Back to News

Anthropic relaunches Claude Code Projects with multi-agent cloud coordination

#anthropic#claude-code#multi-agent#ai-development

Anthropic has redesigned Claude Code's Projects feature, now in beta, to manage multiple AI agents in the cloud. Users describe goals in one conversation, and a coordinator splits work across parallel threads, each running a cloud session with its own branch and repo copy, resolving overlaps as merge conflicts. Separately, Anthropic reported that as of August, Claude leads 26% of its internal AI R&D work, with over 90% at least AI-collaborative, and about 30,000 agents running at any time.

Coverage timeline

  1. The Verge AIStevie Bonifield

    The revamped projects feature in Claude Code allows users to run multiple agents under the same roof, with a shared memory, goals, and library of files and artifacts. Similar to Grok Bot and other tools that manage groups of AI agents, each project has "threads" running different tasks in parallel, with a "coordinator" directing everything: Under the hood, each thread is a Claude Code cloud session working on its own branch and copy of the repo. The coordinator keeps work organized, but if any threads work on the same code, the overlap is resolved as a merge conflict just like any other PR. Each thread can further split its delegated work … Read the full story at The Verge.

  2. Techmeme

    Claude : Anthropic redesigns Claude projects, letting users describe work in one conversation and have Claude manage it across parallel threads, starting in Claude Code — A new experience for Claude projects, now available in beta in Claude Code — https://claude.com/blog/projects- redesigned

  3. 量子位量子位

    刚刚,Claude Code大重构!内部3万Agent管理技术免费开放 Jay 2026-09-18 08:34:34 来源: 量子位 Git白学了??? Jay 发自 凹非寺 量子位 | 公众号 QbitAI 多人云端Coding的时代,来了。 刚刚,Claude Code的项目功能迎来全面重构,原生支持多Agent玩法。 用户下达目标,Claude在云端派生多条Git分支并行干活,写完直接提PR。 此外,A社还新增了共享记忆,保证Agent之间的上下文对齐。 可以说是彻底超进化了,Project不再只是个放提示词和文档的文件夹。 同一时间,在Anthropic披露的最新报告中,也已看到这项能力的上限: 他们内部研发平台,每天同时跑着 3万个AI Agent ,在云端疯狂写代码、调模型、跑测试。 是的,3万个。。。 结果就是,自家核心的AI研发工作,现在已经有整整 26%由Claude主导,来到L4级别。 半年前,这个数字还不到1% 。 一人带一个Agent军团,Claude Code大重构 最新亮相的Claude Code Projects重构版,彻底终结了把Projects当「静态资料夹」的时代。 以前的Projects功能,本质上就是个存放提示词和参考文档的文件夹,聊上几轮上下文就乱成一团。 这次重构直接把形态变成了「以对话为驱动的协作中心」,架构核心是一个名为 Coordinator (协调器)的大脑。 开发者不需要再手把手教AI每一步怎么走,只要在主对话框下达一个高阶目标,比如「把整个鉴权模块重构为OAuth 2.0并补齐自动化测试」。 主协调器接收到指令后,会自动把复杂的工程任务拆解为多个子任务,并调度分配给并行的 工作线程 。 而每一个工作线程,都是一个运行在云端独立环境里的Claude Code实例。 它们拥有完全隔离的代码副本,各自拉出独立的 Git分支 ,在后台同时推进不同模块的代码编写与单测。 活干完了,工作线程会自动在后台提交代码并生成PR。 协调器还会贴心地跟踪各个子任务的上下文依赖,提醒开发者先合并哪个分支、后测试哪个模块。 神了,Vibe Coding几个月,好不容易学会Git。 你现在告诉我,连Git也白学了??? 说回新版Project。 道理大家都懂,以前也不是没人干,但问题在于—— 多个Agent各写各的,其实际体验是不太好的

  4. 机器之心机器之心

    编辑|泽南 AI 距离递归的自我改进(RSI)还有多远? 这个问题是目前 AI 领域最热的话题之一,过去我们通常只能观察一些间接信号:大模型的编程能力跑分,Agent 连续工作的时间,开发人员用 AI 写代码的比例。 现在 Anthropic 给了你一个实际的数字:截至今年 8 月,在 Anthropic 内部的 AI 研发工作中,Claude 已经主导了 26% 的工作。与此同时,超过 90% 的 AI 研发工作已经至少进入了「AI 协作」的阶段。 本周五,Anthropic 发布博客公开了 Claude 研发过程中 AI 自动化的进度。最先进的 AI 现在已经基本上把自我改进当成常规武器在用了。 过去半年时间里,前沿 AI 实验室内部隐藏的自动化曲线已经变得陡峭起来,在 Anthropic 最主要的内部 Agent 平台上,任意时刻都有约 3 万个 AI Agent 正在进行研究和工程工作。仅在今年 8 月,这些 Agent 就产生了超过 10 亿次决策。 某种意义上,一个由人类研究员和数万 AI Agent 共同组成的自动实验室已经开始出现了。 Claude 早已不是 「 帮忙写代码 」 首先需要解释一下所谓「Claude 主导 26% 的研发」究竟是什么意思。 Anthropic 采用了 Epoch AI 提出的一套自动化等级,将 AI 参与工作划分为 AL0 到 AL5 六个阶段。AL0 意味着完全没有 AI 参与,随后依次是 AI 提供少量帮助、AI 辅助、AI 协作、AI 主导,直到最高的 AL5,AI 可以在没有人类参与的情况下完全自主工作。 看起来和自动驾驶的等级划分差不多,目前最值得关注的是 AL4,也就是 Anthropic 所说的「AI leads」。进入这个等级以后,人类通常只需要给 AI 一个大致的目标(就像大家说的许愿式编程),AI 就能够自己完成绝大部分端到端工作。人类不再需要一直守在旁边告诉它下一步做什么,主要只是负责监督、检查以及最后的决策。 Anthropic 举了一个具体的例子。假设一个每天夜间运行的数据管道突然坏掉。在 AL3 阶段,工程师可能需要先检查日志、形成初步判断,然后把信息交给 Claude。遇到意外问题时 Claude 还可能停下来,等工程师决定下一步怎么处理。 到了 AL4,工程师只需要把故障告警丢给 Clau

Related stories

  • Anthropic replaces Git with in-house Scion system as daily merges top 30K

    Anthropic detailed its migration from Git to Scion, an in-house version control system built by seven engineers with Claude in five months, to handle a surge in daily code merges from about 2,500 in May 2026 to over 30,000 in September. The company also developed Antique, a new code review system, and migrated its entire monorepo to Scion, citing Git's decentralized design as a bottleneck for its growing workload.

    4 reports · 4 sources886 points · 1008 comments
  • Anthropic updates usage policy to ban abusive behavior toward Claude

    Anthropic revised its usage policy for the first time in over a year, adding explicit bans on sustained and needless abusive or cruel behavior toward Claude, alongside new rules on election interference, weapons development, surveillance, and deceptive campaigns. The update retains conversation termination as the primary enforcement mechanism for violations. Ordinary frustration and criticism remain permitted.

    3 reports · 3 sources91 points · 225 comments
  • Anthropic launches Claude Haiku 5.5 with effort controls, 90% price cut

    Anthropic released Claude Haiku 5.5, its cheapest and fastest small model, priced at $0.10 per million input tokens and $0.50 per million output tokens for requests up to 100K tokens, a 90% cut from Haiku 4.5. It is the first Haiku model with adjustable reasoning effort, and benchmarks show it outperforming GPT-6 Luna on several tests, including OSWorld 2.1 and GDPval-AA v2.1.

    6 reports · 5 sources1045 points · 487 comments
  • Meta, Microsoft reportedly cut employee use of Anthropic's Claude

    According to The Information, Meta and Microsoft are working to reduce their employees' reliance on Anthropic's Claude, with Meta's Claude Code active users dropping to about 30,000 from 60,000 earlier this yearcars. Microsoft has cut internal Claude spending by over 33%, reducing per-employee monthly token budgets from $100,000 to $10,000 and routing requests to cheaper models via Copilot. Meta is shifting developers to its own tools like Muse Code and MetaCode.

    3 reports · 3 sources375 points · 383 comments

News updates every hour. Subscribe in your RSS reader to get new stories automatically.

Subscribe