AI Builders Digest — 2026-05-31
X / TWITTER
Josh Woodward, VP at Google / Google Labs / Gemini App / Google AI Studio
Josh Woodward highlighted two lightweight Gemini-era creation patterns: visual transformation, framed as “turn your car into a Lamborghini,” and multilingual content creation becoming “ridiculously easy.” The signal is less the individual demo and more the direction: Google is pushing AI Studio and Gemini toward consumer-grade media manipulation and frictionless language localization.
https://x.com/joshwoodward/status/2060443095527989413
https://x.com/joshwoodward/status/2060443093825094091
Josh Woodward 展示了两个 Gemini 时代的轻量创作场景：把普通汽车视觉转换成 Lamborghini，以及让多语言内容创作变得“ridiculously easy”。真正的信号不只是这两个 demo，而是 Google 正在把 AI Studio 和 Gemini 推向更低门槛的媒体处理与本地化创作。
https://x.com/joshwoodward/status/2060443095527989413
https://x.com/joshwoodward/status/2060443093825094091
Boris Cherny, Claude Code at Anthropic
Boris Cherny pointed to Salesforce’s Claude Code rollout as evidence that the real gains from AI come from redesigning work, not merely accelerating existing workflows. Salesforce reportedly compressed a 231-day migration into 13 days, shipped one PR with 21 endpoints at 100% test coverage, and still saw total incidents drop 5%, because security guardrails and quality standards were built into the agentic workflow itself.
https://x.com/bcherny/status/2060390852619272526
https://x.com/bcherny/status/2060390853835726946
https://x.com/bcherny/status/2060390855383400729
Boris Cherny 用 Salesforce 的 Claude Code 落地案例说明：AI 真正的收益不是把原有流程加速，而是重构工作方式。Salesforce 据称把一个原本估算 231 天的迁移任务压缩到 13 天完成，一个 PR 交付 21 个 endpoint 且测试覆盖率 100%，同时总事故数还下降 5%，关键在于他们把安全 guardrails 和质量标准直接嵌进了 agentic workflow。
https://x.com/bcherny/status/2060390852619272526
https://x.com/bcherny/status/2060390853835726946
https://x.com/bcherny/status/2060390855383400729
Thibault Sottiaux, Codex & ChatGPT at OpenAI
Thibault Sottiaux teased a strong Codex adoption metric, saying a number on the Codex dashboard made him happy and that “we are still early.” He also asked whether people still trust benchmarks or mostly rely on friends when trying a new model, which is a useful product-market signal: model adoption is increasingly driven by social proof and workflow trust, not leaderboard position alone.
https://x.com/thsottiaux/status/2060565265906290786
https://x.com/thsottiaux/status/2060563528596287874
https://x.com/thsottiaux/status/2060529970523603099
Thibault Sottiaux 暗示 Codex 的采用数据很强，说 Codex dashboard 上的一个数字让他很开心，并强调“we are still early”。他还抛出一个有意思的问题：大家现在还信 benchmark，还是更听朋友推荐来尝试新模型？这说明模型采用越来越依赖社会证明和工作流信任，而不只是排行榜名次。
https://x.com/thsottiaux/status/2060565265906290786
https://x.com/thsottiaux/status/2060563528596287874
https://x.com/thsottiaux/status/2060529970523603099
Aaron Levie, CEO at Box
Aaron Levie argued that a company spending $500M to build its own version of the application layer is actually bullish for software. His point: if large companies are willing to spend that much to recreate app-layer capabilities internally, it validates how much durable value still lives above the model and infrastructure layers.
https://x.com/levie/status/2060525104384418271
Aaron Levie 认为，一家公司花 5 亿美元自建应用层版本，反而是对软件行业的利好。核心逻辑是：如果大公司愿意投入这么多钱来内部重建 app-layer 能力，说明模型和基础设施之上的应用层仍然有非常持久的价值。
https://x.com/levie/status/2060525104384418271
Garry Tan, President & CEO at Y Combinator
Garry Tan gave founders a blunt fundraising reminder: “Money is not the fire. Money is gasoline you pour on a fire that already exists.” His advice is to stop treating funding as the missing ingredient and first prove that people actually want the thing enough to create organic traction.
https://x.com/garrytan/status/2060600088079356292
Garry Tan 给创业者的提醒很直接：“钱不是火，钱是倒在已经存在的火上的汽油。”他的建议是，不要把融资当作缺失的关键条件，先证明用户真的想要这个东西，先把第一团火点起来。
https://x.com/garrytan/status/2060600088079356292
Nikunj Kothari, Partner at FPV Ventures
Nikunj Kothari shared enthusiasm for a founder going through the Y Combinator interview process and the upcoming batch. The useful signal is investor attention concentrating again around fresh YC cohorts, especially as AI-native startups continue to compress the path from prototype to fundable company.
https://x.com/nikunj/status/2060580468781953169
Nikunj Kothari 表达了对一位创始人经历 Y Combinator 面试以及下一批 YC 项目的期待。这里的信号是，投资人注意力继续集中在新一批 YC 公司上，尤其是在 AI-native startup 把从原型到可融资公司的路径不断压缩之后。
https://x.com/nikunj/status/2060580468781953169
Peter Steinberger, OpenClaw and OpenAI
Peter Steinberger celebrated Vince joining OpenClaw, saying few people understand “the new ways” software is built. The important read is that OpenClaw is positioning itself around a new software-production paradigm, not just another coding assistant UI.
https://x.com/steipete/status/2060306947035832628
Peter Steinberger 庆祝 Vince 加入 OpenClaw，并表示很少有人真正理解软件正在以“new ways”被构建。更重要的信号是，OpenClaw 正在围绕一种新的软件生产范式定位自己，而不只是又一个 coding assistant UI。
https://x.com/steipete/status/2060306947035832628
PODCASTS
No Priors: Building an AI Guardian for Enterprise with Onyx Security CEO Maxim Bar Kogan
The takeaway: enterprise AI security is shifting from “what did employees paste into ChatGPT?” to “who is supervising the autonomous agents now acting with employee-level permissions?” Maxim Bar Kogan, co-founder and CEO of Onyx Security, argues that the center of gravity has moved to agent actions: coding agents, low-code automations, and first-party enterprise agents are generating exponentially more actions than humans can review. His company’s bet is a “secure AI control plane” that trains small specialized models to decide when a smarter oversight agent needs to inspect a risky action.
The counterintuitive point is that old security primitives break because we want agents to use our permissions. Identity tools can restrict static software, but a useful coding agent needs broad access and context. A proxy can see data, but it still may not know why the agent is about to delete a database. Kogan’s line is sharp: “You don’t want to spend too much intelligence where you don’t have to, and you want to spend a lot of intelligence… in situations where there’s high risk.” That is the chess-clock architecture for AI security: cheap intuition most of the time, expensive reasoning only at critical moves.
He also sees an opening for independent AI-security vendors because enterprises may not trust model labs to audit themselves or train on sensitive historical agent data. If AI vendors become huge, the control layer around them can become huge too.
https://www.youtube.com/watch?v=QDsbFLEt9ro
核心判断：企业 AI 安全正在从“员工把什么贴进了 ChatGPT？”转向“现在这些拥有员工级权限的 autonomous agents，到底由谁监督？”Onyx Security 联合创始人兼 CEO Maxim Bar Kogan 认为，风险重心已经转移到 agent actions：coding agents、low-code automations、企业自建 agents 正在产生指数级增长的动作，人类已经不可能逐条审核。他们押注的是一个“secure AI control plane”，用小型专用模型判断哪些高风险动作需要交给更强的监督 agent 进一步检查。
反直觉点在于，传统安全原语在这里会失效，因为我们本来就希望 agent 使用我们的权限。Identity 工具能限制静态软件，但一个真正有用的 coding agent 需要宽权限和上下文。Proxy 可以看到数据流，却未必知道 agent 为什么突然要删数据库。Kogan 有一句话很准：“You don’t want to spend too much intelligence where you don’t have to, and you want to spend a lot of intelligence… in situations where there’s high risk.” 这就是 AI security 的棋钟架构：大多数时候用便宜直觉，关键节点才调用昂贵推理。
他还认为，独立 AI security vendor 有结构性机会，因为企业未必愿意让模型实验室自己审计自己，也未必愿意把敏感的历史 agent 行为数据交给模型公司训练。如果 AI vendor 会变得巨大，那么围绕它们的控制层也可能变得巨大。
https://www.youtube.com/watch?v=QDsbFLEt9ro
Generated through the Follow Builders skill: https://github.com/zarazhangrui/follow-builders