System translated (Gemini)

2026-03-30 AI Daily | Stripe Reveals Paradigm for Agent Delivery at Scale, Axios Supply Chain Attack Triggers Security Alert Link to heading

Today, the AI industry shows a deep trend of transformation from “dialogue boxes” to an “execution layer.” Stripe has revealed an engineering paradigm for its internal agents that deliver 1,300 PRs weekly, signaling that Agent implementation has entered an explosive growth phase. Meanwhile, the CLI is replacing the GUI as the mainstream interaction protocol for agents. Furthermore, the Axios supply chain attack sounds a warning for AI infrastructure security, and multimodal competition continues to intensify with the launch of Qwen3.5-Omni.

📖 In-depth Guide to This Issue’s Watch List Link to heading

Today’s high-quality updates reveal that AI is making a deep shift from “dialogue boxes” to an “execution layer.”

First, the engineering and implementation of AI Agents have entered an explosive phase. In a podcast, Stripe engineers provided a detailed review of how their internal “Minions” agent delivers 1,300 PRs weekly via Slack, offering an excellent efficiency paradigm for engineering teams. Meanwhile, the rise of OpenClaw and Claude Code marks the advent of the “Personal AI Operating System” era. The practices of pioneers like Claire Vo demonstrate the immense potential of open-source agents in managing complex life and work flows.

Second, underlying architecture and business models are reaching a critical turning point. Arm CEO Rene Haas, in an interview with Stratechery, detailed the company’s strategic shift from IP licensing to direct chip sales. The strong rise of Anthropic and OpenAI’s strategic adjustments indicate that the large model competition has shifted from a mere parameter race to a more pragmatic contest for ecosystem positioning.

Additionally, the case of the 230-year-old company STADLER embedding AI into the traditional recycling industry offers profound insights for established enterprises on how to reshape knowledge work with AI. It is recommended to focus on Stripe’s agent practices and Arm’s business model shift.

🌐 AI Hotspots on X Link to heading

Topic 1: Axios npm Package Hit by Supply Chain Attack Link to heading

  • Category: AI · News
  • Summary: Trending for: 6 hours, Related posts: 14,000
  • What happened: The npm account for the popular JavaScript HTTP client library Axios was hijacked. Attackers published malicious versions (1.14.1 and 0.30.4) containing a Remote Access Trojan (RAT), infecting developer systems by injecting malicious dependencies.
  • Why it matters: Axios is a core component in the AI development ecosystem for calling LLM APIs and data collection. Such attacks can lead to the leakage of AI developers’ API keys, compromised model training environments, and exfiltration of sensitive data, posing a significant threat to AI infrastructure security.
  • Discussion summary: Current discussions focus on the fragility of the npm ecosystem, the urgency of enforcing two-factor authentication (2FA), and how developers can defend against such risks by locking version numbers, disabling install scripts (–ignore-scripts), and using supply chain monitoring tools.

Topic 2: Operation Epic Fury Hits Fifth Week with 11,000 Strikes on Iran Link to heading

  • Category: AI · News
  • Summary: Trending for: 1 day, Related posts: 237,000
  • What happened: “Operation Epic Fury” has entered its fifth week, with a reported cumulative total of 11,000 strikes against targets in Iran.
  • Why it matters: The operation highlights the core role and large-scale application of AI-driven automated target recognition, drone swarms, and algorithmic battlefield management systems in modern high-intensity military conflicts.
  • Discussion summary: Discussions are focused on the precision of AI-assisted strikes versus the risk of collateral damage, the ethical controversies surrounding the shift towards highly automated warfare, and the operation’s significant impact on global geopolitics and cybersecurity.

Topic 3: Stanford’s Meta-Harness Lets AI Agents Auto-Improve Their Own Code Link to heading

  • Category: AI · News
  • Summary: Trending for: 11 hours, Related posts: 745
  • What happened: Stanford University researchers have released the Meta-Harness framework, which enables AI agents to automatically iterate, fix, and optimize their own code logic through a self-feedback mechanism.
  • Why it matters: This technology marks a shift for AI from passively executing instructions to becoming autonomous agents with “self-evolving” capabilities. It is expected to significantly improve the success rate of complex, long-horizon tasks and reduce the cost of manual intervention.
  • Discussion summary: Discussions focus on the framework’s generalization capabilities when handling complex real-world tasks, as well as the potential security risks and challenges to logical controllability posed by AI autonomously iterating its own code.

Topic 4: Alibaba Launches Qwen3.5-Omni Multimodal AI Models Link to heading

  • Category: AI · News
  • Summary: Trending for: 18 hours, Related posts: 4,700
  • What it is: Alibaba officially released the Qwen3.5-Omni multimodal AI model, designed to achieve real-time, seamless interaction across various modalities such as text, images, and audio.
  • Why it matters: This release marks a significant advancement in the full-modal integration capabilities of China’s large models, directly competing with GPT-4o and further intensifying global technological competition in the field of multimodal understanding and generation.
  • Discussion summary: Discussions on X focus on the performance rivalry between Qwen3.5, OpenAI’s GPT-5.2, and Nvidia’s Nemotron 3, as well as the impact of the dense wave of AI releases in December on the global AI competitive landscape in 2025.

Topic 5: Claude Code’s New Feature Hits Usage Limit Snags Days After Launch Link to heading

  • Category: AI · News
  • Overview: Trending for: 1 day ago, Related posts: 24,000
  • What it is: Claude Code, an AI command-line programming tool launched by Anthropic, has caused widespread dissatisfaction just days after its release due to users frequently hitting strict usage limits.
  • Why it matters: This incident highlights the immense demand for computing power from high-frequency interactive AI agents when handling complex tasks, and the challenges top AI companies face in balancing operational costs with the developer user experience.
  • Discussion summary: Discussions focus on Claude Code’s extremely high token consumption rate, the interruption of development workflows by usage limits, and the intense cost-performance comparison with existing programming assistance tools like Cursor.

Topic 6: Build in Public Makers Share Latest Projects in Weekly Thread Link to heading

  • Category: AI · News
  • Overview: Trending for: 16 hours ago, Related posts: 807
  • What it is: Developers from the “Build in Public” community on X showcased their latest AI projects and development progress in a regular weekly thread.
  • Why it matters: This phenomenon reflects the rapid iteration happening at the AI application layer, demonstrating how independent developers leverage the capabilities of a large model to quickly build product prototypes and validate them in the market.
  • Discussion summary: Discussions center on the functional innovations of specific tools, suggestions for optimizing product UI/UX, and how to find a differentiated position in the highly competitive market of AI plugins and tools.

Topic 7: Senator Graham Faces Backlash for Disney Trip Amid Shutdown and War Link to heading

  • Category: AI · Other
  • Overview: Trending for: 18 hours ago, Related posts: 92,000
  • What it is: U.S. Senator Lindsey Graham faced widespread criticism on social media for visiting Disneyland while the government was facing a shutdown crisis and international conflicts were ongoing.
  • Why it matters: A potential government shutdown would directly lead to a standstill in the legislative process for AI regulation, affecting the formulation of related technology policies, security reviews, and the implementation of industry funding.
  • Discussion summary: Public opinion is focused on the professional conduct of public officials during times of crisis and the negative impact of political gridlock on key national agendas, including technology governance.

Summary of AI Public Opinion on X Today Link to heading

Today’s public opinion focuses on the security vulnerabilities and autonomous evolution potential exposed as AI technology permeates all sectors, revealing a multi-dimensional interplay from low-level supply chain security to high-level geopolitical games. There is a strong industry consensus on the efficiency gains from AI-driven automation tools. It is widely acknowledged that the supply chain security of the open-source ecosystem has become a critical weakness in AI infrastructure, and multimodal competition has reached a fever pitch. However, public opinion is significantly divided on the ethical boundaries of AI in military applications, the balance between computing power costs and high-frequency interactive tools, and the controllability of autonomously iterating code. Potential risks not only include developer key leakage due to malicious code injection but also the regulatory vacuum that political gridlock could create, and the potential for logical runaway as AI pursues “self-evolution.”

💡 Influencer Insights Link to heading

Hello! I am a senior AI industry analyst. Based on the tweet content from key AI opinion leaders (Influencers) on X over the past 24 hours, I have compiled today’s AI industry briefing for you.

The “Devolution” of Agent Interfaces: CLI (Command-Line Interface) Becomes the New Infrastructure Link to heading

The most significant trend today is the full-scale return of the CLI. Several influential figures have pointed out that in the era of AI Agents, the CLI is replacing the GUI as the most natural “hand” for an Agent.

  • Feishu Open-Sources CLI: Feishu released lark-cli, allowing Agents to directly manipulate documents, calendars, messages, and more. @dotey believes that the CLI, with its built-in manual (--help) and natural fit for text-based interaction, is the native interface for AI. @op7418 even observed that WeChat Work has also started to adopt the MCP (Model Context Protocol).
  • Infrastructuralization: @vista8 mentioned that all products aiming to integrate with Agents are developing CLIs (e.g., Google’s gws). @Gorden_Sun believes the shift of software towards CLI is an irreversible trend.

Claude Code’s “Ecosystem Hegemony” and Functional Leap Link to heading

Anthropic’s Claude Code continues to dominate discussions, with its feature iteration speed being astonishing:

  • Computer Use Implementation: Claude Code now integrates “computer control” features, allowing it to directly manipulate a Mac desktop for UI testing, clicks, and screenshot verification. @AI_Jasonyu calls it “a programmer agent that can do the work itself.”
  • OpenAI’s “Reverse Output”: In a dramatic turn of events, OpenAI officially released the codex-plugin-cc plugin, enabling developers to call Codex for code reviews from within Claude Code. @dotey pointed out that this is OpenAI proactively stepping into a competitor’s territory.
  • Cost and Bug Controversy: @op7418 shared a reverse-engineering analysis on a Claude Code cache invalidation issue that led to a 10-20x spike in Token consumption, warning users to be wary of using the --resume command.

Vibe Coding: A New Aesthetic-Driven Development Paradigm Link to heading

@Pluvio9yte broke down the Vibe Coding workflow in detail: instead of grinding away at CSS, developers now extract visual inspiration with AI (Nano Banana), inject motion effects (Kling 3.0), and let a website-building AI (Lovable) handle the execution. This “aesthetics-first, AI-adapts” model is reshaping the productivity of indie developers.

2. Unique Perspectives and Industry Foresight Link to heading

  • Test Cases are the New Moat: @ruanyf argues that when AI can replicate large software like Next.js at a very low cost, the code itself loses its protective moat. Test cases will become the key asset to prevent rapid replication by AI.
  • AI Security’s “Classical Chinese” Weak Spot: @lijigang shared a paper on the CC-BOS framework, pointing out that the safety guardrails of large models are mostly designed for modern languages. When confronted with classical Chinese, whose intent is often obscure, the safety alignment mechanisms fail nearly 100% of the time. This suggests that current alignment is “pattern matching” rather than “intent understanding.”
  • Agents as the “New Demographic Dividend”: @lijigang proposed that the greatest certainty in the next 10 years is the “new demographic dividend” brought by Agents, and predicted the necessity of an Agent ID system and a monetary system (a “wallet”).
  • LLMs’ Asymmetric Advantage in Vulnerability Discovery: @Gorden_Sun shared the view from an Anthropic scientist that LLMs can now autonomously discover complex Linux kernel vulnerabilities that have existed for 20 years, disrupting the balance in the security industry.
  • Global Pricing Philosophy: @gefei55 suggested that “bold pricing” is crucial for filtering high-quality users and enhancing product vitality. They also noted that search engine algorithms (e.g., dwell time) can paradoxically cause products with a “poor experience” to rank higher.

Development and Data Tools Link to heading

  • Pretext: A pure TS text measurement library developed by React core member Cheng Lou. It can accurately calculate typography without DOM dependency, boosting speed by 500x. @op7418 believes this is crucial for AI real-time UI generation.
  • XCrawl: A one-stop data collection platform for the AI era, supporting Markdown/JSON output and compatible with Agents like Claude Code (recommended by @Gorden_Sun).
  • PaddleOCR: An open-source project by Baidu, its star count has surpassed Tesseract to become number one globally. @AI_Jasonyu recommends calling its Skill within an Agent for high-precision PDF parsing.

Productivity and Multimedia Link to heading

  • Typeless: A speech-to-text tool that automatically formats the output, ideal for quick input on mobile devices (recommended by @AI_Jasonyu).
  • CapWords: A uniquely creative AI foreign language learning tool that combines image cutouts, recognition, and flashcards, providing a strong “game-like” feel (recommended by @nishuang).
  • Live Translate (Google): The new version of Google Translate integrates Gemini, supporting real-time simultaneous interpretation through headphones (recommended by @AI_Jasonyu).
  • Voxtral TTS: An open-source 4B parameter TTS model from Mistral AI, featuring ultra-low latency and support for voice cloning (recommended by @Gorden_Sun).

Security Alerts Link to heading

  • axios Poisoning Incident: @zhixianio urgently warns that some versions of axios (1.14.1, 0.30.4) and the malicious module plain-crypto-js pose a poisoning risk. All Agent developers are advised to inspect their environments.

Analyst’s Briefing: Today’s developments show that AI is rapidly evolving from a “brain in a chatbox” to a “hand in the operating system.” The explosion of CLIs signals the standardization of Agent interaction protocols, while Claude Code’s takeover of desktop control heralds the dawn of the “fully autonomous employee” era. Developers should focus on CLI friendliness and automated testing, which will be more crucial than mere code generation capabilities in the upcoming competition.

📚 Appendix: Today’s Watch List Update Source List Link to heading

Time window: Last 7 days; 16 sources covered; 21 updates in total.

Y Combinator Podcast (B_intro+search) Link to heading

  • How François Chollet Is Building A New Path To AGI
    • Publication Time: 2026-03-27 23:00 Beijing Time
    • Abstract: - You may have heard of OpenClaw (formerly Clawdbot/Moltbot).
      • The sensational open-source AI assistant that can run on your own devices, connect with the messaging apps you already use, and go beyond chatting to actually perform tasks like managing emails, calendars, files, workflows, and more.
      • Now meet the person behind it.
      • YC’s Raphael Schaad sits down with OpenClaw founder Peter Steinberger to discuss the “aha” moment behind the viral personal AI agent, why local-first agents could replace many of today’s apps, and how personal agents will reshape the future of software.
    • EN Highlights:
      • François Chollet has spent years asking a different question than most of the AI world
      • Instead of scaling what already works, he’s trying to understand what intelligence actually is—and how to build it from first principles
      • In this episode of Lightcone, he traces that path from his early work on deep learning to the creation of the ARC prize, and the launch of ARC V3, a new benchma…
      • He explains why today’s systems may be hitting limits, what recent breakthroughs really mean, and why reaching true general intelligence may require a fundament…

All-In Podcast (A_full) Link to heading

  • Anthropic’s Generational Run, OpenAI Panics, AI Moats, Meta Loses Lawsuits

    • Publication Time: 2026-03-28 04:03 Beijing Time
    • Abstract: - (0:00) Bestie intros.
      • : Friedberg for Governor of California.
      • (2:25) Anthropic’s generational run (15:45) OpenAI: getting focused or panic mode.
      • (36:56) AI valuation impacts, moats, and disruption (43:58) Liquidity speaker announcement, 100x AI moment (50:35) Two landmark social….
    • EN Highlights:
      • (0:00) Bestie intros
      • : Friedberg for Governor of California
      • (2:25) Anthropic’s generational run (15:45) OpenAI: getting focused or panic mode
      • (36:56) AI valuation impacts, moats, and disruption (43:58) Liquidity speaker announcements, the 100x AI moment (50:35) Two landmark socia…

  • Bryan Johnson: I Just Took the Most Powerful Dose of DMT in the World… Here’s What It Was Like

    • Release Time: 2026-03-26 08:10 Beijing Time
    • Abstract: - The Pod by 8 Sleep can cool your bed to 55°F and uses Autopilot to optimize your sleep throughout the night.
      • Follow Bryan: Follow friends: Follow on X: Follow on Instagram: Follow on TikTok: Follow on LinkedIn: Intro Music Source: Intro Video Source: (0:00) David Friedberg intros Bryan Johnson (0:54) Why Bryan Johnson did 5-MeO-DMT (12:56) What brain scans actually show (18:36) Psychosis, bad trips, and life-changing decisions (26:23) The next frontier: organoids and gene therapy (33:26) GLP-1, abundance, and human optimization (35:35) The longevity drug no one is talking about?
      • Follow Bryan: Follow friends: Follow on X: Follow on Instagram: Follow on TikTok: Follow on LinkedIn: Intro Music Source: Intro Video Source: (0:00) David Friedberg intros Bryan Johnson (0:54) Why Bryan Johnson did 5-MeO-DMT (12:56) What brain scans actually show (18:36) Psycho… Thanks to our partners for making this happen The Pod by Eight Sleep cools your bed to 55°F and uses Autopilot to optimize your sleep, all night. Use code ALLIN for up to $350 off.
    • EN Highlights:
      • (0:00) David Friedberg intros Bryan Johnson (0:54) Why Bryan Johnson did 5-MeO-DMT (12:56) What brain scans actually show (18:36) Psycho…
      • Thanks to our partner for making this happen
      • The Pod by Eight Sleep cools your bed to 55°F and uses Autopilot to optimize your sleep, all night
      • Use code ALLIN at https://www.eightsleep.com/allin for up to $350 off. Follow Bryan:

Lenny’s Podcast （A_full） Link to heading

  • 🎙️ This week on How I AI: How Stripe built “minions”—AI coding agents that ship 1,300 PRs per week + How to turn Claude Code into your personal life operating system

    • Release Time: 2026-03-30 23:03 Beijing Time
    • Abstract: In this episode, Stripe engineer Steve Kaliski provides a detailed analysis of how they built “Minions”—an AI programmer collective that can deliver approximately 1,300 Pull Requests per week, often triggered by just an emoji in Slack.
    • EN Highlights:
      • ,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F361d81ef-7faf-4d8e-8028-5d5e03432a9a_2329x551.pn…
      • ,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https…

  • How to turn Claude Code into your personal life operating system | Hilary Gridley

    • Publication Time: 2026-03-30 20:04 Beijing Time
    • Abstract: - Hilary Gridley is an entrepreneur, former product lead, and new mom, who previously discussed AI for managers on a podcast.
      • She returns to share how she has transformed her approach to personal productivity using Claude Code as her primary tool for managing both professional work and life.
      • Hilary showcases her “anti-system system”—a philosophy that prioritizes simplicity over complex setups, allowing the AI to learn preferences through observation rather than pre-configuration.
      • Listen or watch on YouTube, Spotify, or Apple Podcasts. What you will learn: - How to use a simple iPhone back-tap shortcut that requires zero app switching to instantly capture to-dos - The “10x Impact Framework” for deciding which tasks to automate versus those requiring human effort - How to use Claude Code’s observation features to build a preference file that improves over time without manual setup - Why the “yappers API” (talking about what you’re doing as you work) eliminates the need for complex OAuth integrations - A workflow for breaking down daunting tasks into 10-minute first steps that actually get done - How to create Claude skills by simply describing the problem instead of writing code or following tutorials - A technique for using “recording mode” to demonstrate workflows without exposing personal information. Brought to you by: WorkOS—Get your app enterprise-ready in minutes. Lovable—Build apps by simply chatting with an AI. In this episode, we cover: (00:00) Intro to Hilary Gridley (02:43) The opportunity cost of time as a new mom and entrepreneur (07:11) The anti-system philosophy (08:05) Demo: Planning your day with Claude Code (10:00) Setting up a simple iPhone task capture shortcut (11:48) How Claude automatically organizes reminders and learns preferences (16:19) Breaking down daunting tasks into manageable first steps (23:40) The yappers API: Talking to Claude instead of building integrations (25:28) Daily logging and observation mode (27:45) Quick summary (30:50) The power of screenshots (32:55) The 10x Impact Framework for automation decisions (37:51) Applying the framework to different career stages (39:29) Building a “skill” for “recording” anonymous demos (44:11) Building a return tracking skill from scratch (48:31) Building muscle memory for using AI tools (50:18) Where to find Hilary. Reference Tools: • Claude Code: • Obsidian: • iPhone Shortcuts: • Cursor: Other References: • Figma file of Hilary’s demo: Where to find Hilary Gridley: Substack: Website: Where to find Claire Vo: ChatPRD: Website: LinkedIn: X: Production and marketing.
      • Watch Now | 🎙️ How a new mom and entrepreneur automates her life management with Claude Code—all without complex setups. Hilary Gridley is an entrepreneur, formerly in product… She returns to share how she has transformed her approach to personal productivity using Claude Code as her primary tool for managing both professional work and life… Hilary showcases her “anti-system system”—a philosophy that prioritizes simplicity over complex setups, allowing the AI to learn preferences through observation… i=1000758198579">Apple Podcasts What you will learn: - How to use a simple iPhone back-tap feature to instantly capture to-dos… How to turn Claude Code into your personal life operating system | Hilary Gridley.
    • EN Highlights:
      • Hilary Gridley is an entrepreneur, former product…
      • She returns to share how she’s transformed her approach to personal productivity using Claude Code as her primary tool for managing both professional work and l…

  • Hilary demonstrates her “anti-system system”—a philosophy that prioritizes simplicity over complex setup, allowing AI to learn preferences through observa… What you’ll learn:

  • How to capture to-dos instantly using a simple iPhone back-tap…

  • From skeptic to true believer: How OpenClaw changed my life | Claire Vo

    • Posted: 2026-03-29 20:31 Beijing time
    • Summary: - Claire Vo is the host of our sister podcast, “How I AI,” a former product lead and engineer, and the founder of an AI startup called ChatPRD.
      • Claire now runs her business, podcast, and family life with the help of 9 OpenClaw agents running on multiple Mac Minis and old laptops.
      • In this episode, Claire shares her journey from an OpenClaw skeptic (it deleted her family calendar the first time she tried it) to a true believer, and provides a masterclass on using AI agents in real life.
      • The exact step-by-step process to install and set up OpenClaw (it’s easier than you think) 2.
      • How to avoid the biggest OpenClaw mistakes (don’t install it on your main computer) 3.
    • EN Key points:
      • Claire Vo is the host of our sister podcast, “How I AI,” a former product executive and engineer, and founder of an AI startup called ChatPR…
      • Claire now runs her business, podcast, and family life with the help of nine OpenClaw agents running on multiple Mac Minis and old laptops
      • In this episode, Claire shares her journey from OpenClaw skeptic (it deleted her family calendar the first time she tried it) to true believer, and gives a mast…
      • The exact step-by-step process to install and set up OpenClaw (it’s easier than you think) 2

  • 🧠 Community Wisdom: When AI velocity outpaces your product strategy, when your estimates keep slipping, one day in San Francisco, pairing Claude Code with Codex, and more

    • Posted: 2026-03-29 01:01 Beijing time
    • Summary: - 👋 Hello, and welcome to this week’s ✨ Community Wisdom ✨ a subscriber-only email, sent every…
      • ,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa1ca8352-9571-4b81-b041-52374176f4fc_2912x1456.p….
      • ,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa1ca8352-9571-4b81-b041-52374176f….
    • EN Key points:

  • 👋 Hello and welcome to this week’s edition of ✨ Community Wisdom ✨ a subscriber-only email, delivered every…

  • ,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa1ca8352-9571-4b81-b041-52374176f4fc_2912x1456.p…

  • ,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa1ca8352-9571-4b81-b041-52374176f…

  • How Stripe built “minions”—AI coding agents that ship 1,300 PRs weekly from Slack reactions | Steve Kaliski (Stripe engineer)

    • Published: 2026-03-25 20:03 Beijing time
    • Summary: - Steve Kaliski is a software engineer at Stripe where he has spent the last six and a half years building developer tools and payment infrastructure.
      • He was part of the team that created “minions,” Stripe’s internal AI coding agents that now send about 1,300 pull requests per week with minimal human intervention outside of code review.
      • In this episode, Steve demonstrates how Stripe engineers activate development work from Slack, utilize cloud-based development environments to enable parallel agent workflows, and shows machine-to-machine payments where AI agents transact autonomously with third-party services.
      • Listen or watch on YouTube, Spotify, or Apple Podcasts. What you will learn: - How Stripe’s “minions” write 1,300 pull requests per week with minimal human intervention - Why a good human developer experience creates better outcomes for AI agents - The key role of cloud development environments in unlocking AI-driven engineering velocity - Machine payment protocols that allow AI agents to spend money to complete tasks - Code review strategies for handling thousands of PRs written by agents - Why non-engineers at Stripe are starting to use Minions to ship code - The future of software businesses built primarily for agent consumers. Brought to you by: Optimizely — The AI agent orchestration platform for marketing and digital teams. Rippling — Stop wasting time on administrative tasks and build your startup faster. In this episode, we will cover: ( 00:00 ) Introduction to Steve ( 02:39 ) Stripe’s Minions and their impact across Stripe ( 04:42 ) Why activation energy is more important than execution ( 05:44 ) What are minions?

  • Technical Architecture (06:52) Demo: Activating Minions in Slack With an Emoji (09:04) Why a Great Developer Experience Is Good for Humans and Agents (11:22) Walking Through the Agent Loop and System Prompts (13:42) Why Stripe Chose Goose as its Agent Tool (16:00) The Role of Stripe’s Developer Productivity Team (17:15) Why Cloud Environments Unlock Multithreaded AI Engineering (21:14) One-Shot Prompt: From Slack to a Delivered PR (22:04) How Stripe Handles Code Review for 1,300 AI-Written PRs a Week (23:44) Non-Engineers Use Minions Across the Company (24:53) Demo: Curating a Birthday Party and Machine Payments With Claude (32:15) Quick Recap (35:08) The Future of Ephemeral, API-First Businesses for Agents (36:36) Lightning Round and Final Thoughts Reference Tools: • Goose (AI agent tool): • Claude Code: • Cursor: • VS Code: • Slack: • Browserbase: • Parallel AI: • PostalForm: • Stripe Climate: Other References: • Stripe Machine Payments: • Blue-Green Deployment: • Git Worktrees: Where to find Steve Kaliski: Twitter: LinkedIn: Where to find Claire Vo: ChatPRD: Website: LinkedIn: X: Production and marketing.

    • EN Key points:
      • Steve Kaliski is a software engineer at Stripe wh…
      • He’s part of the team that created “minions”—Stripe’s internal AI coding agents, which now ship approximately 1,300 pull requests…
      • In this episode, Steve demonstrates how Stripe engineers activate development work from Slack and leverage cloud-based development environments for parallel age… What you’ll learn:

  • How Stripe’s “minions” write 1,300 pull requ…

Stratechery by Ben Thompson (A_full) Link to heading

  • Spring Break

    • Publication Time: 2026-03-30 18:00 Beijing Time
    • Summary: - Stratechery is on a bit of a disjointed Spring Break, as my usual week off will be spread out:.
        • There will be no Update on Thursday, March 19.
        • There will be no Update on Monday and Tuesday, March 23-24; there will be an Update and Interview on Wednesday and Thursday, March 25-26.
        • There will be no Update on Monday, March 30.
      • All other Stratechery Plus content, including my podcast, will proceed as scheduled.
    • EN Key points:
      • Stratechery is on a bit of a disjointed Spring Break, as my usual week off will be spread out:
        • There will be no Update on Thursday, March 19
        • There will be no Update on Monday and Tuesday, March 23–24; there will be an Update and Interview on Wednesday and Thursday, March 25–26

  • 2026.13: So Long to Sora

    • Publication Time: 2026-03-28 01:00 Beijing Time
    • Summary: - Welcome back to This Week in Stratechery!
      • As a reminder, every week on Friday, we send out an overview of the content in the Stratechery bundle; highlighted links are free for everyone.

  • Additionally, you have full control over the content we send you.

  • In that regard, here are some of our favorites from this week.

  • Sora, 2025-2026.** AI Sam came, AI Sam saw, AI Sam stole those GPUs.

    • EN Highlights:
      • Welcome back to This Week in Stratechery
      • As a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; hig…

  • An Interview with Arm CEO Rene Haas About Selling Chips

    • Published: 2026-03-26 18:06 Beijing Time
    • Summary: - Listen to this article: Good morning, this week’s Stratechery interview is with Arm CEO Rene Haas, who I previously spoke with in January 2024. He recently made a major announcement during Arm’s first independent keynote: the long-standing IP licensing company is undergoing a significant shift in its business model and is selling its own chips for the first time.
      • In this interview, we delve into this decision, including the keynote’s metadata, Arm’s history, and how the company has evolved, especially under Haas’s leadership.
      • We then explore why CPUs are important for AI and how Arm’s CPUs compare to those from Nvidia, x86, and other custom Arm chips.
      • Finally, we discuss the risks Arm faces, including supply chain limitations and the changes the company needs to make to support this new direction.
      • As a reminder, all Stratechery content (including interviews) is available in podcast form; click the link at the top of this email to add Stratechery to your podcast player.
    • EN Highlights:
      • Listen to this post:

  • Arm Launches Own CPU, Arm’s Motivation, Constraints and Systems

    • Published: 2026-03-25 18:00 Beijing Time
    • Summary: - Arm is selling its own chips, not just licensing IP.
      • This is a significant change compared to Arm’s history, but it’s not surprising given the evolution of computing.
      • Arm is selling its own chips, not just licensing IP. This is a big change compared to Arm’s history, but it’s not surprising given how computing is evolving.
      • Arm launches its own CPU, Arm’s motivation, constraints, and systems.
    • EN Highlights:
      • Arm is selling its own chips, not just licensing IP
      • It’s a big change compared to Arm’s history, but not surprising given how computing is evolving.

OpenAI Blog (A_full) Link to heading

  • Helping disaster response teams turn AI into action across Asia

    • Published: 2026-03-30 06:15 Beijing Time
    • Summary: Today in Bangkok, we joined the Gates Foundation, the Asian Disaster Preparedness Center, and DataKind to bring together 50 disaster management experts from South and Southeast Asia to discuss how AI can accelerate the response efficiency of governments and non-profit organizations in times of crisis.
    • EN Highlights:
      • AI for Disaster Response in Asia: OpenAI Workshop with Gates Foundation

  • STADLER reshapes knowledge work at a 230-year-old company

    • Published: 2026-03-28 06:00 Beijing Time
    • Summary: - STADLER is a family-owned company with over 230 years of history, specializing in providing automated waste sorting plants for the global recycling industry.
      • With over 650 employees worldwide, the company plays a key role in helping countries advance their sustainable development and circular economy goals.

  • Under the leadership of co-CEO Julia Stadler, the company has adopted a forward-looking approach to modernization—embedding AI as a core productivity layer into daily work.

  • Since 2023, STADLER has followed a clear principle: every employee who works on a computer should use AI to improve speed, quality, and collaboration.

    • EN Key points:
      • Learn how STADLER uses ChatGPT to transform knowledge work, saving time and accelerating productivity across 650 employees.

  • Inside our approach to the Model Spec

    • Published: 2026-03-25 18:00 Beijing Time
    • Summary: This article explores how OpenAI’s Model Spec serves as a public framework for guiding model behavior, balancing safety, user freedom, and accountability as AI systems continue to evolve.
    • EN Key points:
      • Learn how OpenAI’s Model Spec serves as a public framework for model behavior, balancing safety, user freedom, and accountability as AI systems advance.

  • Introducing the OpenAI Safety Bug Bounty program

    • Published: 2026-03-25 08:00 Beijing Time
    • Summary: - OpenAI is launching a safety bug bounty program to identify AI abuse and safety risks, including agentic vulnerabilities, prompt injection, and data exfiltration.
      • This article from the OpenAI blog explains how the introduction of the OpenAI Safety Bug Bounty program is shaping the broader AI and infrastructure landscape.
      • Following the launch of the OpenAI Safety Bug Bounty program, it also has practical implications for founders, operators, and investors.
    • EN Key points:
      • OpenAI launches a Safety Bug Bounty program to identify AI abuse and safety risks, including agentic vulnerabilities, prompt injection, and data exfiltration.

Google DeepMind Blog (A_full) Link to heading

  • Gemini 3.1 Flash Live: Making audio AI more natural and reliable

    • Published: 2026-03-26 23:23 Beijing Time
    • Summary: - Our latest voice model has improved precision and lower latency to make voice interactions more fluid, natural, and precise.
      • This article from the Google DeepMind blog explains how “Gemini 3.1 Flash Live: Making audio AI more natural and reliable” is shaping the broader AI and infrastructure landscape.
      • It also has practical implications for the founders, operators, and investors of “Gemini 3.1 Flash Live: Making audio AI more natural and reliable.”
    • EN Key points:
      • Our latest voice model has improved precision and lower latency to make voice interactions more fluid, natural and precise.

  • Protecting people from harmful manipulation

    • Published: 2026-03-26 00:46 Beijing Time
    • Summary: - Google DeepMind is researching the risks of harmful manipulation by AI in areas like finance and health, leading to the development of new safety measures.
      • This article from the Google DeepMind blog explains how “Protecting people from harmful manipulation” is shaping the broader AI and infrastructure landscape.
      • Following “Protecting people from harmful manipulation,” it also has practical implications for founders, operators, and investors.
    • EN Key points:

  • Google DeepMind researches AI’s harmful manipulation risks across areas like finance and health, leading to new safety measures.

  • Lyria 3 Pro: Create longer tracks in more

    • Release Time: 2026-03-26 00:01 Beijing Time
    • Summary: - Introducing Lyria 3 Pro, which unlocks longer tracks with structural awareness.
      • We are also bringing Lyria to more Google products and surfaces.
      • Introducing Lyria 3 Pro, which unlocks longer tracks with structural awareness. We are also bringing Lyria to more Google products and surfaces.
    • EN Key Points:
      • Introducing Lyria 3 Pro, which unlocks longer tracks with structural awareness
      • We’re also bringing Lyria to more Google products and surfaces.

Two Minute Papers (B_intro+search) Link to heading