About Guillermo Rauch
Guillermo Rauch, CEO of Vercel, has emphasized a shift toward AI coding agents, stating that more than 50% of Vercel deployments are now made by agents, up from 3% six months prior, and that over 70% of logged-out traffic to vercel.com comes from agents. He discussed the concept of "impact maxing" versus "token maxing," arguing that organizations should focus on shipping production-ready applications rather than consuming tokens without results. Rauch also said that writing code by hand is "dead" and that competitive advantage comes from managing agents and building "the harness of agents" for outcomes. He noted that Vercel has partnered with Snowflake to ground coding agents in real business context, aiming to help teams build secure applications in days instead of months.
Rauch has also discussed the importance of simplicity and focus in building a company, describing Vercel's growth arc as finding focus, building repeatability, and scaling. He said that "play long-term games with long-term people" and "conduct yourself with great integrity" are core to his approach. Rauch described using natural language as the new interface for agents and said he uses Vercel's internal agent for weekly "brain dumps." He commented that his memory has "deteriorated so dramatically because of AI use" but that he has leaned into delegation and diversification of tasks as a tradeoff. Rauch introduced a feature called "Vercel Drop" for deploying HTML files, which he said agents began using before the company publicly announced it.
Source: AI-verified profile updated from Guillermo Rauch's recent appearances.
Browse all interviews →
Transcript (90 segments)
J
Jules0:06
Hi, I'm Jules and this is Show Me Your Stack, a series where we're interviewing your favorite techies to understand what their day-to-day tech stack actually looks like. Today we're interviewing Guillermo, who is the founder and CEO of Vercel.
This is Show Me Your Stack.
One thing about G, he loves to hack.
G
Guillermo Rauch0:26
He loves to hack. Benevolently, only good hacks.
I actually love reverse engineering, taking things apart.
J
Jules0:37
Is that to sort of identify the holes in that system as you take it apart?
G
Guillermo Rauch0:41
Something that I focus on a lot is trying to understand what's true about the world.
G
Guillermo Rauch0:45
If someone tells me like this framework or this tool is really successful, popular, etc., I try to understand like is that so? Like can I go out into the world and see how it's being used? I have an attitude of like trust but verify with what I hear.
J
Jules0:59
So are you always first person verifying things or?
G
Guillermo Rauch1:02
I try to, and because I have agents that can do the work for me, I can say send my agents out into the world. And I do that not because I don't trust you, but because I actually don't trust you.
I want to verify technology from an end-user perspective. I want to try it out. Like I'm the first person to ever use this technology, like where I pretend to be a novice.
G
Guillermo Rauch1:28
I click things that are not meant to be clicked. People sometimes like don't look forward to my product reviews.
When I first came to San Francisco, mobile and cloud were like blowing up, much like AI is blowing up today.
I don't know that we had so many billboards dedicated to it. It took a lot of reconfiguring of your priors and your mind of I now have to think more about mobile. And if you succeeded in doing that, this massive market opened up to your company. And so now people have to do that for agents.
So to me, there's a lot of elements of deja vu.
J
Jules2:15
One of the most magical parts of early Vercel days was the ease in which a developer could spin up a UI and create an interface, which is like so ironic that we're now here crafting interfaces.
G
Guillermo Rauch2:26
I think we need to now think about what is the equivalent design philosophy but for building agentic products. The new interface for agents is going to be natural language and conversation. So, WhatsApp, Telegram...
Vercel just announced the chat SDK to help developers build agents that have interfaces in all of the messaging platforms. I think UI is still going to be important that you can dive deep into the things that the agent looked into. Let me read those logs. The models continue to improve. More and more of the world will be the agentic interface.
So I still think a lot about like more and more of our company runs on our internal agent. Right now it's called D0.
G
Guillermo Rauch3:08
V0 and D0.
J
Jules3:10
Unfortunately a little too close.
G
Guillermo Rauch3:12
It's going to get renamed soon.
What are the interfaces that are going to be most important to us in our day-to-day? You were asking about like what are some apps that I use, and I believe that in the fullness of time, and this is what Vercel is enabling, I'm going to be able to go to any application and it's going to be agentic and AI native. For applications that are not fully AI native yet, I can do command shift space brings up Raycast AI. I still use the terminal a lot by the way. I use Claude Code and I use this for automating some things on the operating system side.
J
Jules3:47
How are you actually operating day-to-day?
G
Guillermo Rauch3:49
D0 is a big part of it. I have a bunch of channels where I get periodic reports of company metrics. I also can ask D0 for bespoke questions. I also shared on X recently, every Monday I get the brain dump. The agent does basically like a business analysis. It looks at our metrics. This is what happened to total requests on the platform. This is what happened to total compute on the platform. This is what happens to all tokens on the platform. Agents are really good at spotting trends.
J
Jules4:14
Okay. So, walk me through, this is how you're also reading your email. Yeah. How does that work? You no longer use a Gmail superhuman interface?
G
Guillermo Rauch4:23
I still use it from time to time. One of my predictions that is actually, I'm stealing it from Nikita Bier. He mentioned that the spam problem is going to become more and more insurmountable because of things like OpenClaw. It's very hard to notice if some post on social media is coming from a person or some agent.
J
Jules4:40
So if you were to write me an email right now, how would you do that?
G
Guillermo Rauch4:43
Writing doesn't happen that often. Although I will say that for example I can say here summarize these actions I just took and send them to Valerie on Slack. I have the Slack MCP wired up.
J
Jules4:55
You are not in Slack writing these messages yourself.
G
Guillermo Rauch4:58
Well I am most of the time. I am most of the time.
J
Jules5:00
Every Vercel employee is twice when they receive a message.
G
Guillermo Rauch5:03
It found my EA.
G
Guillermo Rauch5:05
Right. And by the way I think it discloses it. So let's open it.
J
Jules5:09
She's going to be like aren't you in an interview right now?
G
Guillermo Rauch5:12
So perfect. Right. So, here's a summary of what I just knocked out.
J
Jules5:15
So cool.
It's so impressive how in the weeds you are still as a developer. Obviously, you've always been a developer. You love building side projects, but this still occupies a lot of your time.
G
Guillermo Rauch5:31
Well, the funny thing is, is this developing? G from 10 years ago would be like, 'No, man. It's an IDE or using Vim.' Like, I'm a huge Vim user and whatnot, but I haven't really had the need. I use Nano Banana a lot and Grog to take an image of something and very quickly prompt to do something else or to give an idea to the teams. I think creation really replaces a lot of the more bespoke ways that we used to communicate. I really believe that a real link and a real product is worth not a thousand words, it's worth 1 billion words. We've been talking about it and I always say like whenever we catch ourselves talking, we should be v0-ing.
J
Jules6:06
Why? Why are you chatting on Slack?
G
Guillermo Rauch6:08
Yeah.
J
Jules6:08
Get busy building.
G
Guillermo Rauch6:10
Get to v0. I love our documentation, but it's overwhelming. So, let's actually open it. I was telling our designer of like, hey, let's actually make it real. And within two basically v0 shots, I now have a new version of documentation. This is a clear instance of a CEO actually adding value even though it's marginal.
I prefer to do this, which doesn't take me that much time, than give a bunch of unstructured feedback on Slack or meetings or whatever.
J
Jules6:43
Right? When you already have a clear idea of what you want.
G
Guillermo Rauch6:51
So, it's like I asked v0, build me like 20 different progress bars that resemble classic video games.
J
Jules6:58
Gosh, I love this.
G
Guillermo Rauch6:59
And by the way, came up with the ideas, too. This is not an original G idea. This is an original G's agent idea. I invited the agent to augment my idea, which is something that I also recommend. A little bit of like ego death. Don't try to overly complete the idea because you're not that smart.
J
Jules7:14
How he starts the prompt. I am dumb. I am not that smart. Let's see what I did. Please fix my ideas.
G
Guillermo Rauch7:19
What was really, so this is where I was giving the agent a lot of feedback.
J
Jules7:23
How far are you going to go back with the agent in like tweaking these little by little changes?
G
Guillermo Rauch7:28
It depends on how much time and how obsessed I am with the problem.
I created tooling that allows me to count how many keystrokes I entered into a given day.
J
Jules7:43
Is this a reflection of your productivity in a day?
G
Guillermo Rauch7:45
I tend to think so. So, let's look at the data.
J
Jules7:47
The month of February.
G
Guillermo Rauch7:49
It's all fake.
G
Guillermo Rauch7:51
It's all fake. So, I actually have it charted by computer. So, it only has the last few days of data.
J
Jules7:56
Let it be known he was working in Q1. Feb 26. It was my hottest day.
G
Guillermo Rauch8:00
Nice. The Coachella of coding.
G
Guillermo Rauch8:10
I do think that prompting well is an art of sorts because sometimes I found success by saying agent, let's both take a deep breath. Let's reset the approach altogether. It's not unlike managing humans.
A G soundboard.
I'm going to make an example. Now, one of the key things is why is this so slow? Why? Because I know how fast it can be. This is why understanding systems, understanding the raw materials, understanding networking and like the bigger higher level concepts is really good for the prompter because then you're not sending your agent into a dead end and you're not wasting tokens chasing something that's not fundamentally possible, right?
G
Guillermo Rauch9:02
We launched Vercel Cues and I wanted to tell the world about how it's different. I cooked with the agent on a couple things. One, I asked it to visit our documentation and compare it to the other products. I combined the research task. It created the dimensions for me and then I also made it beautiful so I could screenshot it and post it on the X platform, the everything app. Before I post it, I adversarily checked with a bunch of other AI tools. You can create like a consortium of agents to like judge your work and then I also worked with the humans on this one.
J
Jules9:35
And then yeah and then sign off. You're like, I'm optimistically trusting the AI, the agent, to perform tasks on my behalf, to do this research, but I will always, always, always check.
G
Guillermo Rauch9:46
Yeah. The ultimate check that I've always loved to do, I check the final product. The final product has to be great.
J
Jules9:51
This episode is brought to you by v0 by Vercel. v0 by Vercel is like your favorite coding agent, but on the web and no setup required. Whether you're technical or not, take your ideas out of the group chat and into production. It's not just a builder, but an agent builder where you can prompt, build, and ship in minutes. Every prompt becomes real deployable code connected to your stack. It's never been easier to create a new landing page, deploy your next app, or build your first agent. Visit v0.app to go from idea to production in seconds.
G
Guillermo Rauch10:22
One of my predictions for 2026 is that we're going to generate so much software as a collective. So, you have to decide, all right, what are the battles that I'm going to pick? That's why we have a lot of really cool things that just meant to stay internal. They're highly custom to how we work at Vercel. We don't want to pollute the world with noise and like, hey, here's 20 other new things for you guys. There's also an app called Timeless. This is another app we built internally that allows me to whenever I spot a bug, I can press a command and dump a few of the last few seconds of screen recording and document the bug for my teammates.
J
Jules10:56
Oh, interesting.
G
Guillermo Rauch10:57
So, if something happened in the past that I wasn't happy with, you can go back. You can go back in time.
J
Jules11:07
What is an essential part of your non-digital stack?
G
Guillermo Rauch11:10
Due to being on a time crunch, I spend a lot of time on Peloton. Oh, I guess it's digital in some fashion.
J
Jules11:17
Digital thing is still a digital thing.
G
Guillermo Rauch11:19
My non-digital thing is digital.
J
Jules11:20
How do you handle task management?
G
Guillermo Rauch11:22
Don't. I try to memorize the most important things. The reason I do this is if something is so important, how is it falling out of your top of mind? Why do you need to write down a reminder?
J
Jules11:33
I feel like my memory has deteriorated so dramatically because of AI use. Like I rely so much on the tools to be my memory system.
G
Guillermo Rauch11:42
For me, it's not so much memory. I think what's going to deteriorate first is the hard thinking cognitive skills because you don't use those brain paths. Do you feel like any of that has deteriorated for you yet?
J
Jules11:56
I think so, actually. But I've leaned more into the massive diversification of tasks and delegation that I can do. It's been a good tradeoff.
G
Guillermo Rauch12:07
Oh, no.
G
Guillermo Rauch12:11
Thank you so much.
J
Jules12:12
That was super fun.