Guillermo Rauch argues that an AI-era company should not grow merely by hiring more people. It should grow through people who lead each product like founders and agents that amplify them. Vercel calls this Recursive Founder Mode and aims to center product, hiring, customer support, and data analysis on one internal agent, V. Yet Rauch's central idea is not eliminating people. It is a shift that lets humans concentrate more deeply on taste, empathy, direction-setting, and capital allocation.

1. Declaring That the Internet Will Be Rebuilt

The video opens with Rauch's forceful claim that "coding" has become too narrow a word. In the age of AI agents, not only software development but slides, documents, spreadsheets, and business processes can all be treated as code. Work is moving from manipulating separate tools to designing automated systems and flows that produce desired outcomes.

"We are rebuilding the entire internet."

"Coding is a misnomer. Everything is code. A slide deck is code, a document is code, and an Excel spreadsheet is code."

Born in Argentina, Rauch created Socket.IO and Mongoose and co-created Next.js, the globally used React framework. His company Vercel supports frontends for companies such as OpenAI, Stripe, and Nike and has become infrastructure for building software in the AI era.

Through v0, he emphasizes that people can now describe an app in ordinary language and receive working software. But the change is not simply about who can build apps. It can transform how companies make products, how people work, and how society understands creation and labor.

2. Recursive Founder Mode Scales Founder Mode

Paul Graham's "Founder Mode" essay argues that founders should remain deeply involved with customers and product details as companies grow. Rauch agrees with the founder's drive and tenacity: listening directly to customers, breaking bureaucracy, and owning small details all matter.

There is still a clear limit. One founder-CEO cannot personally oversee multiple products, customer segments, and millions of users. Vercel's answer is Recursive Founder Mode.

"A founder-CEO can only go so far alone. Eventually, you need to build the machine that builds the machine."

Vercel does not treat each product as a departmental project. It finds someone with strong vision and ownership and lets that person operate almost as the product's CEO. Rauch cites AI SDK leader Lars Grammel, who was not only a skilled engineer but had a clear view of a model-agnostic TypeScript developer experience and founder-like momentum.

"He is almost the CEO of his product."

Another example is Fluid Compute. Vercel acquired Tom Leonard's company after he had independently handled marketing, communications, product design, and a Rust runtime. Rather than making such a person merely responsible for a feature, Rauch uses them as founder-like leaders with end-to-end ownership.

The model is:

• The founder designs company-wide direction, values, core metrics, and organization.
• Each product leader owns broad areas including development, messaging, marketing, and customer experience.
• The company provides space for autonomy so leaders can experiment and ship quickly.
• AI strengthens the structure. Narrow roles limited to writing code in one language become less important; people who move the whole product become more important.

"What I love most is seeing someone inside the company own an entire business end to end as if it were theirs."

3. Turning Technology and Taste into Agent Capabilities

Rauch jokingly describes himself as something like a Chief Quality Officer. He directly inspects quality and tries to move that judgment into agents and organizational systems so it does not depend only on personal instinct.

Vercel's skills.sh fits this effort. It organizes the company's Next.js and React application knowledge into a skills repository AI can use. Some asked whether Vercel was open-sourcing all its knowledge, but Rauch believes openness strengthens the platform and customer trust.

"Our value is not in hiding knowledge. It is in the whole Vercel platform and the trust we give customers."

He also cares deeply about communication quality. He likes X partly because the constraint of short writing forces a message to become compressed and precise. Vercel employees should not merely launch technology, but help people understand why they should use it.

"You are not here to throw technology into the void. You need to help with storytelling."

"It is not enough to launch. The product has to land so people understand what it is."

He wants to turn years of QA instinct, customer communication, and the craft of landing a product into internal agent skills. Converting exceptional individual judgment into repeatable organizational capability is the core of Vercel's recursive improvement system.

4. AI-Era Hiring: Actual Work over Résumés

Rauch says the traits required in AI-era talent change every week, but two standards remain. One is proof of work—evidence of what a person has actually built. The other is a future-oriented mindset that anticipates change six or twelve months out and repositions accordingly.

"Ultimately, I look for people who can ship something great."

"What matters is not what you built today, but whether you can show what you will ship next week."

Vercel's interviews reflect this through a work prompt. Candidates create a role-relevant deliverable and show how they explain it, how closely they examine details, and whether they explore unexpected, interesting directions.

The best candidates do more than complete the assignment. They identify friction, missing features, and improvement opportunities encountered while using Vercel. Rauch recalls one candidate who submitted beautiful work and then said, "I wasted an hour here. This is an opportunity Vercel should fix."

"She brought not only the result, but a point of view about what she wanted to improve in the company and platform."

The standard applies beyond engineering. Marketing, sales, and customer teams should use agents and platforms to make deliverables directly. A marketer who files a ticket for design or development and waits is too slow in the AI era.

"You are not here to file Jira tickets. You are here to solve problems."

5. Why an 800-Person Company Treats Every Hire as Quality

Vercel has roughly 800 people, but Rauch remains involved in every hire. He does not meet every candidate personally. Instead, an agent summarizes interview progress, work, history, and public activity and incorporates his hiring standards into the prompt.

He may exchange asynchronous questions or request revisions. This is not mere pickiness; he sees hiring as a relationship lasting years or a decade.

"This is a generational project. I will invest in you, and I expect you to be here for ten years."

Hiring is not isolated as HR work. Human quality is company quality just as product quality is.

"Ultimately, everything is a quality issue. We need the best people."

One Vercel value is Iterative Greatness. It does not mean stopping until perfection. It means shipping quickly, receiving customer feedback, and continually improving without blocking product flow.

"At Vercel, nobody has the right to block something. Flow must continue. Tokens must flow, and products must ship."

The exception is security, which can legitimately stop a release. Otherwise, nobody should halt production merely by delaying a review or decision.

"Do not hold a product until I have time to review it. Send the email. The product ships."

Vercel deliberately designs few bottlenecks. Its CTO and CEO analyze organizational throughput, changelog frequency, and how quickly improvements are communicated to customers. For Rauch, operations are also code and system design.

6. Why Founders Stay at Vercel: Autonomy, Platform, and Distribution

AI makes it easier to build a product in a weekend and raise seed funding. Why should entrepreneurial talent remain at Vercel? Rauch does not believe short-term movement following money and fame is healthy. People who immerse themselves for a long time and accomplish something large matter more than those who continually switch companies.

"A founder is also someone with long-term commitment and long-term vision."

He invests not only in famous people or those with large followings, but in people with steep learning curves who can grow quickly. From the hiring stage, he states that Vercel's work may require at least ten years and tremendous effort.

If an employee genuinely wants to found a company, he supports them. He admits that he too was not suited to a lifetime in hierarchy. But some people can gain autonomy and entrepreneurial fulfillment inside Vercel. Recursive Founder Mode creates the feeling of operating your own company within a company.

Vercel is also platform, distribution, and customer channel that can make a founder's product much larger. Something limited when built alone can grow rapidly inside Vercel's broader portfolio.

Rauch disagrees that AI will destroy all SaaS. Some fixed, generic SaaS will weaken, while companies increasingly build internal tools and agents themselves. Customers may buy fewer separate SaaS products and instead build proprietary tools and knowledge on a larger platform.

"Static, off-the-shelf software may not be as good as what customers build and deploy themselves on Vercel."

As making software becomes easier, deployment, distribution, trust, and foundational infrastructure become more valuable. That is why Vercel can attract product founders internally.

7. Speed and Taste Do Not Conflict

Vercel prizes rapid shipping and obsessively pursues aesthetic polish. Speed and taste can appear opposed: quick releases may be crude, while refinement slows work.

Rauch agrees that they conflict under a naive approach. But when tools and primitives raise the quality floor sufficiently, fast work is much less likely to be terrible.

He cites shadcn, creator of shadcn/ui. The component system raises the baseline for buttons, dropdowns, sidebars, and other modern application elements. Vercel extends such components through its own design system so people build on strong tools.

"At Vercel, if you have capability and agency and access to all these primitives, it is quite difficult to produce something terrible."

Rather than hiring designers without limit, the design-engineering team builds agents that incorporate design review and refined detail. Organizational taste and customer-journey judgment become skills.

Rauch strongly criticizes the supposed opposition between AI and taste.

"I think there is a false dichotomy between AI and taste."

"The truly tasteless act is using the model's first output unchanged."

"AI slop" means clichés such as purple gradients, excessive uppercase text, and contextless green online indicators. The answer is not abandoning AI, but encoding good taste and prohibitions in skills, prompts, and review systems.

8. The Engineer's New Role: From Writing Code to Building a Factory

Rauch sees AI less as a new threat than the latest platform transition in software. JavaScript was once mocked as a toy language and web development regarded as inferior to desktop software. Mobile and cloud met similar resistance.

"People who insisted on desktop missed the web. Those who insisted on the web missed mobile. Those who insisted on data centers missed cloud."

When new technology appears, people should try it and leap to the platform rather than defend identity. AI autocomplete, Copilot, ChatGPT, and agents belong to the same progression.

The new essential engineering skill goes beyond telling a model "fix it" or "do not make mistakes." It is deep understanding and effective use of AI. Ultimately, the important capability is not producing one output, but building a software factory that continually produces outputs.

"The core meta-skill is now building a factory that produces software."

Vercel's internal data connects faster shipping through AI adoption with revenue growth. People move from producing outputs to improving the systems that produce them.

AI SDK tests the same philosophy. Customer issues can be resolved largely automatically, and agents can analyze new model capabilities and incorporate them into the SDK.

"Customer reports and model progress can be automatically formalized and shipped."

The emerging role is the product engineer: someone who thinks about customers, product strategy, P&L, retention, and scalability while implementing. Rauch expects flatter organizations, faster information flow, and broader individual responsibility.

9. How Young Creators and Organizations Should Use AI

In the past, a 17-year-old could demonstrate excellence by building open source and earning GitHub stars. AI now lets almost anyone build some software, making the ability to build a baseline rather than a differentiator.

Rauch believes young creators must explain product value, win customers, and produce business outcomes. Indie maker Marc Lou, whom he invited to an internal Vercel discussion, represents a full-stack AI-enabled builder who thinks through landing pages, calls to action, and the relationship between visitors and revenue.

"Making code and publishing it is no longer enough. You have to go farther."

Looking back at Socket.IO's use in Microsoft 365, Rauch says he would now have told customer stories more actively. The boundaries between open source and business, project and startup, are blurring. Products need names, logos, websites, customer stories, and distribution strategies.

Vercel is not immune to unequal AI adoption. Giving every employee an AI subscription does not make everyone ten times more productive. The top 5% may use prompts, tools, and skills at tremendous speed while others use AI shallowly like ordinary office attendance.

The answer is not giving every person a different chatbot, but putting everyone on a shared foundation. Vercel built V for that purpose.

"For an agent to succeed, it must become like the iPhone. It calls, texts, plays games, and does anything—but there is one place to go."

10. V as the Company's Single Front Door

Vercel operates an internal agent named V. When employees need information, automation, or answers about a customer or the business, V should become the first stop instead of many SaaS products and chatbots.

"When anyone in the company does not know something, wants to automate something, or has a question about the business or customer, I want them to go to V."

V is not one model. Beneath it are multiple agents, applications, capabilities, skills, and teams managing them. Vercel also released Eve, a framework for building such organizational agents.

SaaS will not disappear, but the standard for good SaaS will change. Fast websites, rich features, and stable interfaces once mattered most; now interoperability with agents becomes essential. CRM and support tools must be callable and operable reliably by agents such as V, not only by humans clicking screens.

"A platform's rise now depends on how well it interoperates with the V's of the world."

Even an MCP implementation is useless if calls are excessive or frequently fail. Agent-era infrastructure must withstand explosive request volume, not merely expose an API.

Agents can increase service load 100-fold in a short time; Vercel's own load doubled within months. Legacy deployment environments and slow approvals can become severe bottlenecks.

"Agents provide Ferrari speed. You cannot put them into a two-hour queue when they try to deploy."

11. The AI-Era CEO's Job: Use It Personally and Save Human Time

Rauch says every CEO must personally use AI. Without experiencing what products and agents can actually do, leaders cannot correctly change roadmap or operations. It is like a mobile-era CEO discussing mobile strategy without a smartphone.

"Agents are such a large platform shift that you cannot understand it without using them."

He uses AI as an inexhaustible IQ machine. He assigns asynchronous personal research in a CLI and handles company questions by speaking with V in Slack. Work that once required meetings, decks, screenshots, and messages to many people goes straight to an agent.

For example, he can ask an agent to investigate how easy it is to sign up for Vercel AI Gateway and obtain an API token. The agent performs and analyzes the journey from the user's point of view.

"If I send someone a DM, it probably means my agent could not solve the problem."

This does not devalue people. By assigning repetitive research and information retrieval to AI, Rauch contacts humans only when judgment, context, or relationship is genuinely required. Future collaboration will not be "me or the agent," but me, you, and the agent together.

12. Agents Operating Security, Data, and Cost

Vercel uses agents as an operating system, not merely assistants. In security review, an internal tool called DeepSeek gives a model the role of a highly paranoid security expert. It finds code vulnerabilities and compares models' security-analysis performance and cost.

Rauch finds it fascinating that a model can review code it wrote or analyzed and explain what is wrong.

"The model can write its own code and tell you why that code is wrong. It feels a little like therapy for a psychopath."

Using the strongest and most expensive model for every task is irrational. One may excel at vulnerability discovery while costing far more; another may be slightly weaker but seven times cheaper. Agents should select models based on task importance and cost.

"Intelligence can flow like water from a tap, but it can also be as expensive as liquid gold."

Rauch expects token cost to become a major share of operating expense. Companies must manage which departments and tasks use which models and in what quantity. AI adoption is not only productivity improvement; it eventually means redesigning the P&L.

Vercel's data organization is also moving away from employees filing analysis tickets and data scientists replying within an SLA. Data access and security policies are encoded in V so employees can ask it directly.

When the CTO asked for a powerful tweet idea, V discovered that Vercel had recently reached two billion deployments and suggested it. Rauch saw this as an impressive combination of data agents and reasoning.

13. The Company of the Future Starts with an Agent

If Rauch founded Vercel again today, its foundational asset would not only be a website or cloud infrastructure. More fundamental than incorporation, he would build an agent to operate the business alongside him from the beginning.

He recalls the robot in Bicentennial Man and a robotic assistant that follows the protagonist while knowing everything about the business. Future companies may grow through founders and agents like V together.

He does not over-humanize agents. They remain machines and infrastructure. But for a good employee experience, an agent should be refined to carry company-specific voice and personality. One Vercel intern complained that V was too wordy and boring and worked to give it more Vercel character.

"When you start a company, your agent will start with you."

A goal as broad as handling "every pixel and every token on the internet" may still require many people. But Rauch sees agents produce more than teams expected every day. CEOs should test directly and enter the details instead of judging the possibility abstractly.

14. v0 and Infrastructure for the Agent Era

v0 is Vercel's most visible AI product, but Rauch says it did not change the company's essence. Vercel's soul remains infrastructure for rapidly and reliably deploying software built by people or agents.

v0 did not begin as generic code generation. It began with the question: what if someone could start a Next.js project that already reflected Vercel's quality standard? A user begins with a prompt and receives accumulated development and design knowledge by default.

"v0 helps you start faster with something that carries Vercel's quality standard."

Cloud services also change with AI. The early web needed CDNs to distribute static content globally. As the web became dynamic and personalized, Vercel built a modern CDN. The agent era needs not only page delivery but a CDN for tokens, leading to services such as AI Gateway.

Human interfaces are insufficient. Agents need machine-friendly fundamentals such as CLIs, APIs, and MCP servers more than buttons and dashboards.

"Designers now have to design CLIs and APIs."

Companies serving agents need interfaces such as MCP servers. Web pages and page-speed metrics once defined competitiveness; now reliable APIs and servers for agents do.

15. Do Moats Remain in the AI Era?

If anyone can build quickly, does a long-term moat disappear? Rauch says customers ultimately choose the product with the highest quality across every dimension.

Infrastructure cannot be casually vibe-coded. Global distributed systems, security, performance, operational procedure, and reliability do not emerge from a few generated lines. Platform and infrastructure moats may grow.

"You cannot simply vibe-code infrastructure. You need operational rigor and security best practices."

Barriers may fall for small custom software solving a specific problem, but that is not necessarily bad. More solopreneurs or tiny teams may produce tens of millions of dollars in annual revenue, doing work that once needed a 100-person SaaS company and major funding.

Demand for software will not vanish. The idea that everyone will build and maintain all their own apps remains partly unrealistic. But the boundary between venture-scale businesses and small, highly profitable ones will shift.

16. What Remains Human: Empathy, Taste, Choice, and Connection

No matter how powerful AI becomes, Rauch sees a clear human role: empathy, emotion, and genuine relationship. He rejects AI as a friend and values real connection with coworkers.

"Human empathy, emotion, and connection are ours, and they will remain ours."

Human connection may become scarcer and more valuable in the AI era. Social networks can command a larger premium not merely as content feeds, but as places to connect with real people. Physical meetings and shared time likewise gain value.

Humans must decide what to build, what values and taste to reflect, and where to allocate resources. AI may become increasingly strong at bugs, performance, and security. Rauch adapts an old open-source maxim:

"Given enough tokens, all bugs are shallow. All performance problems and security problems are shallow too."

But people decide where those tokens go, which problems matter, and what company and future to create. He calls this capital allocation.

17. A Creative Renaissance, Not the End of Work

Unlike AI leaders warning of mass job loss, Rauch is highly optimistic. He believes AI can bring people into a much more imaginative and creative world like Ready Player One.

He recalls the CD-ROM encyclopedia Encarta. Annual distribution forced creators to invest in detailed illustration and interaction for pages such as human anatomy. The internet prioritized rapid updates and massive access, making the simpler, text-centered Wikipedia style standard.

As AI collapses production cost, the web can become immersive, beautiful, and playful again. Rauch cites a viral AI-made 3D human-body exploration as evidence of revived creativity and expression.

"The web will become a far more whimsical and delightful place."

"We are rebuilding the entire internet."

He also expects artists and small businesses to return. Individual taste, craft, and unique brands suppressed by industrial content production may regain power. Just as family-made olive oil differs from a mass brand, people may seek distinctive products made with care.

Through partnerships such as Shopify, AI can remove barriers that once made people think, "There is no way I can run that business." One day, anyone with an idea may realize it through a prompt.

18. From Coder to CEO: Managing the Factory of Yourself

Rauch candidly says that becoming a CEO after being an exceptional individual developer was difficult. Early founders code, build infrastructure, design, market, and run payroll, but nobody can perform every task at a world-class level.

AI lets leaders do more, but remaining only in one specialty can make them fall behind. A narrow identity as a CSS or JavaScript expert is insufficient. Leaders increasingly need to become full-stack humans.

As companies build software factories, individuals must manage the factory of themselves.

"You are responsible for your own personal factory."

One of his most important CEO responsibilities is managing his psychological state. Businesses do not rise in straight lines; they improve and retreat. He succeeded through product simplicity, but at one point lost focus by attempting many things simultaneously.

"A startup never goes up in a straight line. Things improve and then move backward."

He hates mistakes, but believes most are reversible two-way doors. Even when a mistake feels fatal, breathe, endure for days or weeks, and correct it.

"Most mistakes are two-way doors. Breathe and endure; they pass."

19. Scalable Management and the Meaning of Customer Support

Rauch does not think management is simply holding many one-on-ones. Great managers scale their influence through leading by example, repeating values, and transmitting culture.

He continually mentions Vercel's Iterative Greatness and says making values spread was how he scaled himself. Jensen Huang's operation of Nvidia with many direct reports is mentioned from this perspective.

Despite being CEO, Rauch regularly joins customer-support rotations. Customers are sometimes surprised to receive support from a CEO, but he considers it natural.

"The CEO does customer support? Of course."

"I do not want to perform symbolically. I want to enter the field and feel the pain."

Support lets him understand customer problems directly and demonstrates customer obsession through action. He even says that if competitor CEOs do not help customers themselves, that gives Vercel an advantage.

20. The Source of Energy: Spend Energy to Expand the Budget

Late in the interview, Rauch discusses his energy while raising five children and awaiting a sixth. He exercises every morning, supports customers, follows reactions on X, and operates the company. He recalls kindergarten teachers asking his parents what they fed him because he seemed like pure energy.

He says this is not only innate temperament. He deliberately invests in exercise, diet, and routine for sustainable energy, especially cardiovascular training.

"To increase your energy budget, you have to spend energy."

Exercise is not merely weight loss. It expands his capacity to become a better father, husband, CEO, and colleague. Loving the work also keeps it from feeling like pure suffering.

He still experiences low moods and difficult periods. Rather than necessarily defining them as a fixed state called burnout, he treats them as changes that can be observed and managed. He even used AI to build an app tracking daily keystrokes as a signal of his condition.

"You cannot perform at your lifetime peak every day. What matters is not falling into despair."

21. Rapid-Fire Answers Reveal Product Philosophy and Humanity

Asked for the most overrated startup advice, Rauch unexpectedly chooses ambition. Too much ambition can make startups ship unready products and flood the world with slop.

"Many startups are so ambitious that they pour terrible output into the world and die."

He is not arguing against ambition, but for staged achievement and preserved quality. Reaching a milestone and then scaling 100-fold may be healthier.

The product he most admires is Google Chrome. Claims that AI has recreated a browser imitate only Chrome's surface, he says. The enormous CI pipeline, security, binary-size optimization, and performance are not easily copied.

In AI, he considers voice control of computers overrated. Voice is useful in some moments but not yet natural and efficient enough to replace typing.

He also recounts his last coding without an agent. When Starlink failed on a flight, he opened Neovim and wrote code in the old way. He jokes that he coded "like a Stone Age caveman," while acknowledging that returning to code itself remains useful sometimes.

22. Obsession with Deployment Speed and Human Writing

Rauch especially loves reducing deployment time. As a child, dragging a file over FTP and immediately changing a live site felt astonishing. Modern deployments taking tens of minutes or hours deeply frustrate him.

Vercel aims to deploy applications of every size in single-digit seconds. Cutting Next.js build time by 50% in some cases or saving one or seven seconds each day delights him.

"What I have chased is the experience where saving a file instantly changes the application on the internet."

Speed matters not only to human developers. Agents need near-real-time deployment to generate code, deploy, observe, revise, and repeat quickly.

Finally, Rauch names emotional relationships with family, love, and his own writing as work he never wants to hand to a machine. AI can research and review a tweet, but should not write it for him.

"Does AI write my tweets? Absolutely not."

He compares human sentences to next tokens. But his next token contains the day's experience, current space, DNA, upbringing, life in Argentina and the United States, family, football, and memories of the World Cup. A single system prompt cannot fully replicate it.

"The best connection you can receive is probably the genuine, authentic, farm-to-table next token that comes from me, not a machine."

23. Conclusion: Extraordinary Time Investment and Greater Humanity

To people watching from places such as Argentina or India and dreaming of a similar life, Rauch says talent is insufficient. He was strong in mathematics and performed well in local competitions, but believes long years organized around learning and work made him.

"My life was carefully constructed from Monday to Monday for learning and working."

"Ten thousand hours or 996 is beginner level. You have to truly put in the time."

The effort described is not simply praise for long working hours. Rauch's message is closer to: manage personal energy and health, stay close to customers, encode high quality and taste in systems, and use AI to become capable of broader responsibility.

The strong AI-era company he imagines is not an automation company that replaces humans. It is a company where people with founder-like ownership build faster with agents, take wider responsibility, and protect the connection and choice only humans can provide.