🌐 双语
Archive

AI Builders
Digest

2026-11-15 13 builders · 26 tweets · 1 podcasts · 0 blogs

🔥 热点话题

JEV 创始人:可靠性优先,拒绝 benchmark 疲劳,机器原生才是未来JEV Creator: Reliability Over Benchmarks, Machine-Native AI Is the Future

The Takeaway: AI 的真正潜力不在于更会讨好人的聊天机器人或刷榜模型,而在于打造可靠、机器原生、可组合的决策系统,让软件真正掌控智能。

JEV(TypeSave)CEO Jago 曾在 OpenAI 工作,如今以“系统一”决策模型冲击行业。他坚持认为,当前 LLM 的大部分调用最终应服务计算机而非人类——风格和散文只是极少数场景,机器需要的是高基数枚举和位级操作。JEV 的目标正是从这一认知核心反向工程。他坦承模型仍远未成熟,存在大量 jaggedness(能力锯齿),但故意公开不完美特性,让开发者自行判断可靠性需求。

Jago 对 benchmark 和“形状崇拜”极度警惕:新形状容易,真正有用且可靠的智能才难。他批评行业过度承诺、快速失望的循环已经让开发者对新型号产生疲劳,并希望未来更多协作——软件公司指出 jaggedness,模型方修复,共同解锁科幻用例。他对 coding agent 的“KV cache 暴政”尤其不满:当前设计几乎都围绕可追加状态,限制了路由、子代理共享上下文、语义压缩等更优雅架构。他畅想依赖类型的语义版本、组织级语义安全检查、脱离 KV cache 的多代理共享状态等。

对 Anthropic 等实验室,他既认可 coding agent 的价值,也怀疑部分估值建立在对 AI 能力边界的误解上——“让他们治癌”可能最终像“让他们吃蛋糕”。他估计 coding agent 对工程生产力的平均提升约 20-40%,且高度依赖场景与用户技能。安全问题上,他区分 ASI 与生物/网络风险,认为真正自主智能仍需重大突破,当前更多是工具增强而非失控代理。最终他呼吁开发者重新获得 agency,一起把 AI 变成可玩、可组合、真正赋能创造的力量。

“the let them cure cancer is going to be like Marie Antoinette’s, like, let them eat cake.”
The Takeaway: The real potential of AI lies not in more agreeable chatbots or benchmark-optimized models, but in reliable, machine-native, composable decision systems that let software truly own intelligence.

Jago, CEO of TypeSave and creator of JEV, previously at OpenAI, is pushing “system one” decision models into the industry. He argues that the vast majority of future LLM calls should serve computers rather than humans—style and prose are rare edge cases; machines need high-cardinality enums and bit-level operations. JEV works backward from that cognitive core. He is candid that the models are still far from ready, full of jaggedness, and deliberately ships imperfect properties so developers—not the lab—can decide the reliability profile they need.

He is deeply skeptical of benchmarks and “shape worship”: inventing a new shape is easy; making intelligence actually useful and reliable is hard. The industry’s cycle of over-promise and under-deliver has already created fatigue among developers toward new model releases. He wants far more collaboration—software companies calling out specific jaggedness, model teams fixing it, jointly unlocking sci-fi use cases. He is especially frustrated by the “tyranny of the KV cache” in coding agents: nearly every design decision is constrained by append-only state, blocking elegant routing, shared multi-agent context, semantic compaction, and more. He dreams of semantic dependent typing, organization-level semantic security checks, and multi-agent shared state freed from the cache.

On Anthropic and the labs, he acknowledges coding agents’ value while suspecting some valuations rest on misunderstandings of AI’s actual boundaries—“let them cure cancer” may end up like “let them eat cake.” He estimates coding agents deliver a 20-40% average productivity boost, highly dependent on use case and user skill. On safety he separates ASI discussions from bio/cyber risks, arguing true agency is still multiple major breakthroughs away; current systems are powerful tools, not runaway agents. He ends by urging developers to reclaim agency and co-create AI that is playable, composable, and genuinely generative.

“the let them cure cancer is going to be like Marie Antoinette’s, like, let them eat cake.”
查看原文 →

Aaron Levie:代理将让越来越多工作变成“代码形状”Aaron Levie: Agents Will Make More Work Become Coding-Shaped

Box CEO Aaron Levie 指出,AI 模型在编码、工具使用、计算机操作和数学等任务上必然持续快速进步,因此能够使用软件或从零生成代码的代理,其提升速度将远超其他知识工作领域。结果是,越来越多的工作会主动变成“代码形状”以适应代理擅长的方向。近期模型已经能生成动画视频、游戏、产品 3D 模型等。传统思维认为模型必须“无所不知”,但真正解锁价值的是“会用软件 + 深度领域判断”。一旦具备这一组合,几乎所有知识工作都会加速被代理渗透。
Box CEO Aaron Levie argues that AI models will keep getting better at coding, tool use, computer use, and math, so agents that use or generate software will improve faster than agents in other knowledge-work domains. Consequently more and more work will reshape itself into “coding-shaped” forms that play to agent strengths. Recent models already generate animated videos, games, and 3D product models. The old mental model that models must know everything is wrong; the real unlock is software fluency combined with domain judgment. With that combination, almost every domain of knowledge work accelerates.
查看原文 →

Amjad Masad:自主商业已至,我们进入 Murphy 世界Amjad Masad: Autonomous Businesses Are Here, Welcome to Murphy’s World

Replit CEO Amjad Masad 宣布“自主商业已至”,并指出技术本质上就是递归自我改进的,自主程度一直在上升,AI 只是加速器而非起点——Douglas Engelbart 早就看清这一点。同时他观察到我们已进入 Murphy 世界:任何可能出错的事都会出错。当数十亿事件同时发生、相互依赖并快速加速时,曾经罕见的故障变成了日常。
Replit CEO Amjad Masad declares that autonomous businesses have arrived and notes that technology has always been recursively self-improving; AI merely accelerates a process Douglas Engelbart already understood. He also observes that we have entered Murphy’s world: anything that can go wrong will go wrong. With billions of interdependent events accelerating, once-rare failures are now daily occurrences.
查看原文 →查看原文 →查看原文 →

🛠️ 开发者工具与技巧

Guillermo Rauch:UI SDK + Gemini 让纯 Web Components 瞬间可用Guillermo Rauch: UI SDK + Gemini Delivers Pure Web Components Instantly

Vercel CEO Guillermo Rauch 展示用两段 prompt 加 Gemini Argon 生成纯 Web Components(无框架)的体验。代理直接复用了 14 个 UI SDK 行为:窗口缩放、拖拽、菜单栏悬停切换、typeahead、工具栏焦点漫游、10 段 EQ 滑块、作用域热键、ARIA 以及 WebMCP 代理控制。他同时称赞 fx 上的子代理体验极佳,所有项目都采用“最强模型规划 + 快速模型子代理执行”的一句话指令模式。
Vercel CEO Guillermo Rauch demoed generating pure Web Components (no framework) with two prompts plus Gemini Argon. The agent reused 14 UI SDK behaviors out of the box: window resizing, drag, menubar hover-switching, typeahead, roving toolbar focus, 10-band EQ sliders, scoped hotkeys, ARIA, and WebMCP agent control. He also praised subagents on fx as extremely delightful; all his projects now plan with the smartest model and delegate to fast subagents via a one-line instruction.
查看原文 →查看原文 →

Josh Woodward:Gemini 接入 Google Wallet,个人财务洞察上线Josh Woodward: Gemini Connects Google Wallet for Personal Finance Insights

Google Labs / Gemini VP Josh Woodward 宣布 Gemini App 用户(先从美国开始)可连接 Google Wallet,获得即时消费图表、发现订阅涨价、根据去年实际支出制定节日预算等能力。
Google Labs / Gemini VP Josh Woodward announced that Gemini App users (starting in the USA) can now connect Google Wallet for instant spend charts, detection of sneaky subscription price hikes, and holiday budgets based on last year’s actual spending.
查看原文 →查看原文 →

Peter Yang:Fable 5.2 两个实用 prompt 审计 Claude Code 设置Peter Yang: Two Useful Fable 5.2 Prompts to Audit Claude Code Setups

AI 教程作者 Peter Yang 推荐两个 Fable 5.2 prompt:"/doctor prompt-audit" 审计整个 Claude Code 配置(Claude.md、skills、commands)并给出修复建议;"/claude-api prompt-audit [skill name]" 深度审计单个 skill。他希望 Claude Code 能在每次 Anthropic 发布新模型时自动建议运行这些审计。
AI tutorial creator Peter Yang shared two highly useful Fable 5.2 prompts: “/doctor prompt-audit” audits your full Claude Code setup (Claude.md, skills, commands) and proposes fixes; “/claude-api prompt-audit [skill name]” deeply audits one skill. He wishes Claude Code would auto-suggest running these every time Anthropic ships a new model.
查看原文 →

Dan Shipper:Every 前沿实验源码与 writeup 全部公开Dan Shipper: Every Frontier Experiments Now Public with Code and Writeups

Every CEO Dan Shipper 展示 Fable 帮他写邮件的实时过程,并强调团队把所有前沿 AI 实验(含原始代码和简短 writeup)发布在 frontier feed 上,方便其他 agent 直接读取,是获取灵感最快的方式。
Every CEO Dan Shipper showed a live look at Fable writing an email for him and highlighted that the team publishes all frontier AI experiments—including raw code and short writeups—on their frontier feed so other agents can read them directly. Fastest way to get inspired.
查看原文 →查看原文 →

Garry Tan:Aside Browser 是 AI agent 的巨大解锁Garry Tan: Aside Browser Is a Huge Unlock for Every AI Agent

Y Combinator 总裁兼 CEO Garry Tan 称赞 Aside Browser 对所有 AI agent 都是巨大解锁,并建议与 gbrain 一起使用。
Y Combinator President & CEO Garry Tan called Aside Browser a huge unlock for every AI agent and recommended pairing it with gbrain.
查看原文 →查看原文 →

Thibault Sottiaux:OpenAI Cyber 团队征集 Daybreak 反馈Thibault Sottiaux: OpenAI Cyber Team Wants Feedback on Daybreak

OpenAI Codex & ChatGPT 的 Thibault Sottiaux 代表 Cyber 团队公开征集对 Daybreak 的反馈:模型、codex 插件、蓝/红队访问、文档、红队与防御工厂等,哪里可以做得更好。
OpenAI Codex & ChatGPT’s Thibault Sottiaux asked the community for feedback on Daybreak from the Cyber team: models, codex plugins, blue/red access, documentation, red-teaming and defense factory—what can be improved.
查看原文 →

🌍 其他动态

Swyx:AI UX 仍严重次优,模型已便宜到可随意使用Swyx: AI UX Remains Wildly Suboptimal Despite Models Too Cheap to Meter

smol.ai / Cognition 相关建设者 Swyx 回应称,当前 AI UX 仍严重次优,考虑到模型已经便宜到几乎可随意计量,这一现状更显荒谬。
smol.ai / Cognition-affiliated builder Swyx noted that AI UX is still wildly suboptimal—especially given that models are already too cheap to meter.
查看原文 →

Matt Turck:NVIDIA 是开放权重 AI 最有动力的深度玩家Matt Turck: NVIDIA Is the Deep-Pocketed Player with Clear Skin in Open-Weights AI

FirstMark 合伙人 Matt Turck 认同 NVIDIA 是唯一既有深厚口袋又对开放权重 AI 有明确商业利益、并愿意下创始人式大赌注的玩家,并追问接下来会发生什么。
FirstMark partner Matt Turck agreed that NVIDIA is the one deep-pocketed player with a clear business interest in open-weights AI and the willingness to make founder-style big bets, then asked what happens next.
查看原文 →

Nikunj Kothari:2026 创业书架推荐Nikunj Kothari: 2026 Startup Bookshelf

FPV Ventures 合伙人 Nikunj Kothari 分享了 2026 创业书架书单,并表示期待本周的事故报告。
FPV Ventures partner Nikunj Kothari shared a 2026 Startup Bookshelf and noted he is eagerly looking forward to the incident reports from this week.
查看原文 →查看原文 →