely to grow roughly in line with the number of threads. Anthropic provides project usage tracking and model-level adjustments, but it does not decide whether a task is worth the expense. Team and enterprise editions are not yet available, and cloud threads cannot yet directly access internal code and tools, limiting some enterprise use cases.
Impact on regular people
For enterprise IT: AI coding may evolve from a one-off assistant into a continuous project, but permission isolation, spending caps, code review, and internal network access remain prerequisites for deployment—especially to prevent multiple threads from modifying critical systems simultaneously.
For individual professionals: The immediate change is to how software teams work. More developers will need to decide when to split tasks, review outputs, and control costs—not merely write better prompts.
For the consumer market: There will be no immediate change to the average consumer experience. But if background agents can reliably complete tasks across applications, digital assistants may finally move beyond answering questions to handling tasks on users’ behalf.