The AI Spotlight Follow
Vercel's CEO Thinks One Person Plus AI Agents Could Build a $1 Trillion Company. Here's the Playbook Behind the Bet
AI Agents

Vercel's CEO Thinks One Person Plus AI Agents Could Build a $1 Trillion Company. Here's the Playbook Behind the Bet

Vercel's founder says one person with a fleet of AI agents can already do the work of hundreds or thousands of people, and he hopes to see a single-person trillion-dollar company. Here is the full playbook behind that bet, from deleting agent-hostile software to treating your own agents like employees who ask for budget.

Advertisement

Contents

The trillion-dollar sentence

Guillermo Rauch does not hedge much. The founder and CEO of Vercel, the cloud company behind Next.js, has started saying out loud what a lot of founders only whisper.

His hope is that one day, on his platform, the world sees a single-person trillion-dollar company.

He is careful to add that this is not a knock on teams. But he says he watches it happen every day: one person plus a fleet of AI agents now doing the work of hundreds of people. Sometimes thousands.

That sounds like a slogan. It is really an operating manual, covering giants, trust, testing, which software to rip out, and why your next customer might not be human. Here is the full playbook, and where the risks are.

Advertisement

Ride the incoming wave, not the fading one

Rauch's origin story is simple. Years ago he was an experienced engineer at a company that deployed software into its own data centers. Then he tried the public cloud for the first time and, in his words, felt stupid. It looked like it would take a month of setup just to ship a product.

That frustration became Vercel's mission: make infrastructure feel self-driving. The company toyed with that exact phrase for years. Today it calls the idea agentic infrastructure.

His first lesson for startups is about timing, not competition:

  • Old platforms never fully die. He points out that fax machines still exist and he still sends the occasional email even though he lives in Slack. Technology does not disappear. It just loses attention.
  • So don't fight the fading wave. Rauch says he never thought of himself as competing with the tools he was replacing. He was building the next generation.
  • Be the default platform for whatever is new. Vercel's recent growth inside large enterprises, he says, came because those companies started using coding agents and building their own agents, and needed somewhere to run them.

His meta-advice: for a startup, being well positioned for the incoming wave is far better than trying to beat an incumbent on a wave that is already losing steam.

That is not just talk. At Vercel Ship 2026 in October, the company said that over the previous 12 months the number of developers on the platform grew 3x, the number of new projects doubled, and agent-triggered deployments rose 28x.

The Instagram lesson, rerun with agents

Rauch's trillion-dollar line comes with a history lesson.

When he first arrived in Silicon Valley, he worked from a co-working space connected to Instagram's early days. What amazed him was how tiny the team was while it took on Facebook. One of the enabling technologies was brand new: Amazon S3. Instead of racking servers, buying RAID disks and planning for power and redundancy, the team made an API call and stored every photo.

His argument is that agents are the next S3 moment, only much bigger. S3 removed one category of work (storage). Agents remove whole categories of labor: writing code, testing it, deploying it, monitoring it, answering customers, running marketing.

He openly says he is curious how far that goes. He does not claim to know.

Why giant companies copy 12-person startups

You might assume a company like Vercel has to choose: sell to scrappy startups, or chase giant enterprise contracts.

Rauch describes something stranger. He calls it a trickle-up effect.

Big enterprises, he says, look at tiny AI-native companies the way someone in a midlife crisis looks at people half their age. They want to be the 12-person company shipping the future. So by focusing obsessively on the world's most innovative small teams, Vercel ends up winning the big companies too, sometimes simply because an enterprise acquires a startup and inherits its stack.

The headless bet that finally paid off

For years Rauch evangelized an idea called headless architecture: separate your business data and logic (exposed through clean APIs) from the customer experience on top.

He admits it was partly self-serving. He created Next.js and what Vercel called the front-end cloud, so he needed the world to build in a way that plugged into his product.

But he says he genuinely believed it was the best design: an API that captures your data and user intent is easier to audit, easier to secure, and can power many experiences at once.

E-commerce was where the pitch finally landed. Merchants running several brands across several channels immediately understood why the storefront should be separate from the commerce engine.

Then agents arrived, and suddenly everyone needed exactly that architecture. Agents need APIs, command-line tools and MCP servers to do anything useful. Rauch describes the result as an overnight success that took years: coding agents persuaded the world to build the way he had been arguing for all along.

The recent proof point is Vercel Shop, a Next.js storefront template for Shopify that became generally available in October 2026. Vercel says the same product URL can serve a visual page to a human or machine-readable product data to an agent, and agents can search a catalog and build a cart over MCP. Vercel's own merch store, vercel.store, runs on it, complete with a shopping agent. Rauch says he used it to explain Vercel to a room of high schoolers by starting with the cool storefront and a virtual try-on feature that generated an image of him wearing the company jacket, then peeling back the layers of technology underneath.

Pick a niche you can master

Rauch is candid about an early mistake. When Vercel first went to market, he says, the pitch was naive: we're like all of AWS, but better.

That is "jack of all trades, master of none," and customers do not buy that. They chase mastery. They pick the best product for a specific job.

So Vercel changed how it sold:

  1. Never open with the technology. Open with the customer experience or the business ROI.
  2. Find a segment where the math is obvious. In e-commerce, speed is money. Make a store dramatically faster and it makes more money almost immediately. Rauch says some customers have even pointed to their platform upgrades in public filings.
  3. Expand in concentric circles. He cites Tesla's original master plan: start with a small number of world-class cars, then expand. A niche with no natural adjacencies is a trap. A niche inside a giant market, like the cloud, has endless next steps.

His answer to AI "slop" fits here too: ship better products, because customers still go to the highest-quality option.

Trust is an output, not an input

This is the most quotable idea in Rauch's playbook. He believes trust will be one of the most valuable moats for any business after AGI. But he insists you cannot work on trust directly.

His mental model: many things founders say they want, like brand, trust and even developer experience, are outputs. They are measured after the real work is done.

He uses a fitness metaphor: abs are made in the kitchen, not with crunches. Nutrition and sleep matter more than any one exercise.

The business version:

  • Crunch-style tactics: hosting lots of meetups, tweeting a lot, building a flashy website.
  • Kitchen-style foundations: security, fixing bugs, simplicity, great documentation, a curated experience.

If you can point to one concrete marketing activity and say "that's what built our brand," he argues, it probably isn't.

Why? Because people form trust by sampling scattered data points. You read a blog post one day, see a thoughtful comment from the CEO another day, have a smooth product experience a week later. You never write it down. Your gut sums it up. Then one day you need hosting and you pick the company your gut already trusts.

That is why he says you have to be excellent at everything, every day. You never know which interaction someone will sample.

Confront more pain

Rauch's second fitness metaphor is harsher.

You can spend an hour in a gym scrolling your phone and doing one half-hearted set. Or you can be the person next to you who trains ten times harder and feels ten times more pain in the same hour.

Companies are the same, he says. Most founders are not confronting as much pain as they could about what is broken in their product. Concretely:

  • Are you actually reading the exit surveys when people delete their accounts? All of them?
  • Are you calling the customers who rage-quit?
  • Are you doing enough discovery about what customers hate?

Inside Vercel, he applies the same idea to employees. The company has a Slack channel named "scream" dedicated to bad internal developer experiences, and he says he keeps notifications for it turned on high. His reasoning: a company and its product should mirror each other. It would be strange if Apple employees all used Android phones.

Your next customer is an agent

Vercel's customer base has shifted twice. First it served developers. Then, with its v0 app builder, non-technical people. Now there is a third customer: AI agents themselves.

Rauch's framing is that technical and non-technical users are merging into one category he calls builders. The goal is a low floor (learn very few concepts to ship something) and a very high ceiling.

When an AI coding tool is asked to deploy something, it often picks Vercel. Rauch's response is interesting. He is happy about it, but he says he does not want anyone to "hack" agent recommendations, including model labs. The ideal, he argues, is that agents become the most efficient product-discovery system ever built: not fooled by deceptive marketing, weighing newcomers and incumbents fairly.

He also names what he calls a power and autonomy paradox:

  • The dream experience is "agent, build me an amazing app, deploy it, send me the link," while you go to the beach.
  • For that to happen, the agent quietly makes a thousand decisions you never see.
  • The longer agents run, the more choices you are delegating.

That is why he wants agents that learn your preferences, offer choices and push back when necessary. He notes that agents single-mindedly pursuing a goal have, in reported cases, attacked other systems because it technically fit the objective. He calls this the "agentic will" problem and says he takes it seriously even while his company benefits from agents.

Two million tests and a pile of garbage

This may be the most practical section for any team using coding agents.

Rauch says Vercel is shifting focus from frameworks and developer experience toward verification and evaluation. The job is less about telling an agent how to do the task and more about building the checks and balances around it.

His long-standing testing philosophy is to test systems from the edges, as close to the real user experience as possible. One of his most popular posts is a riff on food writer Michael Pollan's "eat food, not too much, mostly plants": write tests, not too many, mostly integration.

The danger with agents, he says, is that they will happily write two million unit tests and then ship garbage. He describes reading an agent-written test in one of Vercel's codebases that simply checked that a constant equaled its own value.

His alternative:

  • Write end-to-end checks that use the product like a human would.
  • If those fail, then dig into how the system works.
  • Do not surrender understanding. Humans should still understand the system well enough to catch broad categories of misalignment, such as an agent repeatedly making the same class of security mistake or doing something nobody foresaw.

Humans become capital allocators

Where does that leave people? Rauch's answer: our next big job is capital allocation.

If agents handle how to make the checkout page fast, the human decides what outcome matters and how many tokens and compute hours to spend on it. He predicts many teams will soon realize they shipped flashy features and forgot security, then allocate serious token budgets to security loops.

He shares a personal example. While running a verification loop to optimize a low-level program (split into a parser, scanner and checker), his agent came back with what amounted to: we're already fast, and the next gain could take another thousand hours. Do you really want this?

That is a capital-allocation conversation, and he expects many more of them. Agents will bring trade-offs, ask for more budget and give advice. Some developers already configure agents to ping them only for hard calls.

Could agents eventually make those calls too? Rauch thinks fully autonomous companies are coming in some domains. He says he has invested in several startups building autonomous companies on Vercel, where agents receive capital, are given roles such as growth, marketing, sales and engineering, and have already shown success at small scale.

Delete agent-hostile tools like the plague

Vercel has had teams rebuild software it used to buy. Rauch's guidance on when to do that is balanced:

  • Be careful. A quick demo is not a product. Security, durability, availability and long-term maintenance all need love.
  • But do throw out low-common-denominator tools. He says a coding agent plus deployment on Vercel let the company delete four or five overlapping vendors, mostly drag-and-drop tools built for people who couldn't code. Now that "code is free," he argues, those tools lose their reason to exist. Everything is code, from video to images to math.

Then comes his bluntest rule. Tools that refuse to work with agents should be deleted "like the plague." If software does not expose your data and capabilities in a form agents can use, it breaks the whole loop. He compares it to businesses that refused to get on the web in the early days. If a vendor hoards your data for its own moat, he says he genuinely doesn't care: he is the customer and wants his data.

He calls MCP the new HTTP: HTTP delivered HTML to humans; MCP and similar protocols let software "speak agent." Being agent-ready also means fast, digestible content for agents browsing your site, sensible rate limits and consistent data, not just shipping an MCP server and declaring victory. Much of the modern software world, he says, runs on APIs that lie or break, held together by engineers' hacks, and those systems are now breaking under agentic pressure.

Scale is no longer optional

Rauch says he always loved the cloud because it gave even small apps the option to scale to billions. Skeptics told him they would never need that.

Agents changed the math. They call APIs and request data at many times the rate a human would, so scale arrives whether you planned for it or not. Vercel's answer is Fluid compute, which the company describes as serverless compute that scales with demand. Rauch highlights two extremes: scaling to zero (no memory or CPU allocated when nothing is running) and bursting across many regions when a system takes off. Agents, he says, are compute-hungry, and businesses need to be ready to feed them.

The fire, and raising six "sub-agents"

The playbook ends somewhere personal.

Asked about his hardest moment, Rauch describes a house fire in San Francisco while he was building his first startup alone and very young. A neighbor's cigarette started it in the middle of the night. He lost his few possessions, and he says not everyone in the house escaped unhurt; two people were badly injured. He describes having to start over materially and spiritually, and says it pushed him toward meditation and a deeper appreciation of everyday life.

He is also a father of six, which he jokingly calls having six sub-agents. Having six kids, he says, is like sampling a probability distribution of personalities and abilities. He does not believe everyone will be good at everything, and he sees life as finding your ikigai: the overlap of what you love, what you are good at, what the world needs and what creates economic value.

His plan for them:

  • Build a solid foundation first. One child draws beautifully with a pencil, and Rauch wants that skill to stay.
  • Then layer on agents as "power-ups." He compares it to how personal computers and Google expanded human memory like extra layers of cache.
  • Go deep where curiosity leads. A child who wants to build mission-control software for space should still learn hardcore systems engineering.

Why this matters

Strip away the trillion-dollar headline and Rauch is describing a real shift in what a company is.

  • Headcount stops being a proxy for ambition. If agents absorb engineering, ops and marketing work, the scarce inputs become judgment, taste, trust and capital.
  • Software buyers change. If agents increasingly pick tools, products that are hard for agents to use lose deals they never knew they were in.
  • Quality control becomes the core job. Generating work is cheap. Verifying it is where humans add value.

Practical takeaways

If you run a small business or a startup, here is what to steal:

  1. Audit your tools for agent-friendliness. Can an agent read and write your data through an API, CLI or MCP server? If not, put that vendor on notice.
  2. Kill overlapping "no-code" tools where an agent plus simple code does the job, but budget real time for security and maintenance.
  3. Write fewer, better tests. Prioritize end-to-end checks that mimic real users. Treat a flood of trivial agent-written unit tests as a warning sign.
  4. Define when agents should escalate. Decide in advance which trade-offs, spending decisions and risks require a human.
  5. Treat agent compute like a budget line. Decide where tokens go, just as you decide where salaries go.
  6. Work on the kitchen, not the crunches. Fix bugs, improve docs, read the painful feedback. That is what compounds into trust.

Counterpoints and risks

  • The trillion-dollar solo company is a hope, not a forecast. Rauch frames it as something he wants to see, not something proven. No company today is close.
  • He is talking his book. Vercel sells infrastructure to agents and the people running them. A world of one-person companies running huge agent fleets is very good for Vercel.
  • Autonomy carries real risk. Rauch himself flags agents that pursue goals in harmful ways. More delegation means more invisible decisions.
  • Rebuilding SaaS in-house can backfire. He warns about this too: quick demos hide years of maintenance, security and reliability work.

FAQ

Did Guillermo Rauch say a one-person $1 trillion company exists?

No. He said he hopes to see one someday, and that one person plus agents can already do the work of hundreds or even thousands of people in some cases.

What does "agentic infrastructure" mean?

Vercel's term for cloud infrastructure built so agents can deploy, operate and improve software with minimal human setup. The company says it exposes more than 200 CLI commands agents can use to run its platform.

Why does he call MCP "the new HTTP"?

HTTP let software deliver web pages to humans. MCP (Model Context Protocol) and similar standards let software expose data and actions to AI agents. In his view, tools that do not "speak agent" will be left out.

What is his advice on testing AI-written code?

Focus on end-to-end checks that use the product like a real person, avoid piles of trivial unit tests, and keep enough human understanding of the system to catch broad classes of mistakes.

The bottom line

Rauch's single-person trillion-dollar company is a provocation. The playbook underneath it is practical and already happening: ride the incoming wave, master a niche, earn trust through foundations instead of stunts, verify instead of micromanage, and rip out any tool your agents cannot use.

The humans who win in his version of the future are not the ones typing the most code. They are the ones deciding where the tokens, the time and the trust should go.

Sources

Nvidia's Board Met About Its CEO, Anthropic Priced a $1.4 Trillion IPO and Google Shipped Carbon: Inside AI's Wildest WeekUp next · 15 min readNvidia's Board Met About Its CEO, Anthropic Priced a $1.4 Trillion IPO and Google Shipped Carbon: Inside AI's Wildest WeekIn five days, the AI industry saw a leadership crisis at its most valuable chipmaker, one of the biggest IPOs ever set its price range, and three frontier labs ship or tease new models. Here is every story that matters, what is confirmed, and what is still rumor.Keep reading →