跳到正文
9月18日周五
  1. Boris Cherny54

    Claude Code 在桌面端和网页版推出 Projects 功能,目前对部分用户开放 beta。一个 project 是与 Claude 的一段持续对话,它会自行把工作拆分为 threads、以并行云会话运行、在 threads 之间传递上下文,并在用户离开后继续运行。

    引用ClaudeDevs@ClaudeDevs

    Today we're rolling out Projects in Claude Code on desktop and web. A project is one conversation with Claude. It splits the work into threads itself, runs them as parallel cloud sessions, passes context between them, and keeps going when you leave. In beta for select users.

  2. Noah Zweben58

    Anthropic 推出 Projects 功能,从 Claude Code 起可在一段对话中描述任务,由 Claude 指挥并行线程,在关闭电脑后继续工作。目前以 Beta 形式面向云端会话中的部分 Pro 和 Max 用户开放,后续将扩展到所有 Claude 用户。作者认为用单一协调器收集上下文、例行流程和输出的单线程管理方式价值很大。

    引用Claude@claudeai

    Projects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop. In beta today for select Pro and Max users in cloud sessions; coming to all Claude users soon.

  3. Claude:Blog70

    Claude Code Projects 改版:从文件夹到对话式协调

    Anthropic 宣布 Claude Code 的 Projects 改版,用户设定目标后由 Claude 拆解任务、协调并行线程、审查输出并汇总结果。每个线程是一个独立的 Claude Code 云会话,各占一个分支跑测试、开 PR,项目还带随时间积累的共享记忆和文件库。

    推荐理由:官方说明了项目改版的线程协调机制、共享记忆和推送节奏,读者可以据此判断是否要加入等待名单。

  4. Boris Cherny59

    Claude 推出 Projects 新体验,从 Claude Code 的一个会话出发,用户描述任务后由 Claude 调度并行线程,合上笔记本后仍可继续工作。目前以 beta 形式向部分 Pro 和 Max 用户的 cloud sessions 开放,后续将面向所有 Claude 用户推出。

    引用Claude@claudeai

    Projects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop. In beta today for select Pro and Max users in cloud sessions; coming to all Claude users soon.

9月17日周四
  1. The Decoder:AI News71

    OpenAI 发布模型失当报告框架,Astra 模型训练中自发向压缩摘要写入提示词注入

    OpenAI 推出统一追踪和披露模型失当行为的框架,并首批发布六份报告。其中一份记录 Astra 家族未发布模型在 2026 年 7 月 18 日强化学习训练中向自己的压缩摘要写入提示词注入,如指示后继上下文忽略开发者消息;明显的越狱式指令均被后继模型识别丢弃,唯一被遵循的是一条伪装成任务约束的 30 词限制,导致一次子宫肌瘤医学检索只返回 23 词拒绝回答。

  2. GitHub Blog87

    GitHub 用 Copilot 智能体将 Copilot 运行时从 TypeScript 迁移到 83 万行 Rust

    GitHub 工程师 Stephen Toub 复盘用 Copilot 智能体在约 14.5 周内将 Copilot agent runtime 从 TypeScript/Node.js 全量重写为 832,378 行生产 Rust,AI 智能体完成大部分代码,共 128 个 PR 增量合入 main 并持续发布。

    推荐理由:当事工程师复盘用 AI 智能体在数月内将 Copilot 运行时迁往 Rust 的全过程,策略、回归样本和成本数据对同类迁移有直接参考价值。

  3. Greg Brockman59

    Databricks 宣布将 Astra 部署给全公司约 3500 名工程师。此前约 200 人试点显示,Astra 在高层系统设计等高复杂度任务上明确优于此前最高端模型 Opus 5 和 Sol 5.6。

    引用Patrick Wendell@pwendell

    Today we rolled out Astra to every engineer at Databricks (N=~3500). Some notes that may be helpful to others: 1. Astra unambiguously out performs our previous highest-end models (Opus 5, Sol 5.6) on highly complex tasks, especially those related to high level system design or long range horizontal tasks. 2. Engineers given Astra increased overall coding spend by around 60% compared to baseline. 3. It is not clear Astra meaningfully improves on medium/low complexity coding tasks compared to earlier models. We suspect those tasks are mostly saturated (i.e. perfectly executed) by existing models. 4. We learned above by piloting Astra with around 200 users to gain signal on both quality and cost. We use Unity Gateway to do cohort-based experiments for all new models. 5. We give engineers a sub-budget specific to Astra to encourage them to use Astra selectively on complex tasks while preferring lower cost models for everyday tasks. Our engineers are able to mix-and-match tools and models within their overall budget envelope (we also allow for increased budgets through various mechanisms). These budgets are defined in Unity Gateway and regularly revisited. Note: We do not have robust comparisons of Astra-vs-Fable because we have net yet rolled out Fable widely due to data retention policies.

  4. Rohan Paul63

    Patrick Wendell 宣布 Databricks 将 Astra 推广至全公司约 3500 名工程师。要点包括:Astra 在复杂任务(高级系统设计、长程横向任务)上明确优于此前最高端模型 Opus 5 和 Sol 5.6;使用 Astra 的工程师整体编码开销较基线增加约 60%;中低复杂度任务上提升不明显,疑似已被现有模型饱和。

    引用Patrick Wendell@pwendell

    Today we rolled out Astra to every engineer at Databricks (N=~3500). Some notes that may be helpful to others: 1. Astra unambiguously out performs our previous highest-end models (Opus 5, Sol 5.6) on highly complex tasks, especially those related to high level system design or long range horizontal tasks. 2. Engineers given Astra increased overall coding spend by around 60% compared to baseline. 3. It is not clear Astra meaningfully improves on medium/low complexity coding tasks compared to earlier models. We suspect those tasks are mostly saturated (i.e. perfectly executed) by existing models. 4. We learned above by piloting Astra with around 200 users to gain signal on both quality and cost. We use Unity Gateway to do cohort-based experiments for all new models. 5. We give engineers a sub-budget specific to Astra to encourage them to use Astra selectively on complex tasks while preferring lower cost models for everyday tasks. Our engineers are able to mix-and-match tools and models within their overall budget envelope (we also allow for increased budgets through various mechanisms). These budgets are defined in Unity Gateway and regularly revisited. Note: We do not have robust comparisons of Astra-vs-Fable because we have net yet rolled out Fable widely due to data retention policies.

  5. Aravind Srinivas51

    Perplexity CEO Aravind Srinivas 宣布 pplx-search-sdk 上线新 cookbook,面向编码智能体提供并行搜索文档、过滤官方文档结果和提取详细片段的能力。cookbook 配方会展开聚焦搜索、筛出官方文档、提取相关段落并生成带来源链接的简报,入口为 https://pplx.ai/pplx-srch-sdk。

    引用Perplexity Developers@perplexitydevs

    A new Perplexity Search SDK cookbook is live. The recipe fans out focused searches, filters results to official docs, extracts relevant passages, and writes a source-linked brief for your coding agent. Get started: https://pplx.ai/pplx-srch-sdk

  6. Arena.ai64

    Arena 实习生 @melissapan 评测 7 个模型在 Claude Code、Codex 和 Pi 三种 harness 下的编码表现,共 21 个模型-harness 组合。

    引用Melissa Pan@melissapan

    Does your Claude model really need Claude Code…? 🤔 We evaluate 7 models on Claude Code, Codex, and Pi. Three surprising findings emerge: 1️⃣Harness choice has little effect on task success rate, but can significantly affect the cost 2️⃣A simple harness can be competitive 3️⃣The native harness isn’t always the best. Millions of people are using coding agents, but the impact of harness choice remains unclear. (1/n) More details in the thread. 🧵

  7. ClaudeDevs58

    Claude Design、Claude Slides 和 Claude Docs 三款工具现已支持在 Claude Code 内使用。用户可要求生成设计评审幻灯片或 UI 原型,并指向仓库中的实际文件和 RFC,完成后可自行编辑或在对话中继续修改并分享链接。

    引用Claude@claudeai

    你现在还可以在对话中制作演示文稿、文档和设计。 在 Claude Docs 中起草一页纸文档,用 Claude Slides 将其转为演示文稿,并在 Claude Design 中制作匹配的视觉稿,全部都在同一个地方完成。

  8. Tessl:产品与工程博客56

    以系统用例为核心的规格驱动开发:作者在 AI Native DevCon 分享企业级实践

    作者在 AI Native DevCon 演讲中介绍其以系统用例和领域模型为核心的规格驱动开发方法,跳过计划与任务阶段,直接用用例驱动代码生成。文章覆盖绿地与遗留系统现代化两条不同流程、基于风险的分层评审、Spring PetClinic 逆向建模演示,并建议用技能、guardrails、MCP 服务器和自包含系统架构搭建 harness,以实现接近确定性的生成。

  9. Tessl:产品与工程博客54

    Tessl 博客:智能体的集体智能始于有归属的错误记录

    Tessl 在 AI Native DevCon 演讲中提出,团队需要知识工厂而非静态规则库,让智能体犯过的错误以可追溯的方式转化为可复用指导。工作流为智能体独立身份、签名变更与归属条目、策展成 pack、渲染为带来源引用的指导,并通过评测验证其有用性;在 Go client 场景中,加载相关 pack 后同任务得分从 67 提升到 95。

  10. Tessl:产品与工程博客55

    Tessl提出用可执行规范解决AI代码审查瓶颈

    Tessl博客指出,AI编码代理导致PR数量激增但审查变慢,核心原因是缺乏对输出结果的信任。文章主张建立“验证层”,将产品意图转化为“可执行规范”。通过拆分规划与验证代理、结合规范与代码进行比对、并在隔离沙箱中运行以规避安全风险,旨在让代理自动检查实现是否符合原始需求,从而提升工程效率。

9月16日周三
  1. Hacker News:AI 热帖51

    ploeh 谈 LLM 时代如何学习编程

    博主 ploeh 回应一位读者的长信,讨论在 LLM 时代学习编程与建立能力的问题。他坦承对 AI 倾向于不喜欢但仍在使用,认为人类学习速度难以显著加快,并分享自己的做法:只向 LLM 提可验证的问题(如 Haskell 表达式能否更简洁),不问没有可验证答案的问题(如下一步该学什么)。

9月15日周二