Vercel CEO Says AI Assistant v0 Scaled to $4 Million ARR in 13 Months

Vercel CEO称AI助手v0在13个月内ARR达400万美元 随着人工智能将开发重心从编写代码转向人类的品味与方向指引，软件生产的边际成本正降至零。 Guillermo Rauch详细阐述了Vercel如何从一个专注于前端的雄心勃勃的框架，演变为支撑现代Web应用、零售巨头以及ChatGPT等顶尖AI平台的关键基础设施。对话重点探讨了Vercel近期在v0推动下的爆发式增长——这是一款能将普通英语转化为可用代码的AI助手，并在数周内实现了数百万美元的ARR。Rauch还分享了他独特的管理理念，包括维持550人的精简团队规模、以CEO身份保持深厚技术造诣，以及在软件开发演变为由人类品味主导的高度大众化机器能力的未来中从容应对。 Vercel在高流量互联网事件中扮演着隐形基础设施骨架的角色，包括2024年美国大选（《华盛顿邮报》、Perplexity、Polymarket）以及大规模黑色星期五零售业务。 (1:03) Next.js已成为AI浪潮中无可争议的黄金标准前端框架，为超过130万活跃开发者提供服务，支撑着ChatGPT、Claude、Sora和Midjourney的用户界面。 (1:03) Vercel创立的初衷是将Kubernetes和React等强大原始的开源工具打通，为开发者提供自动化、无缝且快速的部署工作流程。 (1:56) Vercel特意押注动态的云端渲染架构而非静态应用，因为他们认识到移动端浏览需要将繁重的计算负荷从用户设备转移出去。 (4:20) 该平台为每个独立的git分支自动生成安全、临时的预览URL，彻底改变了企业协作模式，并消除了本地化工程瓶颈。 (12:08)

Vercel co-founder and CEO Guillermo Rauch argues that abstracting cloud complexity into dynamic, application-centric frontend infrastructure—and increasingly natural-language AI generation—allows small teams to build, scale, and secure modern web applications without managing raw cloud orchestration. Inverting Cloud Infrastructure Through Dynamic Frameworks (36:35) Rauch recalls that around 2015, utilizing modern cloud computing and open-source systems like Kubernetes and React required prohibitive manual configuration for average enterprises. In response, Vercel narrowed its strategic focus to dynamic frontend architecture by developing Next.js. This approach inverted traditional cloud setups by treating infrastructure as an automated byproduct of the application code itself, countering a prevailing industry detour toward static site generation. Rauch argued that real-time cloud rendering was essential to offload processing constraints from mobile client devices. This framework bet established Next.js as widespread infrastructure, supporting 1.3 million monthly active developers and powering major interfaces across retail (Supreme, Bose), election tracking (Polymarket, the Washington Post), and artificial intelligence products (ChatGPT, Claude, Sora, and Midjourney). Rauch asserts that keeping Next.js open source remains non-negotiable for developer adoption, while Vercel monetizes a managed, high-performance global runtime using a usage-aligned utility model comparable to Databricks or Snowflake. Shrinking Feedback Loops with Ephemeral Previews (12:08) Beyond baseline deployment speed, Rauch highlights ephemeral preview environments as a structural shift in software collaboration. By automatically provisioning a secure, live URL for every git branch, the platform eliminated the coordination bottleneck of developers relying on local machines to demonstrate progress. Rauch contrasts this with legacy corporate workflows—such as major financial institutions scheduling multi-person meetings a week in advance just to review an update—noting that instantaneous, mobile-accessible URLs force engineering teams to review real-world user conditions. Rauch notes that traditional developer experience optimization eventually faces diminishing returns governed by the speed of light and hardware constraints. While partnering with cloud serverless architectures reduced build and feedback cycles from hours to minutes, Rauch contends that the primary limitation in modern software creation is no longer compute hardware, but the supply of specialized humans able to build it. AI-Driven Development and the Expansion of v0 (17:07) To bypass the human coding bottleneck, Vercel launched v0 in September 2023, shifting development from a "code-first" to a "code-last" workflow where natural-language prompts compile into functional user interfaces. The product expanded from $1 million in annual recurring revenue (ARR) after ten months to $2 million within 14 days, and reached $4 million 20 days later. Rauch notes that this influx of generative workloads has significantly driven GPU consumption across major cloud regions. Vercel drew on lessons from an earlier failed browser-based product, Next.js Live, by coupling client-side rendering with foundation models, rigorous evaluations, and tailored design systems to avoid low-quality AI outputs. Concurrently, Vercel released its open-source AI SDK, which expanded from 100,000 to nearly one million weekly downloads in roughly a year. Rauch projects that companies will increasingly operate as specialized "token factories," encoding internal engineering or domain expertise into domain-specific assistants across verticals like CAD and medicine. Enterprise Security Hardening and Lean Team Scale (41:06) Rauch attributes Vercel's enterprise credibility to early exposure to cryptocurrency applications, which exposed the platform to frequent malicious traffic and necessitated built-in web application firewalls with sub-second DDoS mitigation directly at the origin. This standardized infrastructure also accelerates corporate acquisitions and enterprise hiring, as shared adoption of Next.js minimizes technical integration overhead to combining frontend layers over modular backend services. Dismissing the risk that high-scale startups will outgrow managed platforms, Rauch cites Polymarket sustaining intense 2024 presidential election traffic with a lean engineering footprint. He points to an external developer utilizing v0 and browser automation to launch a functional agent within 24 hours of OpenAI's Operator release as evidence of accelerated product development. Rauch forecasts that as software production costs fall toward zero, organizations will operate with smaller engineering headcounts while deploying a significantly higher volume of products. While Rauch envisions code generation becoming commoditized like machine translation—elevating the primacy of human taste and problem selection—the broader shift leaves open how traditional engineering organizations and enterprise staffing models will adapt as conversational prompts increasingly displace manual frontend development.

Vercel 联合创始人兼首席执行官 Guillermo Rauch 指出，公司通过开源框架 Next.js 与自动化云基础设施重塑了现代 Web 开发流程；随着生成式 AI 工具 v0 的爆发，软件构建正迈向由自然语言驱动的“代码后置”模式，使极简团队能够以极低边际成本支撑全球规模级应用。 押注动态架构与基础设施抽象化 (4:20) Rauch 回顾指出，2015 年前后虽然 Kubernetes 和 React 降低了技术门槛，但普通开发者将底层工具拼装为高性能网站依然异常复杂。当时行业主流偏向静态网站，而 Rauch 坚信移动端受限于硬件算力，必须将重度渲染与计算卸载至云端，因而坚持押注动态渲染架构。为解决 React 等基础库缺乏完整工程载体的问题，Vercel 主导推出了开源全栈框架 Next.js。目前 Next.js 月活跃开发者已达 130 万，并成为 ChatGPT、Claude 等主流 AI 产品的标准前端框架。 在商业模式上，Vercel 摒弃了早期开源软件依赖人工技术支持的旧路径，转而借鉴 Snowflake 和 Databricks 的按用量计费机制。Rauch 提出“基础设施应成为应用的输出产物”的核心理念，即开发者专注于应用本身，计算、安全与部署环境则由平台自动匹配并计费。据披露，Vercel 在 2024 年 5 月实现了 1 亿美元年化营收，年增速达到 80%，而其全球团队仅保持在 550 人的精炼规模。 预览环境革新与加密货币安全淬炼 (12:08) 在工作流层面，Vercel 通过将每个 Git 分支自动部署为独立的临时预览链接（Preview Environments），从根本上改变了团队协作模式。该机制彻底打破了开发者“只能在本地机器运行展示”的限制，使非技术人员与管理层能够直接在手机端审查真实运行状态，消除了大型跨国机构为核对基础功能而耗费数周协调会议的传统摩擦。 面对大型企业对托管平台稳定性的疑虑，Rauch 透露加密货币项目的爆发在客观上充当了严苛的安全试验场。由于早期 Web3 及 NFT 平台充斥着高度密集的恶意流量，Vercel 被迫将其安全架构强化数倍，直接在核心平台内建 CDN 与 Web 应用防火墙（WAF），将分布式拒绝服务（DDoS）攻击的自动缓解响应缩短至数秒级。这种原生防御能力，反而成为吸引 Fortune 500 企业及大型零售商在黑色星期五与大选流量高峰期放心入驻的关键技术背书。 从代码优先转向 v0 驱动的代码后置 (17:07) 在探索传统开发体验逼近物理极限之际，Vercel 借助大语言模型孵化了 AI 助手 v0，汲取了此前浏览器端纯代码开发产品 Next.js Live 的失败经验。Rauch 分析称，由于主流基座模型已在海量 Next.js 与 React 代码上充分预训练，v0 得以颠覆传统的“代码优先”开发流程，转向直接输入自然语言即可生成可用全栈界面的“代码后置（code last）”模式。v0 于 2023 年 9 月以研究预览版上线后，耗时 10 个月达成 100 万美元 ARR，随后仅用 14 天突破 200 万美元，并在其后 20 天内迅速扩充至 400 万美元。 为避免通用模型产生低质代码，工程团队结合评估基准（evals）、检索增强生成（RAG）与专用设计系统，将资深工程师的品味与代码范式注入推理。此外，Vercel 开源了底层 AI SDK，推动社区衍生出面向 CAD 设计、医疗科研等垂直领域的专家型 AI 工具。该 SDK 的周下载量在约一年内从 10 万次增长至近 100 万次，成为前端 JavaScript 生态中仅次于 OpenAI 原生库的第二大 AI 工具包。 破除毕业风险与软件边际成本归零 (52:04) 针对企业成长后可能因成本或定制需求迁出托管平台的“毕业风险”（Graduation Risk），Rauch 认为高度集成的平台能力让逃离变得不再必要。以预测市场平台 Polymarket 为例，一个极小规模的团队未自建复杂底层设施，即平稳承载了 2024 年美国大选期间的高并发峰值。在企业销售场景中，技术团队利用 v0 截屏老旧系统即可在数秒内输出 Next.js 原型，大幅缩短了系统迁移的技术验证周期。 Rauch 预测，随着“软件生产的边际成本走向零”，开发领域正在经历类似机器翻译对人工翻译的替代历程。在近期实践中，有开发者利用 v0 结合云端浏览器与 DeepSeek 模型，在 OpenAI 推出 Operator 功能后仅用 24 小时便构建出具备完整任务自动化能力的 Agent 产品。这种由机器承担繁杂编码机制的趋势，虽然会对传统工程岗位的人员规模形成下行压力，但将大幅降低软件创造的准入门槛，让更少的人力得以创造并管理面向全球的高并发产品。 尽管 AI 正在快速接管基础代码编写与架构重塑，Guillermo Rauch 强调人类的设计品味、审美把控与问题选择能力依然是核心护城河。然而，随着软件构建的边际成本持续下探，传统软件工程的分工模式与大型企业的既有人力组织架构，如何在这一波高人效变革中完成重构，仍是行业需要长期面对的演进命题。

## Background 本期视频由商业科技播客频道 Acquired 于 2025 年 2 月 18 日发布在 YouTube 平台。节目对 Vercel 创始人兼 CEO Guillermo Rauch 进行了深度访谈，围绕 Vercel 突破 32.5 亿美元估值与超 1 亿美元 ARR 的增长轨迹，深入探讨了其在碎片化 Web 市场中胜出的原因、明星 AI 产品 v0 仅用 14 天达成 200 万美元 ARR 的爆发内幕，以及如何围绕开源生态 Next.js 构建兼顾初创公司与大企业的商业闭环。 ## TL;DR Guillermo Rauch 回顾了 Vercel 如何从打破 2015 年繁琐的云基建门槛起步，通过押注云端动态渲染与垂直整合打造出成为现代 Web 及 AI 界面事实标准的 Next.js。访谈重点探讨了颠覆性 AI 工具 v0 的诞生，阐述了编程正从代码优先逆转为“代码为末”的自然语言交互范式，让开发转变为一种人人可用的通用技能。同时，Guillermo 结合自身技术型 CEO 的管理实践，详述了平台如何依靠经历极端网络安全考验的底层弹性，赋能不足 10 人的微型团队（如 v0 团队与 Polymarket）轻松支撑海量并发洪峰，彻底打破传统平台面临的“客户长大即流失”的毕业风险。 ## Atmosphere & style 整场对话充满极客情怀与高密度商业复盘，Acquired 主持人的宏观商业拆解与 Guillermo 深耕一线工程的独到见解形成默契共振。谈话风格坦诚笃定且兼具冷峻的技术反思，既有对早期行业盲目追逐静态架构的犀利点评，更弥漫着在 AI 降临后见证 24 小时极速交付与软件边际成本归零时的纯粹兴奋与远大野心。 ## Bulletin - [46:27](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2787s) Vercel 年化收入（ARR）已突破 1 亿美元且保持 80% 同比增速，公司估值达到 32.5 亿美元，全职员工仅维持在 550 人。 - [17:31](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1051s) AI 辅助开发产品 v0 增长斜率惊人，从上线后 10 个月达成 100 万美元 ARR，随后仅用 14 天便翻倍至 200 万美元，再过 20 天突破 400 万美元 ARR。 - [04:22](https://www.youtube.com/watch?v=4VG9B05xD7M&t=262s) 开源框架 Next.js 月活跃开发者突破 130 万，成为驱动 ChatGPT、Claude、Sora 等前沿 AI 应用界面以及主流大型网站的事实标准。 - [35:33](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2133s) 开源项目 AI SDK 周下载量在一年内从 10 万次飙升至近 100 万次，成为全网紧随 OpenAI 官方 SDK 之后第二大 JavaScript AI 依赖项。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=4VG9B05xD7M&t=0s) Vercel 核心定位为现代 Web 应用构建与部署的基础设施平台，深度承载媒体大选高并发追踪与黑五电商大促等苛刻场景。 - [02:11](https://www.youtube.com/watch?v=4VG9B05xD7M&t=131s) 2015 年科技巨头专有架构与普通开发者体验存在巨大断层，Vercel 旨在抹平开源基建与高质量即时发布之间的鸿沟。 - [05:30](https://www.youtube.com/watch?v=4VG9B05xD7M&t=330s) 通过软硬件式的垂直整合模式，Vercel 打破“框架归框架、云归云”的分离，帮助企业免除组装繁琐云配置和招聘昂贵运维团队的负担。 - [08:29](https://www.youtube.com/watch?v=4VG9B05xD7M&t=509s) 在社区狂热追捧纯静态架构（Jamstack）时，Vercel 反共识押注动态渲染与云端渲染（SSR），成功顺应了移动端性能减负的历史趋势。 - [13:36](https://www.youtube.com/watch?v=4VG9B05xD7M&t=816s) 每个 Git 分支全自动生成带独立 URL 的即时预览环境，不仅消除了“在我电脑上明明是好的”沟通摩擦，更重塑了跨职能协同与移动端实测流程。 - [17:31](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1051s) v0 将开发范式从“代码为先”反转为“代码为末”，让任何人凭自然语言即可直接构建并部署具备专业无障碍支持的交互式 Web 应用。 - [20:40](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1240s) 软件开发的本质正从专职角色（Role）演进为通用技能（Skill），人类核心价值正在向审美趣味、方向判断和上游创意设定转移。 - [22:26](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1346s) 借助开源的 AI SDK，各行业正在加速复制打造垂直领域的“v0”，印证了企业逐步转变为承载专有知识的“Token 工厂”的产业趋势。 - [24:19](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1459s) v0 的技术护城河源于浏览器原生 Next.js 运行管线与模型对齐工程的结合，并构建了专为 AI 优化的高容错设计系统以消灭“电子垃圾代码”（Slop）。 - [26:17](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1577s) 在企业销售中，现场利用 v0 将客户老旧页面截图秒级转译为现代化交互原型，大幅压缩了售前评估周期并形成降维打击。 - [29:12](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1752s) 单纯依赖优化传统编译部署链路正逼近光速物理极限，而大语言模型的介入彻底突破了开发者体验的边际效应递减天花板。 - [31:13](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1873s) 创业初期必须在单一细分赛道追求世界级卓越，Vercel 通过先在前端云建立垄断口碑，再由客户需求反向推动向全栈开发者云扩展。 - [35:33](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2133s) 坚持开源框架 Next.js 与 AI SDK 零直接变现，以此建立强大的开发者标准与流量飞轮，并依托按用量计费的透明云托管完成商业闭环。 - [37:49](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2269s) 主张“基础设施是应用程序代码的派生产物”，颠覆自底向上的云原生运维模式，让开发者专注于应用本身并天生具备多云与自托管能力。 - [41:20](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2480s) Crypto 行业极端恶意攻击的洗礼倒逼 Vercel 锤炼出秒级拦截的内嵌安全防护与弹性架构，彻底化解了困扰开发者平台的“毕业风险”。 - [44:27](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2667s) Next.js 成为跨公司技术整合的通用语言，大幅降低了企业并购中的前端融合摩擦力，并成为传统巨头吸引顶尖工程师的标配。 - [46:27](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2787s) 技术型 CEO 应深度介入业务细节并保持高频“吃自家狗粮”，利用对产品质感的深刻把握消除冗余层级汇报并实现高质量授权。 - [50:54](https://www.youtube.com/watch?v=4VG9B05xD7M&t=3054s) 不足 10 人的微型团队依托公网商业基础设施孵化出爆发式产品 v0，以及 Polymarket 扛住大选洪峰，均验证了小团队驾驭海量业务的极致杠杆形态。 - [53:47](https://www.youtube.com/watch?v=4VG9B05xD7M&t=3227s) 从 OpenAI Operator 到开源平替 Open Operator 仅耗时 24 小时，标志着软件边际构建成本逼近于零，全民开发者（Citizen Developers）时代已然降临。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=4VG9B05xD7M&t=0s) Vercel 的本质与“只要网站好用，它就可能跑在 Vercel 上” 节目开场，Acquired 团队向 Guillermo Rauch 发问：抛开超过 32.5 亿美元的估值、亮眼的客户列表以及各种光鲜的大数据，他自己会如何向大众定义 Vercel？Guillermo 回应称，Vercel 是一套用于构建和部署现代 Web 应用程序的基础设施平台（infrastructure platform）。从个人情结与公司基因来看，他长期极度痴迷于“开发者体验（Developer Experience）”、“极致性能（Performance）”与“设计（Design）”。对互联网上的普通访客而言，这意味着只要你点开一个链接，发现页面加载飞快、体验极为愉悦丝滑，它很大概率就是由 Vercel 驱动的。 Guillermo 拿具体场景举例：回想 2024 年 11 月的美国大选夜，无数网民在疯狂刷新网页追踪最新选情，这需要极其苛刻的实时数据吞吐能力。不管是《华盛顿邮报》（The Washington Post）、《明尼阿波利斯星坛报》（Star Tribune），还是 PolyMarket、Perplexity 这类新一代前沿创新产品，其前端底座全部托管在 Vercel 之上。同样在黑五（Black Friday）电商大促这种流量暴增、极度考验稳定与转化的时刻，从 Bose、Fanatics、Supreme 等知名大牌，到 Axel Arigato、Rag & Bone 等新兴品牌，都在依靠 Vercel 保证海量并发下的流畅体验。 - **Vercel 的核心定位**：构建与部署现代 Web 应用的一站式基础设施。 - **代表客户**：媒体与预测机构如 The Washington Post、PolyMarket、Perplexity；知名电商与零售品牌如 Bose、Fanatics、Supreme 等。 --- ### [02:11](https://www.youtube.com/watch?v=4VG9B05xD7M&t=131s) 2015 年的困局：开源繁荣与陡峭使用门槛的脱节 谈及创业初期，Guillermo 回忆在 2015 年前后，虽然市面上有一百万种构建网站或 Web 应用的方法，但没有任何一种能够达到 Google、Meta 或 Amazon 等巨头的质量标准。科技巨头拥有极强的基础设施，能够快速迭代、海量实验，并为每一位访客提供高度动态化、定制化的页面体验；而当时的普通 Web 生态却极其静态脆弱，网站稳定性很差。虽然云计算正在兴起，但当时的云平台（如原生 AWS）使用门槛极高，给人的感觉就像是“你必须拥有一个计算机博士学位”才能摆弄明白。 2014 至 2015 年间，Google 开源了 Kubernetes，Meta 开源了 React，业内开始吹嘘“Web 正在变快、现代化工具应有尽有”。然而对于普通开发者或正在进行数字化转型的传统企业而言，要将这些开源基建拼装成自己可用的内建平台，往往需要耗费数月甚至数个季度的时间去配置和调试。Vercel 诞生的初衷，正是为了抹平开源技术可用性与开发者快速发布高质量产品之间的巨大鸿沟。 ```mermaid flowchart TD accTitle: The Web Infrastructure Dilemma in 2015 accDescr: Comparison between tech giants' internal platforms and average developers' high friction setup in 2015. subgraph Giants["科技巨头（Google / Meta / Amazon）"] G1["自研复杂底座"] --> G2["高度动态 & 实时渲染"] G2 --> G3["极速发布 & 个性化体验"] end subgraph Developers["2015 年普通开发者面对的断层"] D1["开源引擎: React / Kubernetes"] --> D2["极度繁琐的云配置 & 组装"] D2 --> D3["动辄数月组装轮子 / 陷入静态困境"] end subgraph VercelSolution["Vercel 的破局点"] V1["全自动化基础设施封装"] --> V2["Next.js + 垂直整合云"] V2 --> V3["开箱即用达到巨头级体验"] end ``` --- ### [04:22](https://www.youtube.com/watch?v=4VG9B05xD7M&t=262s) 聚焦前端开发与 Next.js 的诞生 起初，Vercel 走的是通用路线（任何服务都可以部署），但 Guillermo 很快找回了自己的本源基因——专注于前端开发（Front-end Development）。他自 12、13 岁起便执迷于打磨直接面向终端用户的交互界面。于是，团队构建了两套紧密咬合的核心武器：一套是面向全球、深度优化的高速部署基础设施；另一套则是基于 React 的上层框架 Next.js。 如今 Next.js 已拥有 130 万月活跃开发者，成为了构建高性能 Web 前端的事实标准。Guillermo 指出，放眼当今的现代 AI 产品生态，无论是 ChatGPT、OpenAI Operator、Claude、Sora 还是 Midjourney，其前端界面几乎全由 Next.js 驱动。 - **Next.js 规模**：月活跃开发者（MAU）突破 130 万。 - **AI 领域黄金标准**：包括 ChatGPT、Claude、Sora、Midjourney 等标杆级 AI 应用均选用 Next.js。 --- ### [05:30](https://www.youtube.com/watch?v=4VG9B05xD7M&t=330s) 软硬件式的垂直整合：跳出“自己在 GitHub 捡零件拼车”的荒谬模式 Guillermo 坦言最初写下 Next.js 的第一行代码，源自极其强烈的个人挫败感：“巨头已经把未来指引出来了，论文发表了，代码也开源了，但要落地却难于登天。”他打了一个生动的比方：React 就像是一台性能强劲的“引擎”，但普通开发者和企业需要的是一辆组装好、能直接从 A 点开到 B 点的“汽车”，甚至是通往云端的“航天飞机”。对于企业而言，Next.js 与 Vercel 的结合不仅让网站更快，更能直接提升商业转化率与营收。 Vercel 最独特的差异化在于“软硬件垂直整合”模式（将上层框架工具与底层云基础设施紧密绑定并深度优化）。在当时，主流模式是让企业自己在 AWS 或 Google Cloud 上“玩生存冒险游戏（Choose Your Own Adventure）”，企业得去 GitHub 到处搜刮开源框架，招募年薪 50 万到 100 万美元的稀缺 Kubernetes 工程师去搭集群、造自己的内部框架。Acquired 主持人也对此大有共鸣，戏称当时给电商公司推销“我们给你托管 K8s”十分滑稽，因为电商公司要的是卖货而不是维护集群。Guillermo 幽默地总结道："Vercel 把这群（昂贵的）工程师全都招了过来，这样你们就不用招了。"企业统一使用 Next.js，不仅避免了重复造轮子，还能直接复用全网开发者在 Stack Overflow 上积累的庞大经验库；而在 AI 时代，各大 LLM 更是精通 React 与 Next.js，让使用这套栈的团队直接站在了全球智慧共识之上。 - **商业直觉**：打破“框架归框架、云归云”的分离模式，实现端到端的性能优化。 - **生态外溢效应**：工具标准化带来了庞大的社区问答库、极低的人才培训成本，以及现代 LLM 在代码生成上的天然亲和力。 --- ### [08:29](https://www.youtube.com/watch?v=4VG9B05xD7M&t=509s) 社区走偏的弯路：为何在静态应用狂欢时押注动态渲染 面对主持人“当年 Next.js 的机会究竟因何而生？是否因为 React 抽象层太底层、太重而需要有人为大众造好用界面？”的追问，Guillermo 揭示了一段深层的技术历史分歧。当时，整个前端开发社区经历了一场令人困惑的“静态应用（Static Applications）偏航”（即 Jamstack 等一股脑推崇纯静态生成的潮流），资本也在疯狂追逐各种静态网站框架。 Guillermo 从小喜欢逆向工程互联网上最伟大的产品，早年深入钻研 Gmail 的运行机制，并打造了风靡全球的实时双向通讯库 Socket.io。在他看来，Google 和 Facebook 信息流之所以能牢牢吸引并留住全球用户，核心正在于它们是“高度动态（Highly Dynamic）”的。社区放弃动态渲染、转头去捧静态框架的做法在他眼中不可理喻。因此，在外界盲目追捧静态框架的浪潮下，Guillermo 选择反其道而行之，默默打磨聚焦于“动态渲染”与“云端渲染（Server-side Rendering / Cloud Rendering）”的 Next.js，最终赢得了时代的发展方向。 ### [10:11](https://www.youtube.com/watch?v=4VG9B05xD7M&t=611s) 云端渲染与动态架构的“超额收益” 在移动互联网爆炸式增长的初期，业界的主流思潮大多集中在纯静态站点和极度厚重的客户端架构（fat client）。然而，Guillermo Rauch 却坚定地押注了截然相反的方向：他坚信将渲染逻辑移至云端，才能将设备端的计算负担彻底剥离。随着智能手机越来越普及，用户希望在移动端以极致的速度获取所需内容。将渲染与计算工作负载卸载到云端，正是他敏锐捕捉到的第一个“超额收益（Alpha）”机遇。这在当时是一条艰难得多的技术路线——要构建出能够支撑高并发弹性扩展、安全可靠且支持多租户的云基础设施，团队花费了整整数年时间才将其打磨完善。 与此同时，当时的 React 还处于非常底层的阶段。Guillermo 在与包括 Redfin 在内的诸多大型网站团队交流时发现，各大公司为了满足自身复杂的生产需求，纷纷不得不基于 React 手工搭建自己的上层抽象层。当被问及“React 是否本应从一开始就内置一个类似 Next.js 的框架层”时，Guillermo 给出了极其肯定的回答。他指出，Facebook 内部其实早就拥有一套自研的类似 Next.js 的体系，因为这本就是他们内部使用 React 的原生方式。开源界常有这种现象——就像顶尖科技论文一样，机构往往出于善意无偿开源核心理念，但在没有恶意保留的前提下，很多最关键的工程“配方”和胶水代码被遗漏在了内部。维护庞大的开源生态需要付出难以想象的心血，而市场最终用实际行动给出了答案：开发者真正渴求的是动态优于静态，以及包含服务端渲染（SSR）的完整端到端框架，而非仅仅一个底层的 UI 组件库。 ```mermaid flowchart TD accTitle: 云端动态渲染与传统胖客户端架构对比 accDescr: 对比传统胖客户端模式在设备端的高负担与 Next.js 结合 Vercel 在云端渲染降低移动端设备负担的流程。 subgraph Traditional["传统前端模式 (Fat Client)"] A["庞大客户端 JS 包"] --> B["移动端设备运算"] B --> C["首屏慢 / 体验受设备性能限制"] end subgraph Modern["Next.js + Vercel 架构"] D["云端多租户动态计算"] --> E["服务端渲染 SSR / 优化 HTML"] E --> F["轻量内容瞬时推送到移动端"] end ``` --- ### [12:08](https://www.youtube.com/watch?v=4VG9B05xD7M&t=728s) 从“提交代码即上线”的执念到 Next.js 的诞生 在构建整个商业版图时，Guillermo 最先动手搭建的并非 Next.js，而是承载部署的底层基础设施（后来的 Vercel）。他心中始终被一个极其纯粹的愿景所驱动：“开发者所需要做的唯一一件事，应当只是推送代码，然后立刻得到一个运行中的 Web 应用（push code, receive web application）。”在他看来，任何横亘在这一流程之间的繁琐环节——编写复杂的配置文件、手动组装 Kubernetes 集群、向企业 IT 运维部门提交工单——全都是阻碍效率的繁文缛节，而铲除这些阻碍就是属于他的巨大战略超额机会。 ``` 开发者推送代码 (git push) │ ▼ ┌─────────────────────────┐ │ Vercel 自动化云底座 │ ◄── 彻底消除：配置地狱 / 搭建集群 / IT 审批工单 └─────────────────────────┘ │ ▼ 即时交付可用 Web 应用 (Live App) ``` 然而，Guillermo 早期也踩过坑：最初他试图让这套平台无所不包，支持市面上几乎所有的语言与技术栈，结果导致基础设施过于泛化。痛定思痛后，他决定回归自己的看家本领——深耕前端领域。彼时技术社区中广泛流传着一个经典的梗图：现代前端工程化之复杂，简直堪比在黑板上解一道无休止的超级数学方程式。前端工程师群体向来善于自嘲，而正是这种对现状的自嘲与痛点，催生了 Next.js 的雏形。Guillermo 意识到，既然从零搭建一个高质量前端项目的门槛如此之高，那何不创造一个与底层基础设施天然深度咬合的高阶前端框架，将开发与发布流程一气呵成？ --- ### [13:36](https://www.youtube.com/watch?v=4VG9B05xD7M&t=816s) 预览环境的魔力：终结“在我电脑上明明是好的” 节目中，主持人在准备阶段曾向一位资深开发者朋友询问对 Vercel 最深刻的“高光时刻”。该开发者直言，最让他感到不可思议的体验，是发现自己在 Git 仓库中创建的每一个分支（Branch），都会被系统全自动部署成一个具备独立 URL 的即时预览环境（Preview Environment）。团队成员无需自行在本地搭建完整的模拟环境，就能直接在浏览器中与刚写完的代码实时交互。Guillermo 坦承，自己在最初设计时也没料到这个功能竟会爆发出如此巨大的威力。 - **消除沟通损耗**：过去开发者将代码从本地推向生产，往往需要在内部跨职能沟通中消耗数天甚至数周时间。Guillermo 曾偏执地追求将“敲击键盘到应用上线”的时间压缩到极致的毫秒级。 - **解锁跨职能协同**：企业对这一机制的推崇远超预期。当每个分支都拥有独立的临时 URL 时，设计师、产品经理、市场人员甚至外部客户都能直接点击参与评审，彻底消灭了“你等一个小时，我抱着笔记本走到你工位上给你演示”以及“在我电脑上明明跑得好好的”等经典职场尴尬。 - **淘汰陈旧的企业流程**：很多传统大厂仍在沿用极其低效的工作方式。Guillermo 提到，某全球顶尖跨国银行的首席信息官（CIO）曾告诉他，他们为了看一个正在开发的功能，必须协调 12 个人的日历开会，通过会议屏幕共享看某个工程师的本地桌面。相比传统开发环境中长期开机、极易遭受攻击的常驻测试机，Vercel 生成的预览环境完全是即用即弃（ephemeral）且高度安全的。 - **让开发者直面真实世界的移动端体验**：工程师日常习惯坐在顶配的 Mac 前工作，享受着极其宽裕的算力和屏幕空间；然而真实世界中，某头部电商客户 90% 以上的流量均来自手机。通过随时可分享的预览链接，团队可以在新功能上线前直接在手机上真机实测，逼迫团队直面终端用户的真实环境。 --- ### [17:31](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1051s) v0 爆发：从“以代码为先”到“以代码为末”的范式颠覆 在 Vercel 成立的第九个年头，公司内部孕育出了一款犹如“内部孵化初创”的颠覆性产品——AI 前端开发助手 v0。这成为了 Guillermo 创业历程中最具远见、也最令他震撼的里程碑。大语言模型（LLM）由于在海量公开代码中训练了极高比例的 Next.js 和 React 语料，展现出了惊人的 Web 编程天赋。Vercel 敏锐抓住了重构 Web 开发流程的机会：v0 允许用户仅凭一段简单的英文描述，就能直接生成可交互的 UI 界面和运行中的完整 Web 应用。 这一革新打破了传统的“代码为先（code first）”模式，将其反转为“代码为末（code last）”——“任何脑海中有想法的人现在都可以直接‘开火下厨’（anybody with an idea can cook）”。这种极低门槛不仅激活了现有的专业开发者，更将潜在用户群体从全球数百万开发者陡然放大到数以亿计的普通人。 | 发展节点 | 时间跨度 / 节点 | ARR 业绩表现 | |:--- |:--- |:--- | | **研究预览版发布** | 2023 年 9 月 | 正式上线小范围测试 | | **跨过首个里程碑** | 用时 10 个月 | 达成 $1,000,000 ARR | | **爆发式翻倍** | 仅用 14 天 | 飙升至 $2,000,000 ARR | | **极速狂飙** | 紧接着 20 天 | 迅速突破 $4,000,000 ARR | 这种惊人的增长斜率连经历过 Next.js 爆发的 Guillermo 都直呼平生未见。v0 产生的庞大计算需求，甚至一度导致数家全球头部云厂商在核心数据中心区域的 GPU 算力告罄。Guillermo 分享了他个人对 v0 的“顿悟时刻（Aha moment）”：作为一名从孩童时期就浸润在前端领域的元老级极客，他尝试用自然语言让 v0 重建自己此前手写的 Next.js 个人博客。结果令人震惊——v0 自动写出的代码不仅在桌面端和移动端的响应式设计上表现得比他亲手写的更出色，甚至自动补全了他当时遗漏的无障碍（Accessibility）支持功能，在产出质量上全面超越了人类专家的人工水准。 ### [20:40](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1240s) 从开发者基建到全民工具：开发正在变成一种“技能”而非“岗位” 随着 v0 的诞生与普及，Vercel 平台上的应用创建速率暴增了近四倍，新用户注册量直冲云霄。这一现象背后的本质转变在于漏斗顶层人群的彻底重构：过去，Vercel 归根结底是一套底层基础设施，虽然深受专业工程师的喜爱，但对普通人而言如同天书；而 v0 却将门槛抹平，让所有人都能轻松上手。面对主持人打趣 Vercel 是否正经历一段“从开发者品牌蜕变为消费级品牌”的企业转型，Guillermo Rauch 给出了更深层的产业判断。他认为，“写代码/软件开发”本身正在经历类似“翻译”的演变过程。 Guillermo 以 Google 翻译作比：在过去，将阿根廷西班牙语翻译成英语需要极高门槛的人工专业分工，而 Transformer 架构最初正是为了解决机器翻译而诞生；如今语言转换早已不再需要专门的人工翻译员。同理，大部分软件开发工作本质上就是将人类的“业务意图”翻译成可运行的“机器代码”。未来，软件开发将从一种专属的“职务角色”（Role）转变为一种机器日益擅长的“通用技能”（Skill）。人类的核心价值将彻底转移至上游——审美趣味、方向把控，以及定义让 v0 去解决何种高价值问题的创造力，这些依然坚不可摧。 - 应用创建速率在 Vercel 平台上暴增约 4 倍，新用户注册量大幅激增。 - 软件开发的本质正被重定义为“意图到代码的翻译过程”，AI 在代码生成、内容总结和语言转换上展现出天然统治力。 - 人类的差异化价值退守至更高阶的审美（Taste）、判断力与创意方向设定。 --- ### [22:26](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1346s) “v0 for X”浪潮与黄仁勋的“Token 工厂”预言 v0 的成功迅速向外辐射，在各行各业掀起了一股打造垂直领域“v0”的浪潮。前不久在社交媒体上爆火的“CAD 版 v0”（针对建筑设计和 3D 打印领域的 AI 辅助设计工具）正是基于 Vercel 搭建的，因为 Vercel 将构建 v0 的整套技术秘密通过开源项目 AI SDK 毫无保留地回馈给了社区。从面向医疗临床决策的 OpenEvidence（基于 Next.js 和 Vercel 构建），到处理邮件的 e0（Email 0），再到金融科技领域的垂直生成器，开发者正在各行各业复制 v0 的模式。 Guillermo 提及英伟达 CEO 黄仁勋（Jensen Huang）的一句名言：“绝大多数企业未来都会重构为 Token 工厂（Token Factories）”。企业内部所沉淀的专业知识、独特品味和对世界的认知，正是驱动这些工厂的核心资产。对于 Vercel 来说，其核心资产是顶级的 Web 界面工程与交互设计能力，由此凝练出了 v0；而其他企业也将各自领域的认知转化为专属 AI。Vercel 推动这场变革，不仅确立了自己的技术标杆地位，更让底层承载这些 AI 应用的 Next.js 与 Vercel 基础设施实现了巨大的商业飞轮效应。 ```mermaid flowchart TD accTitle: Vercel AI SDK 赋能生态垂直应用 accDescr: 展示 Vercel 底层开源技术赋能各垂直行业开发类 v0 产品的架构逻辑 subgraph Core ["Vercel 核心能力与开源工具"] A["行业认知与前端审美"] --> B["AI SDK (开源开发框架)"] C["Next.js / Vercel 云基础设施"] --> B end subgraph Vertical ["各行业'Token 工厂'落地应用"] B --> D["v0 (Web 应用与前端设计)"] B --> E["v0 for CAD (建筑与 3D 打印)"] B --> F["OpenEvidence (医疗科研助手)"] B --> G["e0 / 金融等垂直生成器"] end Vertical -.->|反哺平台调用与部署用量| C ``` --- ### [24:19](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1459s) 从 Next.js Live 的失败到 v0：为 AI 量身打造设计系统 当被问及 v0 究竟是自研了大模型还是做了一套产品化包装时，Guillermo 透露了 v0 鲜为人知的渊源。多年前，Vercel 曾推出过一个名为 Next.js Live 的失败产品，试图将整个 Next.js 运行环境直接搬进浏览器中，其初衷是跳过冗长的前端设计环节，让人们直接“用代码做设计”。然而该产品最终折戟，因为即便目标受众极具视觉感知力，他们依然得跨越极其陡峭的代码学习曲线。Next.js Live 就如同苹果当年的 Newton 掌上电脑，而今天的 v0 则是历经迭代后真正成熟的 iPhone。 v0 的独门绝技在于两者的结合：其一是在浏览器端极速运行 Next.js 和 Vercel 的原生渲染环境；其二是通过严密的基础设施对顶尖开源及商用基础模型进行定向调教与驾驭。为了杜绝社区普遍诟病的 AI“电子垃圾代码”（Slop），Vercel 围绕自身标准构建了一整套评测体系（Evals）、基准测试、精准提示词工程、检索增强生成（RAG），甚至开始为 AI 专门研发全新的设计系统（Design Systems）。Guillermo 笑称，十年前专注于为人类开发者编写框架的自己，放到今天恐怕难以取得同样的成功——因为现在的核心命题已经变成了“如何为 AI 编写开发框架与界面规范”。 - **前身教训**：Next.js Live 证实了让非技术视觉人群直接编写代码是行不通的死胡同。 - **技术壁垒**：浏览器内即时 Next.js 渲染管线 + 顶级通用基础模型 + 专项 Web 开发工程对齐。 - **消灭 Slop**：通过专属 RAG、评估基准与专为 AI 优化的高容错 UI 组件库，确保输出代码在无障碍访问（Accessibility）和设计感上保持顶级水准。 --- ### [26:17](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1577s) 将人类顶尖工程师的“审美偏好”数字化与销售现场的降维打击 长期以来，Vercel 内部始终面临着一个无法通过人力扩张解决的痛点：Next.js 火遍全球后，几乎每家大客户都迫切要求 Vercel 团队提供现场技术支持、代码审计与架构培训。然而，即便 Vercel 麾下聚集了 React 的技术负责人、Next.js 核心团队以及 Svelte 团队，这些顶尖技术大脑的收件箱每天都被成千上万的开发者塞满，根本不可能逐一服务所有企业。v0 的深层战略价值，正是将这些行业顶尖工程师的技术偏好数据和最佳实践范例注入系统，让顶级专家的经验与品味在云端无限制规模化。 这种能力在现代企业级销售中演变为令人惊掉下巴的“降维打击”场景。以往推动传统企业从陈旧技术栈迁移到现代 Next.js 架构，总是受制于漫长的评估周期和排期借口——客户虽有心重构，却苦于没有工程资源。如今在销售沟通会议中，Vercel 的销售工程师只需当场截取客户现存老旧网站的一张截图扔进 v0，几秒钟内便能直接生成一个现代化、高性能、像素级还原的 Next.js 交互式原型。 这种立竿见影的体验彻底颠覆了传统的 PPT 宣讲，将冗长拉扯的合同签署前周期变成了高信任度的产品共创。团队成员不再通过枯燥的文档来回扯皮，而是直接互发 v0 生成的预览链接协同修改。客户方案架构师和销售团队仿佛拿到了无所不能的“烹饪铲”，不仅销售周期被断崖式缩短，客户的信任与惊喜感更是达到了前所未有的高度。 --- ### [29:12](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1752s) 物理极限下的开发者体验与反馈循环革命 回顾大语言模型爆发的节点，Guillermo 坦言：如果没有 LLM 的出现，Vercel 尽管依然会是一家非常成功的企业，但在开发者体验（DX, Developer Experience）维度的探索很可能会迅速遭遇边际效应递减。优秀的开发者是这个世界上注意力最宝贵、最容易因外界打断而脱离“心流状态”（Flow State）的群体，他们对极速反馈循环（Feedback Loops）有着近乎偏执的渴望。为了优化这几秒钟的体验，Vercel 多年间深入到底层架构，与 AWS Serverless 团队紧密合作，用无服务器函数替代数百万美元的高冗余笨重基础设施，才换来了毫秒级的预览部署。 然而，单纯依靠传统软件工程去优化构建与部署链路，很快就会逼近“光速传输”这一物理世界的天花板。虽然现阶段不同企业的工程效率差异依然巨大——例如 Guillermo 接触的一家前沿数据库公司，其开发者在未优化前的反馈迭代周期长达整整 60 分钟，而改用 Vercel POC 方案后直接压缩至 2 分钟，为全公司薪酬最昂贵的人才释放了惊人生产力——但底层物理极限的阴影始终横亘在前。AI 的适时降临，彻底打破了这种渐进式优化的天花板，将优化的范式从“帮人更快地跑完编译”直接跃升为“帮人跨越代码编写本身”，为软件交付开辟了一条全新的维度。 ### [03:13](https://www.youtube.com/watch?v=4VG9B05xD7M&t=193s) 降低门槛与专注极致：单点突破才能建立开发者认知 随着部署时间从几分钟骤降至两秒，软件开发的瓶颈发生质的转变：制约创新的不再是机器或网络，而是全世界究竟有多少人具备编写软件的能力。这正是 v0 带来的革命——它让成千上万原本不具备传统工程背景的人也能参与到软件构建中，并直接接入这套经过全球顶尖企业实战检验的基础设施平台。 针对许多工程师兼顾开源项目与商业公司的创业困惑，Guillermo Rauch 分享了自己的核心心法：创业初期必须在单一细分领域做到世界级卓越。当 Vercel 刚起步时，团队将全部精力投注在“打造全世界最快的前端网站与电商橱窗”。在开发者和企业级市场，做“样样通样样松（jack of all trades）”注定失败，挑剔的工程师与大企业只想要各自品类中最顶尖的产品。Vercel 借此确立了在前端领域的霸主地位，随后才顺理成章地向外扩展，完成从“前端云（Frontend Cloud）”向“开发者云（Developer Cloud）”的蜕变。 很多初创公司容易犯一个错误：认为产品最初的形态必须与自身远大的野心 1:1 匹配。Vercel 当年的野心是重构所有运行在移动设备与 Web 端的应用程序，但他们克制住了铺大摊子的冲动。多年来，开发者不断向 Vercel 索要数据库或完整的后端能力，Guillermo 的回答始终是“我们还没准备好”。这种与客户的“双人舞”理念在于：给客户提供足够极致的功能让他们爱不释手，然后让客户反向“逼”着平台推出更多产品，而非一开始就盲目硬塞所有功能。如今，随着开发者构建 AI 应用后端、处理 Token 流式输出与结算系统的诉求爆发，Vercel 顺理成章地推出了全栈后端基础设施。 ```mermaid flowchart LR accTitle: 从专注单点到开发者云的演进路径 accDescr: 展示 Vercel 从最初专注最快前端，演进到受客户拉动拓展为全功能开发者云的过程 subgraph Phase1 ["第一阶段：极致聚焦"] A["专注前端与电商速度"] --> B["在 Next.js/前端云建立垄断口碑"] end subgraph Phase2 ["第二阶段：客户拉动"] B --> C["克制克制再克制<br/>'我们还没准备好'"] C --> D["开发者主动索要更多后端能力"] end subgraph Phase3 ["第三阶段：开发者云"] D --> E["拓展全栈与数据库支撑"] E --> F["构建 AI 应用基础设施平台"] end ``` --- ### [03:33](https://www.youtube.com/watch?v=4VG9B05xD7M&t=213s) 开源与商业化：零直接变现但构建无可替代的飞轮 当被问及 Next.js 带来零直接收入、纯粹作为开源项目运作是否曾考虑换种打法时，Guillermo 断然回答：“绝无可能（never absolutely not）”。对于生于阿根廷的 Guillermo 而言，开源是他走出阿根廷、走向世界的通票。在他看来，现代开发者根本无法接受一套专有的闭源前端框架。 同样的飞轮效应正在 AI 领域重演。Vercel 推出的开源项目 AI SDK 旨在让构建类似 ChatGPT 的应用民主化。Web 的核心精神是去中心化——在专属域名上设立自己的规则，无需向任何平台申请许可，所有颠覆性创新（包括最初用 Next.js 搭建的 ChatGPT 研究预览版）都诞生在 Web 之上。AI SDK 不追求任何直接的商业变现，但它在短短一年内从每周 10 万次下载飙升至近 100 万次，成为紧随 OpenAI 官方 SDK 之后全球第二大 JavaScript/前端 AI 模块。 这项策略的商业逻辑十分清晰：用开源换取庞大的开发者采用量、社区反馈与品牌心智。当开发者需要将项目真正投入生产并部署上线时，他们自然会首选 Vercel，因为 Vercel 已经为他们把繁琐的基础设施细节全部打磨妥当。类似于 Snowflake、Databricks 或 Datadog，Vercel 采用按用量计费的透明模式：项目尚处起步阶段时收费极低；一旦业务爆发席卷全球，再按规模付费。开发者完全清楚自己的每一分钱花在了哪里，这是一笔极度公平的交易。 - **AI SDK 增长曲线**：在约一年时间里，周下载量从 10 万次增长近十倍，逼近 100 万次/周，成为 JS 生态第二大 AI 依赖项。 - **开源的核心定位**：作为纯粹的获客引擎与开发者标准载体，不设任何软件使用授权收费限制。 - **商业化对应逻辑**：采用用量计费（Usage-based Pricing），卸下开发者运维负担，与客户业务规模共同成长。 --- ### [03:49](https://www.youtube.com/watch?v=4VG9B05xD7M&t=229s) 架构哲学：基础设施应是应用的派生产物而非起点 关于“是否该把商业基础设施完全开源”的提问，Guillermo 指出将底层云平台完全开源会陷入收益递减的死胡同。搭建跨全球 20 个可用区、运行成百上千个微服务的基础设施，其核心挑战根本不在于公开源代码，而在于极端复杂的运维工程能力——监控告警、无缝升级、滚动部署以及灰度分阶段发布。 开源的真正价值在于确立一种“编程模型（Programming Model）”，这正是 Next.js 的角色——它定义并声明了应用程序本身。在此基础上，Vercel 提出了颠覆传统的架构理念：**基础设施应当是应用程序的派生产物（Infrastructure as an output of the application）**。绝大多数拥抱云原生架构的公司往往“从底层基础设施出发”，先去搭服务器和容器集群；而 Vercel 将整个流程彻底逆转： > "大部分人去用云的时候都从基础设施开始，我们把它倒过来了。我们告诉你：用 Next.js 框架，只关注你的应用程序就好。你天生就是多云架构，你有能力在任何地方自托管它；但我们替你扛下了这份脏活累活，为你提供 Vercel 这一选项。" ```mermaid flowchart TD accTitle: 基础设施架构思维对比 accDescr: 对比传统云架构与 Vercel 提出的基础设施派生模型 subgraph Traditional ["传统云架构路线（自底向上）"] T1["预先配置基础设施/K8s/多区域集群"] --> T2["搭建网络与复杂微服务运维"] T2 --> T3["编写并适配应用程序代码"] end subgraph VercelModel ["Vercel 架构模型（自顶向下）"] V1["开源 Next.js 框架声明应用逻辑"] --> V2["基础设施成为应用的自动派生产物"] V2 --> V3["开发者自建自托管"] V2 --> V4["一键接入托管云（全球部署、弹性计算与防御）"] end ``` 这种理念与历史上的开源商业模式形成了鲜明反差。早期的 Red Hat 模式本质上是售卖技术支持服务，商业天花板明显且缺乏规模效应。与 Vercel 最具可比性的是 GitHub 和 Databricks： - **GitHub** 并不发明底层版本控制工具 Git，但它构建了一套以代码 URL 为核心的协作平台；而 Vercel 则进一步往前推进，直接为开发者生成“可运行应用程序的 URL”。Guillermo 甚至预言，未来当代码沦为 AI 的生成域时，源代码本身的重要性会下降，可运行的动态体验将更加核心。 - **Databricks** 围绕开源的 Apache Spark 打造了一套高度优化的云端分布式计算引擎。用户完全可以自己 DIY 搭建集群，但全球算力集群的弹性伸缩才是最难啃的硬骨头。 - **Vercel 的核心护城河**：恰恰在于接管了应用运行、内容分发、安全防护、DDOS 缓解以及应对极端流量脉冲（massive spikes of traffic）的所有复杂工程。对于像 Supreme 发售或大型银行迁移这样的严苛场景，这种开箱即用的抗高压能力构成了企业托付关键业务的基础。 ### [41:20](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2480s) 加密浪潮带来的极端考验与安全护城河 Guillermo 谈到，每当互联网出现颠覆性的新技术浪潮时，对于 Vercel 而言既是天赐良机也是巨大挑战。无论是早期的加密货币（Crypto）还是如今的人工智能（AI），初创团队追求的最核心诉求都是“以最快速度把想法变成现实”，因此绝大多数创新者都会首选 Vercel。然而，加密行业在当年吸引了“整个互联网上最底层、最阴暗的恶意攻击者”。这迫使 Vercel 在安全领域成倍加码——无论是用于连接数据库、后端以及加密钱包的底层机密保险库（Vaults）的精密加密技术，还是处理错综复杂的密码学细节与海量流量识别。 运行互联网底层基础设施最棘手的挑战，在于精准区分好流量与恶意流量。Vercel 坚持用“Vercel 的方式”做安全：零配置（Zero Configuration）且开箱即默认安全（Secure by default）。传统 CDN 或外部 WAF 服务商（例如 Akamai、Cloudflare）因为本身并不直接承载和运行客户的应用程序，它们只是在最外层做流量清洗，即便漏掉一部分攻击流量，它们也很难被苛责到底有没有做好；但 Vercel 同时承载着前端与全栈应用本身，“你必须瞬时过滤掉 100% 的恶意流量，否则宕机的就是你自己的系统”。这种商业与技术诉求的高度一致，倒逼 Vercel 打造出了顶级 CDN。 Crypto 时代的残酷洗礼让 Vercel 彻底脱胎换骨。如今炙手可热的 Solana 区块链将其所有的 Web 资产部署在 Vercel 上，Vercel 实际上成为了整个 Web3 世界通往 Web2 的前端门面。无论是各大 NFT 交易平台还是 Meme 币引发的脉冲式流量洪峰，都在反复锤炼 Vercel。其 DDoS 防御响应时间如今达到了业界领先的秒级标准，而同行有时甚至需要数分钟。有时客户误在 Vercel 外层又套了一层第三方 CDN，Vercel 的工程师甚至能清晰监测到底层攻击穿透了外层 CDN 直扑而来，这也印证了其自身防御能力的强悍。 ```mermaid flowchart TD accTitle: 传统外部CDN与Vercel内嵌安全架构对比 accDescr: 对比传统外部CDN/WAF与Vercel应用与CDN原生集成在安全和流量过滤上的差异 subgraph Legacy["传统模式（外部 CDN / WAF）"] L_Traffic["公网流量 / 攻击"] --> L_CDN["第三方 CDN / WAF (如 Cloudflare / Akamai)"] L_CDN -- "流量清洗与穿透" --> L_App["独立应用托管服务商"] style L_CDN stroke-dasharray: 5 5 end subgraph VercelModel["Vercel 统一全栈模型"] V_Traffic["公网流量 / 极端黑客攻击"] --> V_Edge["Vercel 边缘 CDN + WAF（秒级 DDoS 拦截）"] subgraph V_Core["Vercel 内部原生集成"] V_Edge --> V_Sec["机密保险库 (Vaults) / 零配置安全"] V_Sec --> V_App["Next.js 应用原生全栈运行"] end end ``` ### [44:27](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2667s) Next.js 成为并购通用语言与企业数字化转型 Acquired 播客指出，这里存在一个耐人寻味的黑色幽默：如今那些与 Vercel 签下数百万美元多年期大单的财富 500 强巨头，借由 Vercel 进行数字化转型和组织提效，其底层享受的稳定可靠性，竟然全拜当年加密世界极端 DDoS 攻击的“淬火打磨”所赐。Guillermo 指出，Vercel 渗透进超大型传统企业还有一条独特的路径——并购（M&A）。传统巨头为了保持竞争力，会直接买下那些具备颠覆性的小型创新团队。 在这些并购案中，Next.js 与 React 扮演了“通用技术语言”的角色。在过去，两家公司的底层架构整合往往是旷日持久的技术噩梦，但在 Next.js 的体系下，由于“基础设施本身就是应用程序代码导出的直接产物”，跨公司整合甚至精简到了只需要将代码文件复制粘贴到另一个代码仓库中。整合的核心变成了两个前端的融合，而各种后端服务天生就是可插拔的模块。这赋予了初创公司强大的杠杆，促使传统企业意识到：必须登上 Next.js 这趟列车，否则根本无法吸引和留住顶级人才，更无法实现敏捷交付。Guillermo 更是自豪地透露，连他自己最钟爱的一家德国顶级豪华汽车品牌，其 Web 产品线也全面跑在 Vercel 之上。 - **M&A 摩擦力骤降**：Next.js 与 React 标准化了全栈开发范式，后端模块解耦，前端合并化繁为简。 - **人才虹吸效应**：大企业全面拥抱 Next.js，核心诉求不仅是性能，更是为了在开发者市场上争夺与留存工程师。 - **德国车企趣谈**：虽然受限于 Logo 使用授权无法公开点名，但主持人调侃从 Vercel 一贯的极简设计与审美风格，听众便能猜出是哪家德国豪华汽车巨头。 ### [46:27](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2787s) 技术型 CEO 的组织穿透与管理哲学 在被问及作为技术创始人有何“举世皆非、唯我独信”的强烈信念时，Guillermo 直言：技术型 CEO 不仅可以、而且必须深入参与公司业务的方方面面。从前线亲自帮销售团队关单（Close Deal），到复盘季度财报细节并校准销售配额与佣金制度，再到定义未来产品路线图以及主导定价策略，他都事必躬亲。针对外界可能认为这不符合传统大公司职业经理人分工的偏见，他强调当前 Vercel 的规模早已不是小作坊： - **年化收入（ARR）**：截至 2024 年 5 月已突破 1 亿美元，并且依然保持着 80% 的同比增速。 - **开发者生态**：活跃开发者基数在过去一年翻了一倍以上。 - **团队规模**：在全球仅有 550 名员工。 Guillermo 认为，现代企业的 CEO 角色正在发生结构性转变。传统层层设卡、各自为政的信息孤岛正在被打破，企业运作的核心是“优化内部信息流动效率”。在拥有 550 名员工的同时保持对细节的掌控，秘密武器正是“吃自己的狗粮（Dogfooding）”：他每天都高频使用 v0 和 Vercel，随时与一线开发者社区和客户直接对话。由于对产品技术现实了如指掌，公司省去了下级向上级汇报、为高管制作汇报 PPT 的漫长冗余环节。 当被问及这种高强度介入是否会挫伤高管的积极性时，Guillermo 阐明了自己的界限：高管需要对最终的 KPI、数字和结果负全责，而 CEO 的核心职责是成为“连接真实世界的桥梁”，为他们提供最高质量、最接地气的真实反馈。CEO 绝不能把技术领域当成不可触碰的黑盒甩手给下属，而是要保持对底层质感（Texture）的深刻理解。“正因为我自己去深入学习了底层运作机制，我才能给出高水准的反馈，也才能放心地进行真正有把握的授权。” ### [50:54](https://www.youtube.com/watch?v=4VG9B05xD7M&t=3054s) v0：在企业内部验证极简人效与极致杠杆 十年前如果一家公司达到 1 亿美元 ARR 且服务数百万开发者，其员工数量往往需要数千人，而 Vercel 仅用了 550 人。Guillermo 列举了公司内部现象级产品 **v0** 的诞生过程作为极致人效的绝佳例证。在立项之初，v0 被定位为“大孵化器中的创业团队”，其核心原则是：全链路必须 100% 运行在 Vercel 现有的对外商业产品之上。 - **技术全栈闭环**：前端、后端、数据库存储以及 AI 大模型的调度采购，无一例外依托 Vercel 平台搭建。 - **极小团队规模**：这支孵化出全行业增长最快产品之一的核心团队，规模甚至不足 10 人。 - **绝对禁止内部特权**：Guillermo 以一种近乎神经质的偏执严格监督这支团队，严禁他们“作弊”——他们绝不允许私下找到内部底层基础架构团队开后门或索要特殊内部权限，所有调用的必须是任何普通外部开发者都能在公网买到的商业功能。 这一实验不仅带来了令人咂舌的极高人均创收（Revenue per head），更向全球技术界提供了一个铁证：基于 Vercel 构建现代软件，即便是一个不足 10 人的超微型团队，也足以撬动不可思议的商业与产品杠杆。 ### [52:04](https://www.youtube.com/watch?v=4VG9B05xD7M&t=3124s) 破除“毕业风险”迷思：小团队驾驭大流量与 Polymarket 奇迹 在谈到底层架构能力时，Guillermo Rauch 展现出了极强的底气。许多开发者常有疑虑：当产品瞬间迎来每秒上百万次请求（1 million requests per second）的超高并发时，平台是否需要专门为其定制或临时调配特殊资源？Guillermo 的答案非常明确：“零（Zero）”。他认为，正是这种开箱即用的高弹性，彻底消除了困扰开发者服务平台的“毕业风险”（Graduation Risk）——这是所有致力于简化开发体验的初创公司最恐惧的“幽灵”。业界常常有一种声音：“等我的业务做大了，我就会长得太大而不得不离开 Vercel（outgrow Vercel）。” 但 Guillermo 表示，只要自己还在掌舵并深耕技术细节，他就确保平台本身能直接提供所有必要的“逃生门”（Escape Hatches）与高级底层扩展能力，让极小的团队也能安全地支撑起庞然大物般的业务体量。 ```mermaid flowchart TD accTitle: Vercel支撑超大规模流量与极简团队架构 accDescr: 展示小团队如何依托Vercel与AI，免去底层运维并消除毕业风险 subgraph Input["极简团队与AI生产"] A["小型核心团队"] --> B["AI 辅助生成海量代码"] end subgraph Platform["Vercel 托管与抽象层"] B --> C["自动弹性扩缩容"] C --> D["内置高级能力与逃生门"] end subgraph Impact["业务结果与未来形态"] D --> E["支撑极高流量峰值<br/>(如 Polymarket 大选流量)"] E --> F["无需脱离平台<br/>(无毕业风险)"] end ``` 预测市场平台 Polymarket 便是这一理念的绝佳印证。Polymarket 背后的技术团队规模极小，但在美国总统大选期间，该网站迎来了足以让老牌基础设施工程师都感到胆战心惊的超大洪峰流量。Guillermo 指出，那种规模的并发流量，通常需要你在互联网基础设施领域摸爬滚打数十年、组建庞大运维团队才能勉强维系；而 Polymarket 仅凭一支袖珍团队就轻松抗住了压力。基于这一现实，Guillermo 做出了一个坚定预言：未来的团队规模必然会进一步缩小。当 AI 负责编写绝大多数代码，底层基础设施被完美抽象，团队即便业务规模变得极其庞大也无需经历繁琐复杂的“迁离毕业”流程时，人力规模势必会面临下行压力。但这绝不意味着“开发者要集体失业去沙滩退休”，而是意味着极少数人就能打造出触达全球更多用户的全新软件。 - **消除毕业风险**：平台内建全部进阶功能与灵活通道，无需手动为客户调配特殊集群，即可原生承受每秒百万级别的请求压力。 - **Polymarket 案例**：仅凭微型团队依托 Vercel，就平稳扛住了 2024 年美国大选期间让顶级互联网老兵都生畏的海量峰值流量。 - **团队形态预测**：AI 自动生成代码与高度抽象的云底座相结合，将推动团队形态极致小型化，以极少的人力杠杆撬动前所未有的软件覆盖面。 --- ### [53:47](https://www.youtube.com/watch?v=4VG9B05xD7M&t=3227s) 软件边际成本归零与 24 小时上线的 Open Operator 当被问及这种新生产力范式究竟“能有多快、能达到多大体量”时，Guillermo 讲述了一个发生在采访前一周的真实案例。OpenAI 刚刚发布了具备浏览器操作能力的智能体 Operator，仅仅 24 小时内，一位 Vercel 的客户就打造并上线了开源平替版本“Open Operator”。该团队通过 v0 快速生成用户交互界面，利用云端的无头浏览器基础设施承接 Agent 对网页的自动化操作指令，并接入 DeepSeek 等前沿开源模型来驱动智能体决策。从一个全新的底层概念——“Agent 可以接管浏览器并为人代办任务”——到落地成一个面向全球可用的成熟产品，整个周期只用了整整 24 小时。 该项目发布后迅速席卷了整个开发者圈子。Guillermo 在社交媒体 X（原 Twitter）上转发该项目后，收到了多封来自中国开发者的邮件，对方震惊地表示自己“彻底看呆了（losing my mind）”，完全不敢相信这样一个复杂产品竟然能以如此恐怖的速度在一天之内问世。Guillermo 指出，软件的交付周期已经从以季度、月为单位压缩到了几小时、几天；当然，团队随后仍需面对持续创新的考验，不能仅靠一次病毒式传播而缺乏后续迭代。更深远的意义在于，当软件的边际生产成本无限逼近于零时，整个市场的创造门槛彻底被踏平，“全民开发者”（Citizen Developers）时代正式到来。Vercel 的终极目标正是站在这场浪潮的核心枢纽，无论是人类专业工程师、新兴的公民开发者还是自主调用的 AI Agent，都能通过 Vercel 获取一站式构建与发布软件所需的一切。 在访谈尾声，主持人们感叹这是本次对话最完美的收尾点。Guillermo 也向所有听众发出行动号召：若想与团队取得联系或提供产品反馈，可以直接在 X 上寻找 @v0 或 @vercel；如果想要亲身体验下一代开发模式，只需访问 `v0.dev`，输入你的构想并点击提交，“你就能即刻出发开始构建（off you go）”。 - **24 小时极速交付**：OpenAI 推出 Operator 后，开发者仅耗时 24 小时便结合 v0、云端浏览器架构与 DeepSeek 模型上线了全功能替代品 Open Operator，引发跨国开发者的强烈震撼。 - **边际成本逼近于零**：AI 降本增效使得软件构建成本骤降，推动大量没有传统编程背景的公民开发者（Citizen Developers）涌入创作市场。 - **行动呼吁与入口**：Guillermo 邀请听众访问 `v0.dev` 体验自然语言生成前端应用的快感，并在 X（@v0 / @vercel）上积极提交体验反馈。 ## Notable quotes > [05:30](https://www.youtube.com/watch?v=4VG9B05xD7M&t=330s) Vercel 把这群（昂贵的）工程师全都招了过来，这样你们就不用招了。 — Guillermo Rauch > [12:08](https://www.youtube.com/watch?v=4VG9B05xD7M&t=728s) 开发者所需要做的唯一一件事，应当只是推送代码，然后立刻得到一个运行中的 Web 应用（push code, receive web application）。 — Guillermo Rauch > [17:31](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1051s) 任何脑海中有想法的人现在都可以直接‘开火下厨’（anybody with an idea can cook） — Guillermo Rauch > [22:26](https://www.youtube.com/watch?v=4VG9B05xD7M&t=1346s) 绝大多数企业未来都会重构为 Token 工厂（Token Factories） — 黄仁勋 > [03:33](https://www.youtube.com/watch?v=4VG9B05xD7M&t=213s) 绝无可能（never absolutely not） — Guillermo Rauch > [03:49](https://www.youtube.com/watch?v=4VG9B05xD7M&t=229s) 大部分人去用云的时候都从基础设施开始，我们把它倒过来了。我们告诉你：用 Next.js 框架，只关注你的应用程序就好。你天生就是多云架构，你有能力在任何地方自托管它；但我们替你扛下了这份脏活累活，为你提供 Vercel 这一选项。 — Guillermo Rauch > [41:20](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2480s) 你必须瞬时过滤掉 100% 的恶意流量，否则宕机的就是你自己的系统 — Guillermo Rauch > [46:27](https://www.youtube.com/watch?v=4VG9B05xD7M&t=2787s) 正因为我自己去深入学习了底层运作机制，我才能给出高水准的反馈，也才能放心地进行真正有把握的授权。 — Guillermo Rauch > [52:04](https://www.youtube.com/watch?v=4VG9B05xD7M&t=3124s) 等我的业务做大了，我就会长得太大而不得不离开 Vercel（outgrow Vercel） > [53:47](https://www.youtube.com/watch?v=4VG9B05xD7M&t=3227s) 彻底看呆了（losing my mind） > [53:47](https://www.youtube.com/watch?v=4VG9B05xD7M&t=3227s) 你就能即刻出发开始构建（off you go） — Guillermo Rauch ## Mentioned Vercel, Guillermo Rauch, Acquired, Next.js, v0, AI SDK, The Washington Post, Star Tribune, PolyMarket, Perplexity, Bose, Fanatics, Supreme, Axel Arigato, Rag & Bone, Google, Meta, Amazon, Kubernetes, React, AWS, ChatGPT, OpenAI Operator, Claude, Sora, Midjourney, Socket.io, Redfin, OpenEvidence, e0, 黄仁勋 (Jensen Huang), NVIDIA, Next.js Live, Snowflake, Databricks, Datadog, GitHub, Git, Red Hat, Apache Spark, Solana, Cloudflare, Akamai, Svelte, DeepSeek, Open Operator ## Video description Vercel has become the infrastructure platform powering modern web development over the past several years, with companies from Stripe to Adobe to Runway all building their front ends on them. Today we’re joined by founder and CEO Guillermo Rauch, who shares why Vercel has been uniquely successful in the fragmented (to say the least!) world of web development platforms. There are now more than 6 million Vercel users, 80,000 active teams, and users have grown 200% year-over-year. The company also crossed $100m in annualized revenue last May, and Guillermo shared with us that they’ve been growing at 80% since, and were recently valued at $3.25 billion. This is also a particularly interesting moment for Vercel. Last year they launched a new product, “v0”, which lets anyone create and deploy a working website simply by describing it in English and letting AI take care of the rest. Guillermo shares its origin story within the company (and insanely that it reached $2m ARR in the first 14 days!), and how it’s changed their entire thinking about what’s possible now with AI products. We also cover: * How to build a business around an open source project (Next.js) * How they balance both being a fast and nimble platform for startups with being a reliable platform for enterprises * Guillermo's unconventional approach to staying deeply technical as CEO at scale Links: * Vercel ( https://vercel.com/ ) * V0.dev ( https://v0.dev/ ) * Next.js ( https://nextjs.org/ ) *More Acquired:* - Get email updates https://www.acquired.fm/email and vote on future episodes! - Join the Slack http://acquired.fm/slack - Check out the latest swag in the ACQ Merch Store https://www.acquired.fm/store! _Note: Acquired hosts and guests may hold assets discussed in this episode. This podcast is not investment advice, and is intended for informational and entertainment purposes only. You should do your own research and make your own independent decisions when considering any financial transactions._ [Source](https://www.youtube.com/watch?v=4VG9B05xD7M)

[5:30] Guillermo Rauch: Vercel 把这群（昂贵的）工程师全都招了过来，这样你们就不用招了。 [12:08] Guillermo Rauch: 开发者所需要做的唯一一件事，应当只是推送代码，然后立刻得到一个运行中的 Web 应用（push code, receive web application）。 [17:31] Guillermo Rauch: 任何脑海中有想法的人现在都可以直接‘开火下厨’（anybody with an idea can cook） [22:26] 黄仁勋: 绝大多数企业未来都会重构为 Token 工厂（Token Factories） [3:33] Guillermo Rauch: 绝无可能（never absolutely not） [3:49] Guillermo Rauch: 大部分人去用云的时候都从基础设施开始，我们把它倒过来了。我们告诉你：用 Next.js 框架，只关注你的应用程序就好。你天生就是多云架构，你有能力在任何地方自托管它；但我们替你扛下了这份脏活累活，为你提供 Vercel 这一选项。 [41:20] Guillermo Rauch: 你必须瞬时过滤掉 100% 的恶意流量，否则宕机的就是你自己的系统 [46:27] Guillermo Rauch: 正因为我自己去深入学习了底层运作机制，我才能给出高水准的反馈，也才能放心地进行真正有把握的授权。 [52:04] 等我的业务做大了，我就会长得太大而不得不离开 Vercel（outgrow Vercel） [53:47] 彻底看呆了（losing my mind） [53:47] Guillermo Rauch: 你就能即刻出发开始构建（off you go）

[0:00] Guillermo Rauch: the infrastructure platform to build and deploy modern web applications. [1:56] Guillermo Rauch: it felt like you needed a PhD. [4:20] Guillermo Rauch: things like react are like the engine and people need a car to go from A to B. [4:20] Guillermo Rauch: weird detour towards static applications [12:08] Guillermo Rauch: I can only show you on my machine [12:08] Guillermo Rauch: developers sometimes like we are the most coddled people in the planet. [17:07] Guillermo Rauch: startup within a startup. [17:07] Guillermo Rauch: making the hyperscalers run out of GPU capacity [19:30] Guillermo Rauch: since I'm almost in diapers, [22:26] Guillermo Rauch: refactor themselves into token factories. [31:35] Guillermo Rauch: Being a jack of all trades doesn't work in this world. [34:33] Guillermo Rauch: never absolutely not. [34:33] Guillermo Rauch: ticket out of Argentina and into the world. [36:35] Guillermo Rauch: infrastructure has to be an output of the application. [39:12] Guillermo Rauch: URLs to working applications [41:06] Guillermo Rauch: literally the worst darkest actors inside of the internet, [41:06] Guillermo Rauch: delicious irony [46:27] Guillermo Rauch: can and should be involved in every aspect of the business. [49:38] Guillermo Rauch: bridge to reality [52:04] Guillermo Rauch: scary ghost of every startup that tries to simplify things [53:44] Guillermo Rauch: marginal cost of software production goes to zero,

[0:00] hello acquired listeners and welcome to [0:02] another great episode of aq2 today we [0:06] have GMO Rous the founder co-founder and [0:10] CEO gearmo what's the right terminology [0:11] here all of the above everything The Man [0:14] Behind versel aside from all the big [0:17] fancy numbers valued over3 billion [0:19] dollar and amazing customer list and [0:21] blah blah blah we wanted to turn the [0:23] question to you and say how do you [0:25] describe versel versel is the [0:27] infrastructure platform to build and [0:29] deploy Loy modern web [0:31] applications something that's personal [0:34] that is infused into the company is I'm [0:38] obsessed with developer experience with [0:41] performance and with design so the [0:44] platform is a way of deploying the best [0:48] possible web applications with ease for [0:51] the average visitor of the internet what [0:52] it means is that every time you go to a [0:54] link if it's fast if it's delightful [0:57] it's likely to be built on versell [1:00] amazing yeah it's quite the flex yes so [1:03] but I'll give you an example think back [1:05] to November 2024 us election and you're [1:09] checking out new sites you're refreshing [1:12] you want to know the latest and greatest [1:13] you need real time data and so whether [1:16] you're going to Washington Post whether [1:18] you're going to I don't know Minnesota [1:20] Star Tribune or the innovators of the [1:22] space like poly Market perplexity all of [1:26] those are built on versale and so [1:28] increasingly whether it's a start [1:29] startup or with's established Enterprise [1:32] they're leveraging for sale to build [1:33] Innovative products or later in the [1:36] month right like Black Friday e-commerce [1:38] you need to scale you need to serve tons [1:40] of traffic you need to have really good [1:41] experiences so household names like Bose [1:44] or Fanatics Supreme in the up and comers [1:48] the new brands like Axel arato or ragul [1:51] so those companies are all building on [1:52] versel and hopefully delivering products [1:55] that people love all right so sounds [1:57] today like very successful widely [1:59] deployed everyone sort of accepted it's [2:01] this amazing product and platform that [2:03] we should use it was not always that way [2:05] so take us all the way back to 2015 or [2:08] even before and and how did this start [2:11] yeah it's kind of crazy to go back in [2:12] time the company has grown so much now [2:15] but the original idea was simple so [2:18] there were a million ways to build a [2:21] website or a web application there still [2:23] are a million waves to build a website [2:26] or application but none of them that met [2:28] the quality bar of the Giants of the web [2:31] it was almost like for sale was that [2:33] democratization effort if you think [2:35] about companies like meta or Google or [2:38] Amazon they build really fast they try [2:42] lots of things they experiment a lot and [2:45] their experiences are highly Dynamic and [2:47] tailored to every [2:49] visitor but the web at the time was [2:52] actually super static it was really [2:55] difficult for customers to stay online [2:57] the cloud was coming up but the the [2:59] cloud was exceedingly difficult to use [3:01] like it felt like you needed a PhD and [3:03] and this was my personal experience [3:05] because when I started this company I [3:07] didn't set out to build exactly what you [3:08] see now I started trying to build a [3:11] website and I reached for the latest and [3:14] greatest again my aspirations were super [3:17] bold super ambitious it has to be like [3:19] Google or Amazon [3:20] quality and what is the latest and [3:22] greatest at this point in 20145 that [3:25] you're you're thinking oh this might [3:26] solve my problem yeah so Google had just [3:28] open sourced kubernetes an incredible [3:30] sophisticated way of deploying [3:32] infrastructure and managing it and meta [3:35] had open source react which is a UI [3:38] library for it's the engine that powers [3:40] user interfaces The Narrative at this [3:42] time is like oh the web is fast like [3:43] there's react there's kubernetes like [3:45] all this stuff is open source it's out [3:46] there but like exactly but think about [3:49] the average developer or think about the [3:52] company that's going through digital [3:53] transformation what tools do they grab [3:55] for if they grapped those tools they [3:58] would have to spend month [4:01] quarters configuring them assembling [4:03] them into their own platforms but verel [4:07] was born to basically automate all of [4:10] that we wanted to bridge that gap [4:12] between the availability of Open Source [4:15] technology and you creating delightful [4:17] experiences that you could publish [4:18] really fast one distinction also was we [4:22] started out very general you could [4:23] deploy anything on verel but where I [4:26] personally found sort of like this true [4:29] DNA of my own personal Journey was [4:31] focusing on front end [4:33] development so for many many years ever [4:36] since I was like 12 or 13 years old I [4:38] kind of became obsessed with refining [4:41] the side of applications that are [4:43] customer facing I built a platform that [4:46] focus a lot on that so I built the infra [4:49] it delivers your applications globally [4:51] really fast really well optimized but [4:53] also this framework nexs which now has [4:57] 1.3 million monthly active developers [5:01] and nextjs is all about giving you the [5:03] tools and the perfect guard rails to [5:06] build those awesome front experiences [5:08] and it builds on react so it's sort of [5:10] the standard now for how a lot of [5:12] developers build the fastest web [5:14] applications in fact to give you some [5:16] context anytime you use any modern AI [5:20] tool whether it's chbt operator Claud [5:23] Sora mid Journey all of those are built [5:26] on XA so it's sort of become the gold [5:28] standard for front end [5:30] and did you start nextjs you know you [5:32] wrote the first line of code it was an [5:34] open source thing and absolutely it was [5:36] I had a personal frustration to be [5:38] honest because again think of it this [5:40] way the ways of the future had already [5:43] been discovered the papers had been [5:45] published the open sourcing had happened [5:48] thanks to Google meta and many other [5:49] companies but to actually put that into [5:52] practice it felt like you needed a PhD [5:54] sometimes the metaphor I give people is [5:56] look like things like react are like the [5:57] engine and people need a car to go from [6:00] A to B and so next shayz and versel [6:02] became that car or even like that space [6:05] shuttle to the cloud both the lighting [6:07] developers but also businesses because [6:09] for the business if you use NEX sh as [6:11] and forell the website is faster it [6:13] makes you more money it converts at a [6:15] higher rate and I guess looking back [6:17] what made verel unique is that we we did [6:19] both it's like the vertical integration [6:21] between software and Hardware we built [6:22] the tools for the developers and we [6:25] build the highly optimized [6:26] infrastructure in the cloud for it and [6:28] it was somewhat controversial [6:30] the way that people would use the cloud [6:31] back then was oh okay I'll choose AWS or [6:34] I'll choose Google cloud and forell [6:37] builds on top of AWS but it was [6:39] basically like Choose Your Own Adventure [6:41] for how to build the app it's kind of [6:42] silly that you would tell companies well [6:45] go shopping on GitHub for some open [6:48] source framework and spend months [6:50] binding it to your infra that's so funny [6:52] yeah I remember back in those days like [6:54] you there were companies like oh we're [6:56] going to you know host kubernetes for [6:57] you what I mean we funded one when I was [6:59] madro it was actually a great outcome [7:01] but like great going to host kubernetes [7:04] for like what are you going to do you [7:06] know you're an e-commerce company like [7:08] uh thanks I've got kubernetes like that [7:10] doesn't help me you know well kubernetes [7:12] Engineers were fetching half a million [7:15] to a million dollar salaries because [7:17] they were so rare and so valuable and [7:19] all these companies were like well [7:20] apparently we got to figure out [7:22] kubernetes right one way to think about [7:24] it is versel hired them all so you don't [7:27] have to if you're building a cool web [7:29] application or or if you're facing [7:32] strong competition right like you have [7:34] to modernize really quick you have to [7:37] adopt AI features you have to build new [7:40] interfaces you want to cut to the chase [7:42] you don't want to be managing clusters [7:45] you don't want to be writing your own [7:46] Frameworks in fact some of the companies [7:48] that I mentioned it was so interesting [7:50] like they were developing their own NEX [7:52] js's we kind of developed the last [7:54] framework so companies would not have to [7:56] do their own adventure in fact the joke [7:58] I would make at the time was by [8:00] standardizing on xjs now you have access [8:03] to all the stack Overflow questions and [8:06] answers that every other developer has [8:08] run into so for the Enterprise it made a [8:10] lot of sense like standardized on [8:11] tooling you can recruit talent that is [8:14] up to speed with the tool and it's all [8:16] open source and now it's even more so [8:18] because the llms are experts in nextjs [8:22] and react so it's like you're joining [8:24] This Global Zeitgeist of knowledge and [8:27] wisdom okay I'm obsessed with this idea [8:30] of why did the opportunity for any given [8:32] company exist and hopefully this isn't [8:35] too political of a question in the web [8:36] development world but why is it that the [8:40] opportunity for nextjs was there did [8:42] react miss the mark in creating the [8:46] right level of abstraction did they [8:48] create something way too powerful way [8:49] too configurable way too heavy and it [8:52] required someone else to come along and [8:54] build the actually usable interface for [8:56] the masses what a great question there's [8:58] a couple things that happened one was [9:00] that the community took this weird [9:02] detour towards static [9:05] applications which always befed me I'm [9:08] not saying I'm some kind of Genius that [9:09] saw the future but I was so puzzled [9:12] because when I grew up one thing I loved [9:15] doing was reverse engineer how the [9:17] greatest things on the internet were [9:18] built I remember when Gmail first came [9:21] out I just nerded out for months trying [9:23] to understand how it works and so I [9:25] mentioned that versel exists to give you [9:27] the oh yeah didn't you build like a [9:29] websockets Library prior to forell [9:32] socket.io I became obsessed with [9:33] real-time communication I became [9:35] obsessed with democratizing how this [9:37] greatest applications worked if you know [9:41] how Google works if you know how [9:43] facebook.com Newsfeed works and they're [9:46] highly Dynamic that's how they attract [9:49] and retain users why were people [9:51] building static Frameworks for react it [9:54] puzzled me because investors were [9:56] chasing a lot of those Frameworks I was [9:57] like why are you funding that [10:00] and I quietly kept building this nextjs [10:02] thing that was focused on Dynamic it was [10:05] focused on rendering in the cloud as [10:07] opposed to on the device and I also knew [10:11] very firmly that if you rendered in the [10:12] cloud you were taking the burden of [10:16] experiencing a website off of the device [10:19] mobile kept getting bigger and bigger [10:21] and bigger and you just want to give [10:24] people the thing that they want on [10:25] mobile as fast as possible I wanted to [10:27] sort of offload all that rendering and [10:30] compute to the cloud and so that's the [10:32] first Alpha opportunity that I found [10:35] everybody else is focusing on static and [10:38] sort of like the very very fat client [10:40] approach and I focus more on cloud and I [10:43] focus more on Dynamic which is by the [10:45] way way harder for me so scaling up the [10:49] infrastructure and making it secure [10:52] making it [10:53] multi-tenant doing all of that took [10:55] years to perfect so it was a very long [10:57] bet and to your point react was also [11:00] very low level and that also helped us I [11:02] remember talking to a lot of really [11:04] large sites at the time I think it was [11:06] red fin they were creating like really [11:08] cool abstractions on top of react for [11:10] their needs so there was also that [11:12] opportunity to bridge the gap and give [11:14] you the complete tool set to actually be [11:17] productive it's interesting should react [11:20] have had a nextjs like layer built in [11:24] from the start Absol absolutely and [11:26] Facebook actually had it it was just how [11:29] they actually use the tool and this is [11:31] something that happens in open source [11:32] which is and it happens a lot with [11:34] papers like the company will and to [11:36] their credits like doing a great service [11:38] to humanity like they'll give you some [11:41] details and then others that are super [11:43] important ingredients are left out not [11:45] intentionally and if you think about how [11:47] Facebook actually used react they had [11:50] their own nextjs internally it's a lot [11:53] of work to actually maintain massive [11:54] open source projects like this so yeah I [11:56] think ultimately the market spoke [11:59] and what they wanted was Dynamic over [12:01] static and a complete end to-end [12:04] framework with server sight rendering [12:05] over just the lowlevel component and so [12:08] then did you conceive of these two [12:12] things at the same time the library of [12:14] nextjs and then the very complex [12:17] infrastructure for hosting it that would [12:19] become versell I started with the infra [12:21] because I was obsessed with the idea [12:23] that all a developer should do is push [12:27] code receive web [12:30] application anything that sits in [12:31] between in the form of configuration [12:34] creating clusters filing tickets with it [12:37] was nonsense it was pure Alpha for me to [12:39] grab and so first came the infra the [12:43] mistake that I made initially was it was [12:45] too General I was trying to deploy [12:48] everything under the sun and when I [12:51] actually made it true to my story of [12:53] like focus on the front end GMO like [12:56] that's where you know your expertise [12:58] lies in some way you can credit a meme [13:01] to the Genesis of this story there was [13:03] this meme going around that using [13:05] frontend was like solving this massive [13:08] Blackboard of math equations the [13:11] frontend ecosystem loves to be very [13:13] self-deprecating so they make a lot of [13:15] jokes about like their own job and if [13:18] you paid attention to the memes you [13:19] would arrive to nextjs because it was so [13:22] difficult to actually get started [13:25] building this projects that's when like [13:27] the later arrived to like hey I can [13:29] create a frontend framework and it's [13:32] going to play really really well with [13:33] this infrastructure to concretize this [13:36] idea for listeners the show of just [13:39] getting some of the crap out of the way [13:41] where you write code you deploy it you [13:44] receive application back I asked a [13:46] developer friend in prep for this [13:47] episode what was the first sort of [13:49] magical experience you had with versell [13:51] and he said he specifically remembered [13:53] the moment where he realized all of his [13:56] git branches were automatically deployed [13:59] as previews and he could just go and in [14:01] real time play with the code he had just [14:04] written even though he didn't actually [14:06] spin up a whole environment and deploy [14:08] it can you tell me about the origin of [14:10] that feature and did you conceive that [14:13] to be as needle moving as he experienced [14:16] it as a user I didn't realize that it [14:18] was going to be so powerful to give [14:21] people URLs even before they hit [14:24] production so what I knew for a fact is [14:26] that Obsession of I told you like [14:27] keystroke to live application like even [14:31] down to like a mathematical measurement [14:33] like how many milliseconds should it [14:34] take and I was going to bring it down [14:36] from weeks which I had experienced in my [14:38] previous job from weeks in conversations [14:41] with humans to a developer pressing [14:44] enter on a terminal or running git push [14:47] that was my obsession what I [14:49] underestimated and it's been a wonderful [14:51] gift for our business is that companies [14:55] value even more the ability for their [14:58] developers to collaborate with other [15:00] personas in the [15:02] organization so when that git Branch [15:04] gets automatically deployed and you have [15:05] that URL now everybody in the company [15:08] can participate in building software oh [15:11] it ends that horrible thing that I [15:13] remember going through when I worked as [15:15] a developer where you're like yeah I can [15:17] only show you on my machine and actually [15:20] my machine is doing something weird [15:22] right now let me come over to your desk [15:23] in an hour holding my laptop hovering [15:26] over your desk so you can see what I'm [15:27] talking about there's a lot of silly [15:29] things that still happen like developers [15:31] fighting one staging machine there's a [15:35] huge security Risk by the way to Dev and [15:37] test environments in the cloud that [15:39] they're like computers that are like [15:41] just there with developers running tests [15:44] in them whereas vel's preview [15:48] environments this is what we call these [15:49] URLs are completely ephemeral and [15:52] secure and so you have all these horror [15:54] stories of developers having to wait for [15:57] hours in order to shows somebody [16:00] sometimes is the developer shows their [16:02] client as a freelancer sometimes his [16:04] companies cooperating internally [16:06] collaborating internally the other one [16:09] that's crazy is you know I had a [16:10] conversation last week with a CIO of one [16:12] of the largest banks in the world and [16:15] they don't have this technology so [16:17] instead what they do is they set up [16:19] meetings they say okay next week let's [16:22] coordinate the calendar of like 12 [16:24] people and someone will screen share [16:27] their machine to show the feature [16:29] they're working [16:31] on yeah it's nuts and they don't even [16:33] get to test it on on the real thing that [16:35] they should be testing on you can take [16:37] this URLs I can put it on my phone and [16:40] now I can experience what the end user [16:42] will experience which is you know most [16:44] people on the internet are on their [16:45] phones most developers are on their [16:47] desktop it's kind of silly one of our [16:49] largest Ecom customers was telling me [16:51] like 90% of my traffic is mobile thank [16:55] God for those URLs I can load up the [16:57] site before it goes out on that phone [17:01] and have the developers face reality [17:04] developers sometimes like we are the [17:05] most coddled people in the planet [17:07] sometimes besides the you know Cate [17:09] lunches and whatnot we have this massive [17:11] Mega computers you know the M4 M5 power [17:16] super Mac and then okay are you actually [17:19] relating to that end user that's going [17:20] to visit your campaign your application [17:23] your internal tool most likely in a [17:25] mobile device and versell is basically [17:27] solving that problem as well that's [17:29] great Okay so we've been living in this [17:31] like early versel company history land [17:34] there's always a few magic moments in a [17:37] company's history as we move through the [17:38] progression to today where you really [17:41] went from feeling like I don't know if [17:43] this is going to work to oh my God [17:45] people love it and I now believe this is [17:47] going to work what were some of those [17:49] along the way for you I think the most [17:53] precient one that blows my mind still to [17:55] this day is the launch of [17:57] vzero which is our AI assistant for web [18:00] development it's crazy you're giving me [18:02] an example nine years into the company's [18:04] history yeah when did this happen yeah [18:06] so vzer is truly like a startup within a [18:09] startup the way that it happened is this [18:13] large language models became really good [18:14] at [18:15] coding they became really good at our [18:17] tools as I mentioned they've been [18:19] trained on our tools they've been [18:20] trained on nextjs react is truly a [18:23] blessing but we saw an opportunity to [18:25] sort of disrupt how web applications are [18:27] built so vzer transforms prompts so [18:32] English descriptions into working [18:34] applications and user [18:36] interfaces and it's a true disruption in [18:38] that it's not code first it's code last [18:42] anybody with an idea can cook you come [18:45] to vzer you type in what you want it's [18:48] time to cook and so to give you context [18:50] on this right like we launched it as a [18:52] research preview in September 23 we're [18:57] very early because we're so excited [18:58] about what language models were able to [19:00] do with code it took us 10 months to get [19:03] to 1 million in AR and then 14 days to [19:08] get to 2 million and then 20 more days [19:11] to get to 4 million and that pays [19:14] continues and the trajectory is [19:16] something that I've never seen in my [19:17] life before and I've been involved with [19:19] successes like nextjs and whatnot I mean [19:21] this is to the point where we are that [19:23] company that is making the hyperscalers [19:25] run out of GPU capacity in the world's [19:28] largest regions so it's been fascinating [19:30] to be a part of this okay so how does it [19:33] work why are you running out of GPU [19:34] capacity so think of it this way to [19:38] create a nextjs application we just made [19:40] it as easy as possible for developers [19:43] but it still takes what like several [19:46] steps setting up tools whatever now you [19:48] just open a website you go to v.d you [19:52] type in a prompt and all of that process [19:55] that before was reserved to a couple [19:56] million developers that exist on the [19:58] planet now the Topp of final is hundreds [20:01] of millions and of billions of people [20:03] anybody can create software with this [20:05] and the kicker is this and this is was [20:07] my H moment with this tool by the way [20:08] I've been doing frin since I'm almost in [20:11] diapers like that's how crazy like you [20:13] like I've been in the arena for a [20:15] while it can create code better than [20:18] what I can do by hand one of the H [20:20] moments was I recreated my own personal [20:23] blog which of course I wrote with nexs [20:26] and the output of what vzero creat it [20:29] was better it was better in a responsive [20:32] way so like it was better in both [20:34] desktop and mobile and it had better [20:37] accessibility [20:38] capabilities I had forgotten to set up [20:40] some configurations and some tags and [20:42] some code that came naturally to V VZ [20:45] knows everything and remembers [20:46] everything the rate of application [20:48] creation is something that has maybe [20:51] like quadrupled for versel signups are [20:54] through the roof and it's also because [20:56] this top of the funnel is people that [20:59] before couldn't make sense of versell [21:01] versel is infrastructure at the end of [21:03] the day right it's really great for [21:04] developers this is great for everybody [21:06] so funny so versel you have a corporate [21:09] journey ahead of you of going from a [21:12] developer brand to almost like a [21:13] consumer brand I think developers will [21:17] become this more common skill if you [21:20] will development [21:22] itself think of it as like anything that [21:25] was done by humans that then machines [21:29] took over like [21:30] translation Google Translate is a really [21:32] good example for me because first of all [21:35] Google translate sort of originated [21:36] Transformer architecture as an [21:37] interesting side effect but taking [21:40] something in language a like Argentinian [21:42] Spanish and converting into English no [21:44] longer requires human specialization a [21:47] lot of software processes software [21:48] development processes are like that [21:50] they're like translating intent into [21:53] working code and my take is that [21:56] development will go from being a role [21:59] like translator was a role to a [22:02] skill that AIS will increasingly [22:05] dominate AI are awesome at summarization [22:08] translation and they're awesome at [22:10] writing code now what's really cool too [22:12] is that the taste of the human being the [22:17] direction the human being sets and the [22:20] creativity of the problems that they [22:21] want vzer to solve those are still [22:23] intact those are extremely valuable and [22:26] the other thing we're seeing at versel [22:27] is that we've ins insired other [22:29] companies to pursue and create their own [22:31] vzer literally just a couple days ago I [22:34] heard about the vzer for CAD it's going [22:38] viral in Social right now computer [22:39] automated design correct so like for [22:42] Architects and for 3D printing and [22:45] whatnot that company is using for sale [22:48] because we've open sourced the secret [22:50] sauce of how we built vzer so vzer has [22:53] been this double interesting disruption [22:55] right on one level it's making for [22:57] celles so approachable to everybody and [23:00] on the other hand because its [23:03] infrastructure components are open [23:04] source it's inspiring other companies [23:07] and startups to go after the disruption [23:09] of other Industries you tweeted the [23:11] other day about ezero email zero and I [23:13] was I saw that tweet like oh hell yeah [23:16] vzer for email vzer for CAD v0 for [23:19] medicine in fact there's a vzer for [23:21] medicine called open evidence that is [23:23] built with NEX shs and verell there's [23:26] this open source framework that we [23:28] create called AISD K so think of it as [23:31] the tools to build your own vzer and so [23:35] it massively benefits versel if the [23:38] world gravitates towards this expert Ai [23:41] and by the way this is not just business [23:43] strategy of sorts I remember when Jensen [23:45] said most companies will refactor [23:48] themselves into token factories you have [23:51] some expertise you have some taste you [23:53] have some opinion of the world within [23:54] your organization for us that was web [23:56] development design engineering ing so we [23:58] created VZ there's going to be ezo for [24:01] email there's going to be Finance vzer [24:04] so I'm really really excited about [24:06] what's happening right now and it's not [24:07] just our success it's a your success of [24:09] the community what is vzero did you guys [24:12] create your own foundational model what [24:15] is your infrastructure and what actually [24:17] is the product that you're providing [24:19] yeah so many years ago I tried a product [24:21] that failed and it's called nextjs live [24:24] so nextjs live was the idea of helping [24:27] you run the entire development [24:28] environment of nexj inside a web browser [24:31] and my aspiration was because front end [24:34] is all about what the user sees can we [24:38] actually cut out the design process can [24:40] we design with code and it turned out [24:43] not to work so well because even though [24:46] we're targeting this Persona that is [24:48] very visual you still needed to learn a [24:50] lot of code so vzer is like us learning [24:54] from the past or us trying next shs live [24:56] again it was like our Newton and this is [24:58] our iPhone right part of vzer that's [25:01] very unique is that we can run the [25:04] rendering environment we can run sort of [25:06] the foundations of nexj and versel [25:08] inside the web browser and we combine [25:10] that with the state-of-the-art [25:11] foundation models that we steer towards [25:14] becoming experts in web development so I [25:17] mentioned how vzer is an expert in web [25:20] accessibility vzer people tell us has [25:23] good taste like the design that it [25:25] outputs looks better than the average [25:27] thing that sometimes people call slop [25:30] instead of outputting slop we created [25:33] all these guard rails in the form of [25:34] evals and benchmarks and prompts and rag [25:38] we even creating Custom Design Systems [25:41] that are tailored for AI and by the way [25:43] this is speaks to the disruptions going [25:45] to happen even in the framework space [25:48] Rouch G myself from 10 years ago maybe [25:51] wouldn't be so successful today because [25:53] I was creating Frameworks for humans [25:56] what we're doing now is starting to [25:57] create Frameworks for a a design systems [26:00] for AIS just to make sure I understand [26:02] through these rules and evals and [26:05] prompts and trying to remember [26:06] everything else you said rag are you [26:08] imbuing [26:10] versel employees taste into it to make [26:14] sure that the output always adheres to [26:16] The Taste level of your company when I [26:18] created next and it became so successful [26:20] literally every company in the planet [26:22] was begging me for support support and [26:25] training support and training support [26:26] and training and I was like guys I'm [26:29] creating infrastructure over here I want [26:31] to scale myself I don't want to go and [26:33] teach everybody one by one how to hold [26:36] this thing and that dream remained [26:39] elusive for many years the idea that I [26:41] could go myself to every company and [26:43] you've probably seen that I still go to [26:45] a lot of conferences I meet with a lot [26:47] of developers I meet with Enterprises [26:50] but it just doesn't scale I want to send [26:52] my best Engineers to this company you [26:54] know we employ the tech lead of react we [26:56] employed the next sh team the zelt team [27:00] all these open source projects and these [27:02] people have pretty full inboxes as you [27:05] could imagine they support tens of [27:07] millions of developers and so you're [27:09] exactly right we are IMB the preference [27:13] data the demonstration data of what they [27:17] think is the best way to build software [27:20] so it's our answer to how you actually [27:22] make support advice audits migration set [27:28] scale sustainable and efficient a lot of [27:31] people are using vzer because it's [27:33] really good as an emergent property to [27:35] migrate from Legacy Frameworks to nextjs [27:39] in modern technology oh which I imagine [27:42] had to be a sales hurdle for you for a [27:43] long time of they really want to but [27:46] gosh they just can't find the dev [27:48] schedule time to actually rebuild the [27:50] whole thing on verel another hum moment [27:52] and I literally made a fellow industry [27:56] practitioner jaw drop the other day [27:58] is that you can get on a call with a [28:01] prospect and you can build the web [28:03] application of their dreams in that call [28:06] it's just nuts we take a screenshot of [28:08] their current website built on Legacy [28:11] technology we throw it into vzero and [28:14] outcomes the high performance version [28:16] that otherwise would have been a sales [28:18] pitch it would have been a trust me bro [28:21] we can do this together we can bring in [28:23] an agency literally it's happening [28:26] within calls so say sales cycles for [28:29] verell are shortening customer delight [28:31] and trust is going up they share this [28:34] links the way that people work is [28:36] changing a lot of our customers are now [28:38] sending these vzero links the groundhog [28:40] day of myself is like how can I make [28:43] people collaborate with links instead of [28:45] words or documents or whatever right so [28:48] this vzer links are how the [28:50] conversations are happening oh can you [28:52] change this can you change that okay [28:53] here's the new vzer for this so for the [28:56] sales team Solutions engineering team [28:58] those people they feel so [29:00] empowered they feel like they can [29:03] actually cook as opposed to wait until [29:06] you know arrangements are made and and [29:08] contracts are signed we get to software [29:11] much much faster it's such an [29:13] interesting fork in the universe of your [29:14] company because what if llms hadn't [29:18] happened at the moment that they [29:19] happened you guys would still be a [29:21] plenty successful company but your time [29:24] would be spent differently the company [29:26] would be moving in a different direction [29:27] right now this opened up this whole [29:29] different world for you I think if llms [29:31] hadn't happened you know I almost felt [29:34] like we were going to hit diminishing [29:36] returns on all those developer [29:39] experience ideas that we had so when you [29:41] think about developer experience to make [29:43] it very crisp to listeners developers [29:46] need very fast feedback [29:48] loops if not they're the most the people [29:52] on the planet they get distracted they [29:53] get out of the Zone they lose that flow [29:57] you know mythical flow flow Zone that [29:59] they love and so nextjs and versel [30:01] obsessed with like okay what are the [30:04] classical software engineering [30:06] techniques that we can use to optimize [30:08] those processes for example we partnered [30:11] deeply with aws's serverless [30:13] infrastructure team for those deployment [30:15] previews instead of creating massive [30:17] amounts of redundant infrastructure and [30:18] clusters that cost millions of dollars [30:21] we create serverless functions under the [30:23] hood it's a very specialized [30:25] infrastructure [30:26] Innovation but at some point we're going [30:28] to hit the limits literally of the speed [30:31] of light that's how fast we're going to [30:33] give developers feedback and we're still [30:35] far away from that by the way like [30:37] hitting those limits is like we saw that [30:39] supersonic playe yesterday it's hard [30:42] engineering to get to the limits of [30:44] physical reality but I saw the writing [30:47] on the wall and I was very excited about [30:49] it by the way because it was like I [30:51] talked to a Cutting Edge database [30:52] company the other day the feedback loops [30:55] for their developers today are 60 [30:57] minutes [30:58] they're doing a versel POC it cuts them [31:01] down to two so to make something that [31:04] fast for some of the highest leverage [31:06] individuals on the planet best paid [31:08] individuals on the planet is incredible [31:10] returns for the organization but the [31:11] writing was on the wall that we can [31:13] bring it down to two we can bring it [31:15] down to two seconds and then at that [31:17] point the limiting factor becomes how [31:19] many humans on the planet can do this [31:22] job and this is what VZ is [31:24] revolutionizing we can bring so many [31:27] more people into the world of building [31:28] software and now they can pig it back on [31:30] this infrastructure has been battle [31:32] tested by some of the largest companies [31:35] on the planet it's fascinating I'm sure [31:37] we'll come back to a lot of this vzer AI [31:40] stuff I do want to ask you some company [31:43] building topics sort of advice for other [31:45] Founders out there there's a lot of [31:48] Engineers who have a dream of creating [31:50] something that is two-pronged an open [31:53] source project and a for-profit [31:55] corporation the devil is in the details [31:58] and many people fail at doing this well [32:01] what would be your advice for people on [32:03] how to think about doing that [32:05] well my number one priority is to be [32:08] excellent at of one thing I want to be [32:11] world class at a one thing people call [32:13] it like dominating a niche so when I [32:16] started out verell and then I focused on [32:19] making the fastest websites the fastest [32:21] front ends it fastest the storefronts [32:24] that's when things started working for [32:25] me being a jack of all trades doesn't [32:27] work in this world both developers and [32:30] the most Discerning Enterprises they [32:31] just want the best product in the [32:33] category and versel became that for [32:36] front and from there we can expand [32:39] outwards and provide a lot more services [32:41] to developers we call it the transition [32:44] from being a frontend cloud to being a [32:46] developer Cloud you know developers are [32:48] asking themselves for more from versel [32:52] and I think I had it upside down in the [32:54] beginning I thought well I have to give [32:55] them everything instead it's actually [32:57] much better establish a dance with a [32:59] customer you give them enough that they [33:02] love the product and then they should be [33:04] the ones pulling product out of you and [33:08] so Developers for many many years would [33:09] say like can I get a database on versell [33:12] you would say we're not ready yet we're [33:13] not ready yet can they do more backend [33:15] on verell we're not ready yet we're not [33:17] ready yet and now we're ready because [33:19] you viewed yourself as very tightly [33:21] coupled to nextjs for you know the first [33:24] six seven years of the company right [33:25] exactly we focus a lot on nextjs in [33:27] front and Frameworks that had the same [33:29] shape if you will there's others for [33:32] example we support zelt that empowers [33:35] companies like Logitech on versel it [33:37] Powers companies like Ikea so there's [33:40] others it's not just nexj but this were [33:42] all front and focused Frameworks and [33:45] what's happening now is people are [33:46] saying well I want to build more I want [33:48] to build more of the backend for my [33:50] checkout process I want to build the [33:52] entire back end for my SAS application I [33:55] want to build the backend for how I [33:57] process the tokens coming out of the AI [34:00] models and now vercel is giving them [34:03] that but we started out focused and then [34:04] expanded from there and so that I think [34:07] is something a lot of entrepreneurs Miss [34:09] which is they feel like the initial [34:12] presentation of their product has to [34:15] match one to one the size of their [34:17] ambition and our ambition was to disrupt [34:19] all applications even not just web and [34:21] beond so anything that runs on a phone [34:24] for example but if we didn't start with [34:27] some something that was world class and [34:29] small in the beginning I think we [34:31] wouldn't be here today and to maybe ask [34:33] the question a different way there's a [34:36] great business model around versell and [34:38] there's zero Revenue generation around [34:41] the open source project of nextjs did [34:43] you ever think about doing it [34:45] differently never absolutely not so open [34:48] source for me has always been my [34:50] life when I was in Argentina open source [34:54] was sort of my ticket out of Argentina [34:56] and into the world I wanted to build the [34:58] best possible technology I couldn't [35:00] fathom as a developer today if I'm [35:02] starting an application imagine if the [35:03] framework was proprietary it just [35:04] doesn't make [35:06] sense and we're doing it again by the [35:08] way so I mentioned the AISD K we saw an [35:11] opportunity to democratize building [35:14] applications like chat GPT we knew that [35:16] there wasn't going to be just one AI [35:17] rule at all right the web is all about [35:20] decentralization it's about you have [35:23] your own domain name literally your own [35:26] domain where you set the rules and [35:28] versel is so aligned with making the web [35:31] better making it reach more places more [35:34] surfaces the deployment model of the web [35:37] the deployment model of the web is that [35:38] you don't have to ask permission you [35:39] don't have to ask for a license from [35:41] anybody and web is where disruption [35:44] happens to right so chat PT was a [35:46] website built with nextjs it was a [35:49] research preview that someone put out [35:51] into the world and so I'm very excited [35:54] to continue to do this so AISD K to give [35:57] you contest went from 100,000 downloads [35:59] a week to now almost a million downloads [36:03] a week in about a year is the second [36:07] most popular way of getting AI into a [36:10] JavaScript or front end application [36:12] after open AI second largest AI module [36:15] and there's actually no Direct business [36:17] model for it we want the developer [36:20] adoption we want their feedback we want [36:21] their Community we want to get our brand [36:24] out we want to learn we want to learn [36:26] from as many people as possible possible [36:28] and then when they decide to build and [36:30] deploy their application they'll [36:32] probably choose verell because they know [36:35] that we have sweated the details of the [36:37] parts of infrastructure that they don't [36:39] care about some people do on some level [36:42] there is an a keyi effect too of like I [36:45] enjoy the process of learning a metaphor [36:48] that I view sometimes is that you know [36:50] you have to do it at least once so that [36:52] you learn what it's like to create this [36:54] infrastructure as like going to school [36:57] but if you think about the successful [36:58] companies out there that share the same [37:01] business model as versel like snowflake [37:04] like data bricks like data dog the value [37:08] they give to the world is that you've [37:10] given them the heavy lifting of [37:12] infrastructure and you pay for what you [37:14] use if an application on for sale is as [37:16] small then you don't pay a lot if an [37:19] application takes over the world you pay [37:22] more so it's a very fair deal to [37:25] developers and they understand exactly [37:28] what they're paying for okay so one more [37:30] question along this line I'm a Founder [37:32] I'm considering something with this [37:33] business model would you caution me [37:35] against taking it all the way to the [37:37] logical extreme where I say every single [37:40] thing I do is open source and I'm the [37:43] best in the world at you know creating [37:44] the infrastructure and the hosting and [37:45] the deployment so I can just trust that [37:48] I open source everything that I do and [37:50] I'm not giving away too much of the [37:51] golden gooth the bottom line is that [37:53] part that I just mentioned most [37:56] companies in the world want to off load [37:58] the infrastructure burden to verell so [38:01] having that piece be open source so that [38:03] we teach you how to set up 20 Global [38:05] regions of infrastructure with hundreds [38:08] of microservices each open source [38:10] actually hits diminishing returns there [38:12] it gets so mind boggly complex and most [38:15] of the software engineering that goes [38:17] into that by the way is [38:18] operational rather than based on the [38:20] source code it's the monitoring is the [38:24] upgrades is the rolling systems is the [38:27] staging techniques actually I would [38:30] argue that's not where open source even [38:31] shines for that matter open source [38:34] shines when it gives you a programming [38:36] model so nextjs defines and declares the [38:40] application the way we think about this [38:42] ADV verel which is actually quite [38:43] disruptive [38:45] is infrastructure has to be an output of [38:48] the application most companies that [38:51] reach for the cloud start with the infra [38:54] we're inverting that we tell you here's [38:57] the nextjs framework focus on the [38:59] application you're going to be [39:00] multicloud you're going to have the [39:02] ability to self-host it yourself or [39:05] we've also taking that burden here's [39:06] versel as an option and versell doesn't [39:08] need to be open source I think it [39:09] wouldn't bring value to that audience [39:12] when you were creating the business [39:14] model were there any companies that had [39:16] perfected this that you like oh this is [39:17] the inspiration we should just do that [39:19] or how much of this was Innovative cu [39:20] the open source business model basically [39:23] used to suck there were successful [39:26] companies out there you know like red [39:28] hat the company around it is just [39:29] support you're just selling support it [39:30] doesn't scale totally there weren't a [39:33] ton of examples to be honest I think the [39:35] closest thing is probably data breaks or [39:38] GitHub GitHub is a lot about [39:41] collaboration and it builds on git as an [39:45] open source component they didn't create [39:47] it so I think that's a it's a big [39:48] difference but they give you this URLs [39:50] to code so there is a huge parallel [39:53] there instead of verell gives you URLs [39:55] to working [39:56] applications I'll argue that code over [39:58] time will become less important as it [40:00] becomes the domain of AIS but GitHub is [40:03] a phenomenal example of that part of the [40:05] pre-production side of our cell and then [40:07] on the infrastructure side companies [40:09] like data brakes or snowflake where in [40:11] the data brakes case you adopt an open [40:14] source component like [40:16] spark and then they have a highly [40:18] optimized compute engine for that [40:21] component in the cloud and you can [40:24] totally do the DIY yourself but scaling [40:26] global clusters of compute very [40:29] difficult job that's sort of the secret [40:31] sauce of versel for the parts of the [40:34] application that deal with the run the [40:36] serving the securing of the application [40:39] the DDOS [40:40] mitigation the making sure that it stays [40:43] up during massive spikes of traffic [40:47] that's a beauty of sell it is a [40:49] beautiful business model there's [40:51] actually like a lot of stuff you got to [40:53] do for this to be viable right like for [40:56] corporations to trust you to be their [40:58] infrastructure you can't just be like Oh [41:00] I'm a seed stage startup you know and [41:02] like uh Hey Supreme come trust me with [41:04] your infrastructure like what were [41:06] inflection points there like when did [41:08] you feel like Whoa We got a customer [41:10] that it was hard to get them to trust us [41:11] and now they trust us I feel like it's [41:14] when it rains it pors cuz it's now [41:16] happening a lot you know there's all [41:17] these banks that are migrating to versel [41:20] I think by the way I'll give credit to [41:22] the crypto space so what happens to [41:25] verell by the way is whenever there's a [41:27] new wave of technology and this has been [41:29] a blessing and also sometimes a [41:31] challenge anything that's new and [41:32] disruptive people tend to build it on [41:34] for sale why because they're motivated [41:36] by the speed at which they can get their [41:38] idea out so if you're building on crypto [41:40] and you're building on AI most of the [41:42] startups are choosing for sale crypto [41:45] though was fascinating because it [41:46] attracted literally the worst darkest [41:49] actors insid of the [41:53] internet so it really made us triple [41:56] down quadruple down 10x on security [42:00] services and products from the careful [42:04] encryption of the vaults of how people [42:07] store the secrets to connect to their [42:09] databases and backends or wallets and [42:13] all kinds of cryptographic details that [42:16] these companies need as well as the [42:19] traffic the hardest most difficult thing [42:22] about the internet and running [42:23] infrastructure is that you have to [42:25] discern between good traffic and bad [42:28] traffic and we set out to do security [42:30] the versel way the versel way is that [42:32] it's zero configuration is secure by [42:34] default and we needed to be secure by [42:37] default because I think something is [42:39] different between us and the CDN for [42:42] example for context versel builds in the [42:45] CDN and WAFF capabilities that you [42:47] typically would have to like go to [42:48] another infrastructure provided to to [42:51] get AC Cloud Player blah blah blah yeah [42:53] the difference between the Legacy ones [42:56] of the on cloud ones or the outside the [42:59] cloud ones like Akamai Cloud flare is [43:00] that because they don't host the [43:03] application you know they filter out all [43:05] the traffic it's hard to sa they're even [43:07] doing a good job sometimes whereas [43:10] versel has to filter out a 100% of the [43:14] malicious traffic instantly otherwise [43:17] it's your own problem correct that your [43:20] incentive is aligned to be a really [43:21] really good CDN correct crypto really [43:24] put us through the ringer and [43:28] and I loved it to be honest you know I [43:30] look back and so for example Solana is [43:33] really hot right now Solana host all of [43:35] their web properties we're like the Web [43:37] Two Face to web 3 for a lot of these [43:40] great blockchains and crypto companies [43:43] all of the large nft companies you [43:46] picked for sale and you could imagine [43:48] also just the massive upswings of [43:50] traffic right like there is a meme coin [43:53] that that goes hot there is an nft that [43:55] goes hot there's a new cryptocurrency [43:56] that goes hot hot but it bolstered our [43:59] security to the point where we have [44:01] industry-leading response times for [44:03] example on DDOS mitigation we do it [44:06] within seconds our competitors do it in [44:08] minutes sometimes and we see this [44:11] because sometimes customers layer on [44:13] another CDN in front of our cell which [44:15] is not necessary and so we see the [44:17] attacks coming from the CDN and we're [44:19] like what you're were supposed to [44:20] protect the customer that was fun and [44:23] yeah you learn a lot by being exposed to [44:25] all of this iess to say there is a [44:27] delicious irony to the fact that these [44:31] Fortune 500s that are signing these big [44:33] multi-year Enterprise contracts with you [44:36] and effectively using for s for their [44:38] own digital transformation and changing [44:40] their workplace and their culture is [44:41] benefiting from being hardened by dos [44:45] attacks from the crypto world but [44:46] there's another fascinating effect the [44:48] way that we've gotten into large [44:50] Enterprises sometimes is because they [44:52] acquire no P intendent acquire podcast [44:55] they acquire the disruptor they acquire [44:57] the innovator I wouldn't be surprised if [44:59] there is you know in the next 10 years [45:01] more of a merger between crypto and [45:03] traditional Finance that's happened a [45:05] lot with AI you know n xjs has become [45:09] the common language that speeds up m&a I [45:12] also give credit to react for this when [45:14] you acquire a nextjs powered company [45:16] your integration times go down to like [45:19] what you copy and paste some files into [45:22] another repository why because the [45:24] infrastructure is an output of the [45:26] application literally the job of [45:28] integrating companies becomes [45:30] integrating the two front ends the [45:32] backend services are pluggable the more [45:35] people stand their eyes on this tooling [45:37] there's this like greater than the some [45:39] of the parts effect that happens and [45:41] we're starting to see it in the form of [45:42] m& where the startups get really really [45:45] really big and intimidating to their [45:47] incumbent competitors and they get [45:49] bought and of course there's also [45:51] Enterprises realizing I need to hop on [45:53] this train this is how I get Talent this [45:55] is how I retain Talent this is how I Mo [45:57] faster it's been awesome to see that we [45:59] have Global impact and it happens with [46:02] the brands that I love my favorite [46:04] automobile company from Germany uses for [46:07] sale and it's just delightful so all [46:10] right which [46:11] one I think we don't have logo rights [46:13] for that one but like uh have we covered [46:15] it on acquired or not you could probably [46:17] extrapolate from vel's sort of design [46:20] and taste yeah that's pretty awesome [46:23] great we we won't push you into [46:25] something uncomfortable there [46:27] one question that's been a fun one for [46:29] us to ask here on aq2 the last few [46:31] episodes is what is something that you [46:33] believe as a leader or a founder or a [46:36] technologist that you have real [46:38] conviction in and you feel others don't [46:40] yeah I think there's an easy one for me [46:42] here because I experienced it over the [46:43] last couple weeks that reaffirmed my [46:46] conviction I think CEOs especially [46:49] technical CEOs can and should be [46:52] involved in every aspect of the [46:55] business whe whether it's closing a deal [46:59] whether it's looking at the numbers of [47:01] the last quarter and tuning sales quote [47:04] as in sales [47:05] compensation whether it's informing [47:08] future product road map whether it's [47:10] pricing [47:11] discussions I involved in every aspect [47:13] of the business and and how many [47:15] employees are you [47:17] 550 okay maybe also for context versel [47:22] is now at about well last number we [47:25] disclosed was we hit $100 million in [47:27] annualized Revenue May [47:28] 2024 growing 80% year-over-year in [47:32] revenue and developers using verell more [47:36] than doubled over the past year so we're [47:39] operating at a significant scale and I [47:41] think the role of the CEO and the [47:43] conception of the CEO is evolving and [47:45] changing companies are becoming a lot [47:47] more transparent it's about optimizing [47:49] the information flow within the company [47:52] rather than creating silos of [47:54] information we're doing it a little bit [47:56] different here at versus sell how do you [47:58] stay that involved with 550 [48:01] employees I think using our own product [48:03] has been a massive acceleration I don't [48:07] need to get brought up to speed on [48:09] things I'm in direct connection with the [48:12] developer Community I talk to [48:15] customers and whenever I find time I use [48:18] vzero I use versel I'm staying very [48:21] grounded on on the reality of the [48:24] product so there is just no time being [48:26] wasted on like briefing Executives the [48:29] expectation is that you understand the [48:31] customer deeply you understand the [48:32] product deeply we try to hire people for [48:35] every area of verell that have a [48:37] technical background that have used [48:38] verell in the past if possible and so we [48:41] tend to move really fast we optimize our [48:44] entire product and platform around [48:46] product velocity right so the way that [48:48] the company behaves is the way that the [48:51] product [48:52] behaves when I'm seeing success with [48:54] some of the largest more established [48:56] Brands house names when the cios CTO [49:00] CEOs come to me they don't want a [49:03] specific aspect of our technology they [49:05] want us to teach them how we build they [49:08] want us to export how versel Works they [49:11] can tell that we're moving fast and so [49:13] by continuing to make the company and [49:16] the product a sort of symbiote a two-way [49:20] street I think this is how we're saying [49:22] really fast and lean makes sense so an [49:26] outside Observer would then say there's [49:28] got to be some trade-off here how can [49:30] you possibly hire Executives that you [49:33] want to empower as much as possible but [49:35] you also want to be extremely involved [49:37] how do you deal with that trade-off I [49:39] think the fast exchange of feedback is [49:40] huge right of course the executive is [49:43] the one that's on the line that is [49:44] responsible for a set of numbers kpis [49:48] outcomes whatever you want to call it [49:50] but I'm involved in giving them high [49:52] quality feedback that's my number one [49:53] job I'm the bridge to reality in many [49:56] ways I bring the customer stories from [49:58] the outside world I hold them [50:00] accountable to that I put them on the [50:02] line you know like you have to be able [50:03] to use our product to Dem or product and [50:07] so it's not that I take their job is [50:10] that I'm able to get on the same [50:11] altitude I go into the texture of of [50:14] what they're supposed to do so I think [50:17] it's rooted in [50:19] understanding rather than Outsourcing [50:22] the labor and a principle that I use [50:24] internally a lot is as a CEO you have to [50:28] understand you can't just say well this [50:31] area of infrastructure I have no idea [50:33] how it works talk to Johnny and so I [50:36] spend time learning and that's how I can [50:39] give high quality feedback and that's [50:40] how I can you know delegate with [50:43] confidence okay so in some ways for the [50:46] revenue scale and customer scale you're [50:48] operating at I would reframe wow you're [50:50] 550 people is you're only 550 people now [50:54] a decade ago you would have been way [50:56] more yes I have a concrete example for [50:59] that so vzer was amazing I called it a [51:02] startup within a startup because we said [51:05] the will be entirely full stack for sell [51:09] let's imagine that another company is [51:11] innovating and they only want to use [51:13] versel to build the idea top to bottom [51:16] front end back end database AI model [51:19] procurement everything is happening [51:21] onell and so that team is just tiny the [51:26] Vue figure per head is astonishing and [51:28] it's a proof point that if you build on [51:31] the versel platform you can be so lean [51:34] you know a thing that I monitor [51:35] obsessively is oh wait I'm sorry is [51:37] everything that that team used a [51:40] publicly available product from versel [51:42] so theoretically vzer could have been [51:45] written by an outside team of just a few [51:47] people corre yes like less than 10 [51:50] people and it's one of the fastest [51:52] growing products in the industry right [51:53] now and so the thing that I monitor very [51:56] very very neurotically is they're not [52:00] allowed to cheat they're not allowed to [52:02] go to the infra team and say oh I'm I'm [52:04] having a million requests per second [52:06] right now can you provision some special [52:08] thing for me zero and that gives me the [52:11] confidence that our platform has no [52:13] graduation risk this is the sort of [52:16] scary ghost of every startup that tries [52:18] to simplify things yes oh I'm so glad [52:20] you went here you know oh I'm going to [52:22] outgrow verel so for as long as I'm [52:25] alive and I'm in the the details for my [52:27] previous answer I know that all of the [52:31] scape hatches all of the Advan [52:34] capabilities all of that is actually [52:36] being provided by the platform itself [52:38] and that has allowed that team to be [52:40] just tiny and so I can project it can [52:43] forecast that there's going to be [52:46] companies that are going to be born on [52:47] this platform I mean I already have [52:49] examples right like I like to use po [52:51] Market as an example of like a tiny team [52:54] that during the election saw traffic [52:57] that would scare infrastructure teams [53:01] it's traffic that you have to have been [53:03] around the internet for decades to be [53:05] able to live up to and to sustain and so [53:10] I think teams are probably going to get [53:11] smaller by the way I mean this would be [53:12] my prediction if you have ai generating [53:15] a ton of [53:16] code and the infrastructure abstracted [53:19] away and there's no sort of [53:22] graduation thing that you need to do [53:25] when you get big and if you can stay big [53:28] with this level of Simplicity then the [53:31] downward pressure will be on head count [53:33] but not in a sense of like oh people are [53:35] going to retire like developers are [53:37] going to retire into a beach is that [53:39] we're going to see more products more [53:40] product creation that reach more people [53:44] built by fewer people how fast and at [53:47] what scale okay so I'll give you a [53:48] really good good [53:50] example I think it was last week that [53:52] Operator by open AI came out within 24 [53:56] hours [53:57] open operator launched it was built by a [54:00] versell [54:02] customer they used vzer create the [54:06] interface they used browser [54:08] infrastructure that exists in the cloud [54:10] to sort of automate the operator needs [54:12] to use a web [54:14] browser and they used models like deep [54:17] seek and others to sort of like [54:19] provision the agent so the fact that you [54:21] can go from the idea of okay so there's [54:24] this new idea agents can control a web [54:27] browser and self task for [54:29] people and to put it out into the world [54:32] it can be 24 hours and that thing sort [54:34] of like took over the internet I tweeted [54:36] about it and I was getting emails from [54:38] people in China saying I'm losing my [54:41] mind that that this is a thing how did [54:43] this happen so fast so it can be hours [54:47] it can be days it can be weeks of course [54:49] you have to sustain Innovation right [54:50] like you can just like launch something [54:52] goes viral and then you don't have the [54:54] thing to follow it with but it's amazing [54:57] marginal cost of software production [54:59] goes to zero what ends up happening is [55:02] more and more people join the market [55:04] right like more people are going to be [55:05] able to [55:06] create and hopefully versel is in the [55:09] middle of every transaction meaning [55:11] we're facilitating both the agents and [55:13] the developers and the new developers [55:15] the citizen developers if you will with [55:17] everything they need in order to ship [55:19] all right well I can't imagine a better [55:21] place to leave it than that GMO thank [55:23] you so much for joining us today that [55:25] was super fun thank you thank you for [55:26] having me G if you want to direct [55:28] listeners to any calls to action where [55:30] can they check you out where can they [55:31] play with VZ if you want to directly get [55:33] in contact with us just find us on X [55:36] vzer or versel reach out to us feedback [55:40] start building head to v.d put in your [55:44] idea press submit and off you go awesome [55:48] thanks so much