2026-04-18
AI Builders Digest — 2026-04-18
X / TWITTER
Aaron Levie (CEO Box) on agents and the future of software
Box CEO Aaron Levie thinks through the agent economics of software platforms, and lands on a conclusion that cuts against the "SaaSpocalypse" narrative: agents using software 100X more than people isn't zero-sum — it dramatically expands the addressable market for enterprise platforms. Software is going headless because agents consume via API at a scale people never could. The thread gets specific: instead of reviewing one contract at a time, an agent reviews all of them; instead of running one marketing campaign, you run 10X; instead of being bottlenecked by human steps in onboarding, agents accelerate. Levie's framing of the new business model is clean: seats for the people, consumption for the agents. He's bullish on systems of record being especially well-positioned because their data can be leveraged 100X more via agents. The take is contrarian and has teeth — 1576 likes, 161 RTs. Link
Box CEO Aaron Levie 对软件平台与 Agent 的经济学逻辑进行了深入思考，他的结论直接反驳了"SaaS 消亡论"：Agent 使用软件的频率是人类的上百倍，这不是零和游戏，而是极大扩展了企业平台的可寻址市场规模。软件走向 headless 是因为 Agent 能以人类从未达到的规模通过 API 消费。Levie 的论述很具体：Agent 不是逐个审查合同，而是批量审查所有合同；不是运行一个营销活动，而是跑 10 倍数量；不是在入职流程中被人工步骤卡住，而是全面加速。他对新商业模式的概括很精准：人按席位付费，Agent 按消耗量付费。Levie 看好系统 of record（类似 Box）尤其处于有利位置，因为其数据可被 Agent 放大百倍使用。这个观点有锋芒，有反驳 — 获得了 1576 个赞和 161 次转发。 链接
Kevin Weil (VP Science at OpenAI) departing after two years
Kevin Weil announced his last day at OpenAI, as OpenAI for Science is being decentralized into other research teams. He joined two years ago as Chief Product Officer before moving to the research side to start the OpenAI for Science initiative. Wishing the team well, he says the work continues — and he's moving on to next. Link
Kevin Weil（OpenAI VP of Science）宣布了他在 OpenAI 的最后一天，OpenAI for Science 项目正在拆分并入其他研究团队。两年前他以首席产品官的身份加入，后转至研究团队并创办了 OpenAI for Science 项目。他表示加速科学将是迈向 AGI 过程中最令人惊叹的成果之一，并对 Sam Altman、Mark Chen 等团队成员表达了祝福。 链接
Alex Albert (Research at AnthropicAI) and Matt Turck (FirstMark Capital) on Claude Opus 4.7
Anthropic researcher Alex Albert confirmed that bugs in Opus 4.7 that users encountered on day one have been fixed. Matt Turck noted Anthropic is on a "generational run" with Claude Code, Skills, Cowork, and Design all shipping in close succession — a rare moment of Anthropic moving at startup speed. Link Alex Link Matt
Anthropic 研究员 Alex Albert 确认，Opus 4.7 上线首日用户遇到的 bug 已全部修复。Matt Turck 则指出，Anthropic 正在经历一个"划时代"的密集发布期：Claude Code、Skills、Cowork、Design 接连推出，这种速度在 Anthropic 身上极为罕见。 链接 Alex 链接 Matt
Google Labs announces Flow Music
Google Labs rebranded ProducerAI to Google Flow Music — a standalone site for creating, sharing, and remixing original music using natural language prompts. The tool transforms imagination into fully produced tracks. Flow now covers images and videos, while Flow Music handles songs and playlists. Link
Google Labs 将 ProducerAI 更名为 Google Flow Music — 一个独立站点，用自然语言提示词来创作、分享和混音原创音乐。Flow 家族现在涵盖图片和视频（Flow），以及歌曲和播放列表（Flow Music）。Josh Woodward（Google Labs VP）同步发帖，并展示了新的 remix 功能。 链接
Ryo Lu (Design at Cursor) on what makes a best-in-class agent
Cursor designer Ryo Lu sums it up: the best agent combines the best harness, the best models, and runs anywhere. Cursor is positioning itself at the intersection of all three. Link
Cursor 设计师 Ryo Lu 给出了一个精炼的公式：最优秀的 Agent = 最优秀的 harness + 最优秀的模型 + 能在任何地方运行。Cursor 正在将自己定位在这三者的交汇处。 链接
Garry Tan (President & CEO Y Combinator) on OpenClaw
Garry Tan 开了个玩笑：OpenClaw 就像法拉利，而 Hermes Agent 大概是……本田雅阁。👀 他还吐槽自己花了几个小时眼睁睁看着 OpenClaw 对自己的配置文件进行"脑外科手术"然后把自己弄死了。Y Combinator 掌门人对 OpenClaw 的评价既调侃又带着一种独特的认可感。 链接
Nikunj Kothari (Partner at FPV Ventures) on the Fintool acquisition
FPV Ventures 合伙人 Nikunj Kothari 以天使投资人的身份支持了 Fintool，现在 Microsoft 将其收购。他指出，Fintool 在大模型出现之前就已经在做复杂的推理任务——这是一个产品理念超前于时代的案例。Microsoft 获得了一支非常出色的产品团队。 链接
PODCASTS
No Priors Podcast — Scaling Global Organizations in the Age of AI with ServiceNow CEO Bill McDermott
Bill McDermott runs ServiceNow at $13B+ revenue growing 20% annually. This conversation is less about product and more about operating philosophy — which is exactly why it's useful.
Bill McDermott 掌舵的 ServiceNow 年收入 130 亿美元以上，年增长率 20%。这场对话的焦点不在产品，而在经营哲学——这恰恰是它最有价值的地方。
His core argument cuts through the AI hype: language models are excellent at thinking, but platforms like ServiceNow are what actually acts. A language model will tell you the three steps to solve a compensation issue — in milliseconds. But it won't close the case. Closing the case requires going in and out of HR, Finance, Legal, Compliance, and multiple databases. That's what a workflow platform does. McDermott calls it the AI control tower — integrating all the hyperscalers, all the language models, and all the systems of record.
他的核心论点戳破了 AI 炒作的泡沫：语言模型擅长的是思考，而 ServiceNow 这样的平台才是真正行动的那个。语言模型会告诉你解决薪酬问题的三个步骤——毫秒级完成。但它无法结案。要结案需要进入和退出 HR、财务、法务、合规等多个部门和数据库。这正是工作流平台做的事。McDermott 称其为 AI 控制塔——整合所有超大规模云商、所有语言模型和所有记录系统。
On the "SaaSpocalypse" theory that AI will eat software: McDermott did the math. Replacing a simple ServiceNow application with a language model costs 10X more — when you factor in the replacement cost, the human capital cost of building it yourself, and the ongoing token costs. "People that run businesses understand that people make mistakes. They never will forgive software for making a mistake."
关于"SaaS 消亡论"——AI 会吃掉软件——McDermott 算了一笔账：用语言模型替代一个简单的 ServiceNow 应用，成本要高出 10 倍——这还没算重置成本、自己构建的人力资本成本，以及持续消耗的 token 费用。"做生意的人都明白，人会犯错。但他们永远不会原谅软件犯错。"
He has a vivid sense of the agent future: 2.2 billion agents entering the workforce in the next few years. ServiceNow's own customer service is already 90% agent-managed. Net new headcount growth will be dramatically reduced — not because the company is shrinking, but because agents handle the scale.
他对 Agent 未来有清醒的认知：未来几年将有 22 亿个 Agent 进入劳动力市场。ServiceNow 自己的客服已经有 90% 由 Agent 处理。未来净新增员工数将大幅减少——不是因为公司在萎缩，而是因为 Agent 在承接规模。
His broader philosophy is personal and grounded. He started at 16 buying a deli with a $5,500 note — miss a payment and they take everything. The lesson he's carried through SAP and now ServiceNow: the customer, and the customer alone, determines whether you win or lose.
他的经营哲学根基在个人经历里。他 16 岁时用 5500 美元的期票买下一家熟食店——错过一次付款就血本无归。这条经验他从 SAP 一直带到 ServiceNow：客户，只有客户，决定你的输赢。
"People that run businesses understand that people make mistakes. They never will forgive software for making a mistake."
"做生意的人都明白，人会犯错。但他们永远不会原谅软件犯错。"
The quote that best captures McDermott's view of AI: "AI thinks, but workflow acts." Not as poetic as some of the Valley's louder pronouncements, but probably more accurate about what enterprise buyers actually need.
最能概括 McDermott 对 AI 看法的金句是："AI 负责思考，工作流负责行动。" 没有硅谷某些高调宣言那么诗意，但可能更准确地反映了企业买家真正需要什么。
Generated through the Follow Builders skill: https://github.com/zarazhangrui/follow-builders