Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarHacker News AI→ original

Anthropic turns Claude Code into a coordinator: parallel threads open PRs on their own

Until now, large tasks in Claude Code had to be split across sessions by hand, and the results stitched back together by hand. Now you set a goal in a project, for example cutting p75 checkout latency, and the coordinator hands out work to parallel threads, checks the output, and assembles the result. Each thread is a cloud session on its own branch, and overlaps surface as ordinary merge conflicts. The practical takeaway: long, multi-step tasks can be handed off whole, and you only need to review the PRs.

AI-processed from Hacker News AI; edited by Hamidun News
Anthropic turns Claude Code into a coordinator: parallel threads open PRs on their own
Source: Hacker News AI. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceHacker News AI ↗
Editor-in-Chief
Betainitially for some Pro and Max subscribers with cloud sessions
Key metric #2Each thread is a full session, so limits run out faster
Key metric #3Shared memory remembers team decisions, such as moving the release to Friday

If you have tried running a large task through several Claude Code sessions, you know how it goes: you cut the work into pieces yourself, hand them out, keep track of context passing between them, and then assemble the result by hand. Anthropic has decided to remove that routine and has updated projects in Claude Code. Now you describe what needs to be done, and Claude takes over the planning, distribution, and assembly.

How it works: threads and the coordinator

There are two roles at the core. Threads carry out the work, and the coordinator directs them. When you create a project, you choose a goal and a repository or a set of documents, then configure the cloud environment, connectors, plugins, instructions, and model. Claude itself suggests tasks that can be picked up right away.

Anthropic offers two examples. In the first, you set the goal of reducing p75 checkout latency in an app, and ask Claude to profile each endpoint, test optimizations, and open pull requests in parallel threads. In the second, you connect the API, web, and mobile client repositories and set the goal of removing the deprecated v1 endpoint. Claude opens one thread per repository, rewrites the calls, runs the tests, opens PRs, and suggests which of them should be merged first.

Technically, each thread is a cloud Claude Code session with its own branch and copy of the repository. If two threads touch the same code, the conflict is resolved the same way as in regular team work: through a merge conflict in the PR. In addition, a thread can split its assignment into parts on its own, using subagents, loops, and workflows, so that a large job finishes faster.

You can follow progress in the project's main chat or open a separate thread and edit details. The company advises setting tasks the way you would brief a chief of staff: the coordinator will route the work to a new or an existing thread. You can even manage the project from a phone, and it keeps working after you step away from the computer.

Memory and library instead of complex prompts

The second key part is shared memory. Each thread writes facts into it and pulls context from it. Anthropic gives examples: Claude can remember that the release was moved to Friday, why an export was removed, or who needs to sign off on changes in the billing service. This reduces the need to craft elaborate prompts for every new task.

Claude also remembers its working and communication style. You can ask it to report more or less often, to open new threads more or less frequently, and to make updates shorter or more detailed. Alongside memory, a library has appeared: it collects the files you add and the artifacts Claude creates, so new work builds on past work.

Access, limits, and what is unknown

This is a beta. The updated projects are open to selected Claude Pro and Max subscribers who use cloud sessions in Claude Code and who do not have older projects in the web or desktop app. Over the course of a week, access will expand to other Claude Code users on these plans, and then it will reach all of Claude and the Team and Enterprise plans. Existing projects on Pro and Max will continue to work as before and will be updated as the rollout reaches chat and Cowork.

There is also an obvious cost. Each thread is a full Claude Code session, so several parallel threads burn through limits faster. Anthropic lets you monitor usage by project and choose the model and effort level separately for the coordinator chat and for the working threads. For now, threads run only in the cloud. The company promises that running them on your own machine, alongside local tools and behind a corporate network, will arrive "very soon", but it gives no timeline.

What this means for developers

For teams, the shift is quite practical. Tasks such as migrating API calls across several repositories, profiling and optimization, or closing a long list of similar tickets map well onto parallel threads. Your role shifts toward setting goals and reviewing. The bottleneck becomes PR review: the more threads, the more code someone has to read and merge in the right order. It is worth setting review rules and limits in advance, before handing large goals to a project.

Story

Key Questions Answered

How it works: threads and the coordinator?

There are two roles at the core. Threads carry out the work, and the coordinator directs them. When you create a project, you choose a goal and a repository or a set of documents, then configure the cloud environment, connectors, plugins, instructions, and model. Claude itself suggests tasks that can be picked up right away. Anthropic offers two examples.

What are the key details on Memory and library instead of complex prompts?

The second key part is shared memory. Each thread writes facts into it and pulls context from it. Anthropic gives examples: Claude can remember that the release was moved to Friday, why an export was removed, or who needs to sign off on changes in the billing service. This reduces the need to craft elaborate prompts for every new task. Claude also remembers its working and communication style.

What are the key details on Access, limits, and what is unknown?

This is a beta. The updated projects are open to selected Claude Pro and Max subscribers who use cloud sessions in Claude Code and who do not have older projects in the web or desktop app. Over the course of a week, access will expand to other Claude Code users on these plans, and then it will reach all of Claude and the Team and Enterprise plans. Existing projects on Pro and Max will continue to work as before and will be updated as the rollout reaches chat and Cowork. There is also an obvious cost.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…