Anthropic has completely reworked Projects in Claude Code. Now, within a single project, Claude can run multiple agents in parallel, assign them different parts of the work, monitor their execution, and finally collect the results together.
Anthropic has completely reworked Projects in Claude Code. Now, within a single project, Claude can run multiple agents in parallel, assign them different parts of the work, monitor their execution, and finally collect the results together.
This is reported by Anthropic.
The user needs to describe the overall goal of the project and connect the necessary repository or other context. After that, Claude will evaluate the task itself, break it into parts and launch separate threads for them - parallel work sessions. This is observed by a "coordinator" who organizes all the work. He monitors progress, distributes tasks and collects the results of the agents' work.
For example, a developer can ask Claude to speed up checkout in an application. Then one thread can check a separate endpoint, another can test optimization, and yet another can prepare a pull request. And if the project covers an API, a web version, and a mobile application, Claude can create a separate thread for each repository. And the user can see what is happening in each thread, write to agents separately, or manage everything through the main project chat.
Also, individual threads within a project will have shared memory. Previously, individual Claude sessions had to explain the context anew each time. Now, each thread can add new information to the shared memory and use what others have already collected.
However, because multiple Claude Code sessions are running simultaneously, they can quickly consume the user's available limits. Therefore, different models and levels of computational effort can be selected for the coordinator and individual threads. For now, threads are only running in the cloud. Anthropic says that support for local tools and code is planned to be added "very soon."
The updated Projects are being tested starting September 17th with select Claude Pro and Max users who have access to Claude Code cloud sessions. Anthropic plans to gradually open the feature to all Pro and Max users, as well as Team and Enterprise plans. Over time, the new Projects should also appear in regular Claude and Cowork chats.
Previously, dev.ua wrote about Agent View in Claude Code — an interface in which you can simultaneously launch and control multiple agent sessions. Now, Claude itself helps distribute work between individual agents and maintains a common context for them.
