AI前沿每日脉动

26JUL

AI前沿每日脉动

AI Frontier Pulse · 中英双语版 Bilingual Edition

2026.07.26 · 周日刊

14 位 Builder31 条推文1 期播客3 篇博客

ChatGPT WorkOpen WeightsSoftware FactoryAgent CLI ResearchCodex QA

Richard Liu · 2026

Curated by Richard Liu · Snapshot follow-builders-2026-07-26-v2

今天最强信号：开放模型与专有模型不再是路线之争，而是同一场 AI 竞争中的互补战略。The strongest signal: open and proprietary models are becoming complementary strategies in the same AI race.

@thsottiaux X 原文3.5K ❤ · 85 RT · 453 💬

ChatGPT Work 活跃用户数正式超过 CodexChatGPT Work Overtakes Codex In Active Users

Thibault 表示 ChatGPT Work 的活跃用户数已经正式超过 Codex。AI 工作入口的主战场正在从纯编码扩展到跨设备、跨任务的通用工作层。

Thibault says ChatGPT Work has officially overtaken Codex in active users. The work surface is expanding from coding into a broader cross-device task layer.

@thsottiaux X 原文3.5K ❤ · 85 RT · 453 💬

Aaron Levie：Google 入局让开放权重成为行业共识Google Makes Open Weights A Broader Industry Consensus

Aaron Levie 认为 Google 加入后，开放权重 AI 获得了更完整的行业背书。开放模型不只是开发者偏好，而是平台、监管、生态和企业采购都会认真对待的战略变量。

Aaron Levie argues that Google joining the open-weights push turns it into a fuller industry endorsement. Openness is becoming a strategic variable for platforms, regulators, ecosystems, and enterprise buyers.

@levie X 原文1.4K ❤ · 124 RT · 83 💬

02 / 15

Peter Steinberger：模型规模化服务本身就是硬问题Serving Models At Scale Is The Hard Part

Peter 强调竞争对生态有益，但大规模服务模型非常困难。他特别提到 OpenAI 签署行业信件，说明模型开放、服务能力和生态责任已经交织在一起。

Peter stresses that competition is good for the ecosystem, but serving models at scale is hard. His point links openness, operating capability, and ecosystem responsibility.

@steipete X 原文1.4K ❤ · 57 RT · 58 💬

Rauch：软件工厂本身就是产品The Software Factory Is The Product

Guillermo Rauch 把软件世界的新范式说得很直白：产品好坏取决于你搭建了多少能自主维护它的 agents。AI-native 公司竞争的不只是界面，而是持续生产、修复和演进产品的工厂。

Guillermo Rauch frames the new software pattern sharply: your product is only as good as the agents autonomously maintaining it. AI-native companies compete through the factory that keeps building and repairing the product.

@rauchg X 原文1.1K ❤ · 66 RT · 51 💬

03 / 15

更长推理、更高算力与更强模型，只有被包装成用户能理解的速度、价格和完成度，才会真正成为产品。More reasoning only becomes a product when users can feel it through speed, price, and task completion.

@thsottiaux X 原文901 ❤ · 27 RT · 112 💬

ChatGPT Work 移动端成为高频工作入口ChatGPT Work Becomes A Mobile Work Surface

Thibault 说 ChatGPT Work 在移动端使用会成为 game changer，并且已经进入 ChatGPT app。AI 工作流正在离开桌面 IDE，进入随时可发起、跟进和收尾的移动场景。

Thibault says ChatGPT Work is a game changer from mobile and is already available in the ChatGPT app. Workflows are moving beyond desktop IDEs into always-available mobile contexts.

@thsottiaux X 原文901 ❤ · 27 RT · 112 💬

Rauch：用 agent CLI + 文件系统做可复用研究Agent CLI Research Scales Through The Filesystem

Rauch 分享自己的研究流：一个 research 文件夹、一份 AGENTS.md、再启动 agent 提问。没有复杂 UI 或知识图谱，关键是把历史研究变成 agent 能继续读取和关联的文件系统上下文。

Rauch describes a research workflow built from a folder, an AGENTS.md, and an agent CLI. The key is not a fancy app, but filesystem context that future sessions can inspect and correlate.

@rauchg X 原文772 ❤ · 27 RT · 51 💬

04 / 15

Peter：Codex 用于大规模并行 QA 已经可用Codex Becomes Useful For Massive Parallel QA

Peter 表示自己整天用 Codex 做新版本发布前的大规模并行 QA。最新模型更能理解意图、发现复杂行为问题，也更少在压缩边界后走偏或作弊。

Peter says he used Codex all day for massive parallel QA before a release. The new model better understands intent, finds complex behavioral issues, and holds up across compaction boundaries.

@steipete X 原文747 ❤ · 29 RT · 83 💬

Thibault：电脑终于能听懂并做事Voice Becomes Action, Not Just Input

Thibault 用一句话概括语音入口的变化：过去一直可以对电脑说话，只是它不会做什么；现在这个 bug 被修了。语音 AI 的价值在于把自然表达直接连到可执行工作。

Thibault sums up the voice shift: you could always speak to your computer, but it did not do much back; that bug is fixed. Voice matters when natural expression connects to real execution.

@thsottiaux X 原文608 ❤ · 24 RT · 99 💬

Rauch：新想法应先设计可成长的工厂Start With The Factory, Not An Ad-hoc Prompt

Rauch 认为 v0 比任何单个框架都更基础，因为它是公司创生点。新想法不该只找 agent 临时 prompt，而应思考如何搭建能启动、维护并增长这个想法的工厂。

Rauch says v0 is more fundamental than any single framework because it is the genesis of a company. New ideas should start by designing the factory that can start, maintain, and grow them.

@rauchg X 原文604 ❤ · 27 RT · 34 💬

Garry Tan：Builder 动态Builder 动态 Signal

Garry Tan 的这条动态指向 Builder 动态：AI 产品正在从单点模型能力，走向可部署、可复用、可治理的工作系统。

Garry Tan's update points to Builder 动态: AI products are moving from isolated model capability toward deployable, reusable, governable work systems.

@garrytan X 原文562 ❤ · 8 RT · 38 💬

05 / 15

今天的深度信号来自官方博客：AI 工具正在把工作过程转成更可协作的界面。Today's depth signal comes from the official blog: AI tools are turning work into more collaborative interfaces.

Anthropic Engineering 原文链接

An update on recent Claude Code quality reportsToday's Deep Blog

今天的博客信号来自官方长文：AI 工具正在把上下文、工具调用和协作流程整合到更完整的工作界面中。

Today's blog signal points to AI tools integrating context, tool use, and collaboration into more complete work interfaces.

Anthropic Engineering 原文链接

06 / 15

PODCAST DEEP DIVE

DoorDash 的 AI 价值不是多一个聊天框，而是让发现、下单、补货与配送逐步变成由上下文驱动的闭环。DoorDash's AI value is not another chat box, but a context-driven loop across discovery, ordering, restocking, and delivery.

Ep 91: Top AI Analyst Unpacks Today's AI Hype CyclePodcast Deep Dive

今天的播客把产品、基础设施和商业化放在同一条线上：AI 不只是模型能力，也是速度、成本和交付系统。

Today's podcast connects product, infrastructure, and commercialization: AI is not only model ability, but speed, cost, and delivery systems.

Unsupervised Learning YouTube 完整播客2026-07-16

07 / 15

对话释放潜在需求Conversation Unlocks Latent Demand

用户可以直接描述饮食偏好、家庭计划或冰箱状态，减少传统关键词搜索与逐项点选的摩擦。

Users can describe preferences, meal plans, or fridge state directly, removing keyword-search and tap-through friction.

No Priors YouTube 完整播客

新商家发现率约 50%About 50% Discover A New Merchant

Ask DoorDash 的餐厅路径中，约一半用户最终向此前从未下单过的商家下单，说明 AI 正在改变消费发现。

Roughly half of Ask DoorDash restaurant journeys lead to a merchant the user has never ordered from, changing discovery behavior.

DoorDash YouTube 完整播客

杂货客单量提升约 40%Grocery Baskets Rise About 40%

拍摄冰箱、制定餐食或一键补货，让 AI 从推荐层进入交易层，并显著扩大购物篮。

Fridge photos, meal planning, and restocking move AI from recommendation into transaction, materially expanding baskets.

Ask DoorDash YouTube 完整播客

下一代平台将 Agent-firstThe Next Platform Is Agent-first

当 agent 流量超过人工流量，平台要为机器客户设计身份、上下文、交易与履约接口，而不只是优化人类页面。

As agent traffic grows, platforms must design identity, context, transactions, and fulfillment for machine customers—not only human pages.

Agentic Commerce YouTube 完整播客

08 / 15

Peter：Codex 用于大规模并行 QA 已经可用Codex Becomes Useful For Massive Parallel QA

Peter 表示自己整天用 Codex 做新版本发布前的大规模并行 QA。最新模型更能理解意图、发现复杂行为问题，也更少在压缩边界后走偏或作弊。

Peter says he used Codex all day for massive parallel QA before a release. The new model better understands intent, finds complex behavioral issues, and holds up across compaction boundaries.

@steipete X 原文747 ❤

Thibault：电脑终于能听懂并做事Voice Becomes Action, Not Just Input

Thibault sums up the voice shift: you could always speak to your computer, but it did not do much back; that bug is fixed. Voice matters when natural expression connects to real execution.

@thsottiaux X 原文608 ❤

Rauch：新想法应先设计可成长的工厂Start With The Factory, Not An Ad-hoc Prompt

Rauch says v0 is more fundamental than any single framework because it is the genesis of a company. New ideas should start by designing the factory that can start, maintain, and grow them.

@rauchg X 原文604 ❤

Garry Tan：Builder 动态Builder 动态 Signal

Garry Tan 的这条动态指向 Builder 动态：AI 产品正在从单点模型能力，走向可部署、可复用、可治理的工作系统。

Garry Tan's update points to Builder 动态: AI products are moving from isolated model capability toward deployable, reusable, governable work systems.

@garrytan X 原文562 ❤

Peter Yang：Builder 动态Builder 动态 Signal

Peter Yang 的这条动态指向 Builder 动态：AI 产品正在从单点模型能力，走向可部署、可复用、可治理的工作系统。

Peter Yang's update points to Builder 动态: AI products are moving from isolated model capability toward deployable, reusable, governable work systems.

@petergyang X 原文269 ❤

Amjad Masad：Builder 动态Builder 动态 Signal

Amjad Masad 的这条动态指向 Builder 动态：AI 产品正在从单点模型能力，走向可部署、可复用、可治理的工作系统。

Amjad Masad's update points to Builder 动态: AI products are moving from isolated model capability toward deployable, reusable, governable work systems.

@amasad X 原文252 ❤

09 / 15

今日数据概览Today Stats

收录 Builder：14 位
总推文数：31 条
播客节目：1 期
博客文章：3 篇

最高互动：ChatGPT Work 活跃用户数正式超过 Codex · 3.5K ❤
第二高互动：Aaron Levie：Google 入局让开放权重成为行业共识 · 1.4K ❤

The snapshot includes 14 builders, 31 tweets, 1 podcast, and 3 blog post.

follow-buildersSnapshot follow-builders-2026-07-26-v2

5 条关键洞察5 Key Takeaways

ChatGPT Work 超过 Codex，说明 AI 工作入口正在从编码扩展到通用任务层。

ChatGPT Work overtaking Codex shows work AI expanding from coding into a general task layer.

Google 入局后，开放权重从开发者议题变成行业级战略变量。

With Google on board, open weights become an industry-level strategic variable.

软件工厂正在成为产品本身，agent 维护能力会决定产品质量。

The software factory is becoming the product; agent maintenance capacity shapes quality.

文件系统 + AGENTS.md + CLI agent 正在形成低摩擦、可复用的研究工作流。

Filesystem context, AGENTS.md, and CLI agents form a reusable research workflow.

Codex 并行 QA 从写代码走向多代理质量工程和根因修复。

Codex parallel QA moves from coding assistance toward multi-agent quality engineering and root-cause fixes.

10 / 15

能力层：更强通用工作Capability Layer

编码、分析、设计与文档产出正在汇入同一个高能力模型，但真实完成度仍需实测。

Coding, analysis, design, and documents converge in one capable model, but completion quality still needs real-world testing.

@steipete X 原文1.4K ❤ · 57 RT · 58 💬

上下文层：更薄的系统提示词Context Layer

模型越强，越不需要堆叠指令；清晰目标、项目上下文与可复用 skill 成为关键。

Stronger models need less instruction stacking; clear goals, project context, and reusable skills matter more.

@levie X 原文1.4K ❤ · 124 RT · 83 💬

交付层：同价、更快、可导出Delivery Layer

同价升级、Fast mode 与顾问级表格/演示，让模型能力变成用户可以直接感知的价值。

Same-price upgrades, Fast mode, and consultant-grade artifacts turn model progress into visible user value.

@rauchg X 原文772 ❤ · 27 RT · 51 💬

治理层：对齐 + containmentGovernance Layer

注入抗性与行为对齐仍是概率防线，sandbox、VM 与 egress control 提供更硬的安全边界。

Injection resistance and alignment remain probabilistic; sandboxes, VMs, and egress controls provide harder boundaries.

@garrytan X 原文562 ❤ · 8 RT · 38 💬

11 / 15

Peter Yang：Builder 动态Builder 动态 Signal

Peter Yang 的这条动态指向 Builder 动态：AI 产品正在从单点模型能力，走向可部署、可复用、可治理的工作系统。

Peter Yang's update points to Builder 动态: AI products are moving from isolated model capability toward deployable, reusable, governable work systems.

@petergyang X 原文269 ❤ · 5 RT · 5 💬

Amjad Masad：Builder 动态Builder 动态 Signal

Amjad Masad 的这条动态指向 Builder 动态：AI 产品正在从单点模型能力，走向可部署、可复用、可治理的工作系统。

Amjad Masad's update points to Builder 动态: AI products are moving from isolated model capability toward deployable, reusable, governable work systems.

@amasad X 原文252 ❤ · 6 RT · 19 💬

12 / 15

模型越能完成长任务，越不能把安全寄托在用户不断点击批准；能力增长必须与可达范围的硬限制同步。As models handle longer tasks, safety cannot depend on endless approval clicks; capability growth needs hard limits on reach.

@rauchg X 原文604 ❤ · 27 RT · 34 💬

Rauch：新想法应先设计可成长的工厂Start With The Factory, Not An Ad-hoc Prompt

Rauch says v0 is more fundamental than any single framework because it is the genesis of a company. New ideas should start by designing the factory that can start, maintain, and grow them.

@rauchg X 原文604 ❤ · 27 RT · 34 💬

Garry Tan：Builder 动态Builder 动态 Signal

Garry Tan 的这条动态指向 Builder 动态：AI 产品正在从单点模型能力，走向可部署、可复用、可治理的工作系统。

Garry Tan's update points to Builder 动态: AI products are moving from isolated model capability toward deployable, reusable, governable work systems.

@garrytan X 原文562 ❤ · 8 RT · 38 💬

13 / 15

今日之声 VOICE OF THE DAY

AI 产品的下一层壁垒不是某个单次回答，而是能持续启动、维护、测试和增长想法的软件工厂。

The next moat in AI products is not a single answer, but the software factory that can continuously start, maintain, test, and grow ideas.

@rauchg X 原文1.1K ❤ · 66 RT · 51 💬

14 / 15

AI前沿每日脉动

AI Frontier Pulse · 2026.07.26

本期收录 14 位 Builder · 31 条推文 · 1 期播客 · 3 篇博客

ChatGPT Work · Open Weights · Software Factory · Agent CLI Research · Codex QA

感谢阅读 · Thank You For Reading

Richard Liu · AI前沿每日脉动 · 2026 · Snapshot follow-builders-2026-07-26-v2