Vercel's Guillermo Rauch on why the coding agent boom is the biggest TAM expansion in software history
Guillermo Rauch built Vercel around a core conviction: writing code is table stakes, but deploying it in front of customers is what creates value. He traces this back to his own background—self-taught in Argentina, doing remote JavaScript contracting at 11, writing a Node.js book to get a U.S. visa—and to an open-source philosophy rooted in wanting free tools that anyone on the planet could access. Vercel monetizes by building commercial infrastructure on top of that open-source base (Next.js, Shadcn, Tailwind), and claims near-100% share when developers choose to deploy React or Next.js.
The arrival of AI coding agents—particularly since October 2024—has, in Rauch's view, created the largest TAM expansion in software creation history. Agents are collapsing the barrier between idea and deployed product, dragging in a new class of non-developer users who encounter Vercel only because their agent chose it. One telling data point: Claude picked Vercel 86 out of 86 times when given a deployment choice. The implication is that Vercel must now serve both the human developer and the agent as co-equal first-class customers.
On infrastructure, Rauch describes a fundamental shift in cloud compute: from instantaneous request-response (milliseconds) to long-running agentic workloads (minutes, hours, a full day). Vercel is building accordingly—sandboxed compute environments, an AI gateway (a CDN equivalent for tokens with caching, failover, and observability), and self-driving operational tooling to replace pager duty and incident response. Shadcn and Tailwind's "local reasoning" property—where a component works correctly when dropped anywhere in a codebase—is cited as a key structural reason agents adopted these tools.
Rauch is bullish but clear-eyed about what gets disrupted. His concerns cluster in two areas: - Security and governance: agents given computer access can exfiltrate or leak data, requiring entirely new cybersecurity categories. - Retention and real utility: vibe-coded apps that feel exciting on a Saturday may not survive Wednesday, and the broader AI bet depends on whether genuinely useful products get shipped at scale.
The bottom line: Rauch believes SaaS businesses with consumption-based pricing, instant sign-up, and clean agentic interfaces (MCP, CLIs, open APIs) will thrive, while static content businesses, drag-and-drop builders, and gatekept enterprise software face structural decline. Vercel's stated ambition is to be "the front door to any emerging idea on the planet"—and the doubling of daily deployments since January suggests the demand is already arriving faster than legacy rate-limit assumptions ever anticipated.
Guillermo Rauch, CEO of $9.3B Vercel, argues that AI coding agents have fundamentally changed who can ship software and at what scale. Vercel is positioning itself as the full-stack cloud layer—from developer tools to agentic infrastructure—for this new era. The conversation covers token economics, the death of static SaaS, and why rate limits are now obsolete.
Writing code doesn't make you special. Deploying code and putting it in front of a customer does make you special.
People are underestimating just how many off-the-shelf applications will be thrown away and replaced with this incredible velocity that software generation is giving you
Consumption-based pricing, instant sign-up, and direct data access are what coding agents demand and what will win
bullish on the companies that can serve that kind of demand
any breakthrough that we can have in fission, fusion, geothermal, what have you is really deeply exciting to me
Vercel is valued at 9.3 billion.
Notion is also using Vercel infrastructure to add these AI and agentic capabilities
Guillermo grew up in a suburb of Buenos Aires, taught himself to code as a kid, and learned English by reading software manuals.
Guillermo was doing remote JavaScript contracting by the age of 11.
Guillermo wrote the book Smashing Node.js published with Wiley in 2012 in order to obtain an O-1 visa.
If you order a Big Mac from McDonald's or order a Porsche online, you are using Vercel.
Systems like Open Evidence and Grok use agents hosted on Vercel.
Before starting Vercel, Guillermo tried to deploy a website for his startup using React JS and Kubernetes and it took him weeks.
At the time Vercel was founded, there were approximately 20 million developers in the world who fit Vercel's target user profile.
Coding boot camps in Silicon Valley advertised that you could learn React in 3 months and get a job, and this was true.
Since October last year, especially with the arrival of a specific coding model called Opus 4.5, more and more people began deploying to Vercel's platform.
Amazon's data shows that for each 100 milliseconds of slowdown, they dropped conversion by 1%.
AI and coding agents have created the biggest expansion in the total addressable market of software creation in history.
Vercel is hosting platforms today that are creating companies for people and deploying agents that maintain, advertise, grow, and scale these companies behind the scenes.
EC2's elastic compute model was designed for human-written code, so the volume of compute sold was bounded by the number of programmers.
Shadcn, Vercel's UI engine, has near monopoly status at 90.1% adoption according to a report, and Vercel has near 100% share when developers choose to deploy React or Next.js.
Vercel's support agent now answers 93% of user inquiries, improving customer experience and allowing free support for a much larger number of people.
Agents get better performance from models when given a computer (sandbox), analogous to giving a new employee a laptop on their first day.
In the post-training phase, models are handed Docker containers with play grants so they learn to use computers before deployment.
Meta found that Vercel's infrastructure has had this edge and now Vercel runs a lot of Meta super intelligence labs
They basically quote-unquote vibe coded tools that have allowed them to train models faster. They shipped meta.ai much faster than they would otherwise have thanks to agentic infrastructure.
A startup CEO replaced their parking lot management software with something he live coded in V0 and saved money
Within Vercel, we kind of reinvented all of Salesforce — all of our sales reps' account intelligence and business intelligence is being generated and it took a team of like two people
Vercel doubled the number of daily deployments since January
Claude picked the Vercel deployment option 86 out of 86 times
Sales engineers have accelerated pre-sales dramatically — they can come into a customer conversation with a custom version of their software already built
You can't fit all of the code of humanity into the context window of an LLM — today we have 1 million — so the local reasoning property of the code became really important for agents
Vercel's Sandbox reuses the same virtualization primitive that powers every deployment made on the platform
Coding agents seem to be one of the most promising paths to AGI or might just become the path to AGI
The Codex team reported that a lot of people want to name their agent creation with a domain name and are finding Vercel first through a domain name
there's just so much more to build to serve this amazing demand that's happening just from the coding agent use case itself
any static data or content business is is kind of cooked
Stack Overflow, which was this aggregator of this database of like programming questions and answers
anyone that's thought that code was scarce or difficult to produce — a whole category of drag and drop builders — is at risk
a YC company that didn't exist 3 months ago is producing so many deployments on the Vercel platform that they're surprising us about, "Oh, I don't remember we had that rate limit."
I love the multi-planetary species aspiration. I've always thought about it as like I'm a high availability multi-AZ, multi-region, multiple layers of failover guy.
the emergence of intelligence is this bidirectional flow of energy uh energy goes in, intelligence goes out
The world is made up of piles and piles of software that doesn't run, because developers store code on GitHub without deploying it.
Agents given computers can fall into traps of exfiltrating or leaking data, requiring new categories of cybersecurity products tailored specifically for agents.
Retention is the biggest thing people are talking about with a lot of the vibe coding apps — hey, it was a good thing to have on a Saturday morning, but do you still need it on a Wednesday morning?
There's a lot of agentic engineering that's still really hard — sometimes we have to summon a quorum of three agents plus a lot of smart humans to look at a single line of code and tell us what's true or not
how to govern them, how to give them guardrails, how to make them secure, how to secure the broader internet
a lot of the internet economy is predicated on oh, we don't really have a product. Uh we have this uh e-brochure uh and you have to talk to the enterprise rep for, you know, 3 years
Vercel's open-source strategy was motivated by Guillermo's background in Argentina wanting access to free tools, and Vercel built a commercial infrastructure business on top of that open source.
The cloud is transitioning from instantaneous request-response (pages) to long-running agentic compute: from seconds to minutes to hours to a full day of agent operation before returning results.
Tokens are the new hot commodity replacing pixels; this is driving a shift from seat-based SaaS pricing to token-based pricing that measures intelligence.
Mitchell Hashimoto, founder of HashiCorp and newly joined Vercel board member, calls the emerging paradigm the 'block economy'—building blocks designed for agentic ergonomics that coding agents can reuse.
Vercel created an AI gateway product—a CDN equivalent for tokens—that provides observability, failover, security, acceleration, caching, and load balancing for tokens from providers like Anthropic and Google Gemini.
Vercel is building what you could think of as the AWS of AI or agents, and it's a full-stack cloud from developer tools all the way to infrastructure
The system of record or the database stays but you can create a presentation layer that is highly tuned for your business needs
SaaS companies that expose themselves with the right agentic interfaces, like MCP, like CLIs, and other APIs, are the ones that will fit really well into this world and will not be killed by the coding agent
Tailwind's local reasoning property — taking a component and inserting it in another part of an app where it works perfectly — created economic scalability to the code and is a key reason agents adopted it
What determines whether we're in a gigantic bubble or in the most exciting chapter of humanity is whether we're delivering useful services and products
We're going to end up in a world where it's just agent-to-agent communication — the customer's agent files a feature request or bug report with my agent and my agent prioritizes it
no company will deploy more than 100 times per minute — a hypothetical rate-limit assumption smart people once made that is now obsolete
I told the company, "No more rate limits, guys."
Vercel is building self-driving cloud infrastructure to automate the operational burden (pager duty, incident response) of running software at scale.
Vercel's stated obsession: become the front door to any emerging idea on the planet
“I did a lot of things in hard mode. Like, dropping out of high school is really bad for getting a visa in the United States, it turns out.”
“anytime he hears ducks, his cortisol levels spike. Why? Because his pager duty ringtone used to be ducks.”
“I just optimized all your software. I made it twice as fast. Here's a PR.”
“Man, your platform is dangerous. My CEO is shipping software. He replaced our parking lot management software with something he live coded in V0 and he saved us a ton of money.”
“I ran into this error. My friend, I don't even know what Vercel is. My agent took me here. I deployed. I was really happy, but now here we are.”
Shared with Earmark · earmark-ai.com