Vercel Is Building the Infrastructure Layer for Human-Verified Agents

Mason Nystrom sits down with Tom Occhino, Chief Product Officer at Vercel, to explore how Vercel is evolving from the front-end cloud into agentic infrastructure, and why their partnership with World ID is giving humans a way to be first-class citizens of the internet again.

The insight: proof of human is not just bot protection. It inverts the entire model. Instead of asking whether something is a bot, developers can now build software that is human-only by default and then decide what access to grant to agents operating on a human's behalf.

Key Topics:

• Vercel as agentic infrastructure: three parts, deploy software with agents, build and run agents, and a self-improving platform that sends pull requests to fix its own errors
• From framework-defined infrastructure to agentic infrastructure: the evolution of Vercel's core product philosophy
• Why proof of human is the perfect poster child for Workflows: a multi-step async process that becomes one line of code
• CAPTCHAs are archaic: why the entire model of bot detection is broken and needs to be inverted
• Human-only by default: the new model where software starts as human-only and then grants access to agents that have a verified human backing them
• Trust scores and Uber-style ratings for agents: how proof of human enables differentiated experiences for humans versus bots
• Getting started: one NPM package, one line of code, and the Workflow SDK makes World ID integration dead simpleConcert tickets, sneaker drops, and high-cost actions: the real-world use cases where human verification is non-negotiable

The views expressed in the podcast are those of the individual personnel quoted and are not the views of Pantera Capital Partners LP or its affiliates ("Pantera"). The podcast is provided for informational purposes only to provide market commentary and for general educational purposes, and should not be relied upon as legal, business, investment, or tax advice. The podcast is not directed at nor intended for use by any investors or prospective investors and may not under any circumstances be relied upon when making a decision to invest. Please see additional important disclosures related to the content discussed in the podcast here.

Transcript

SPEAKER_00 0:00

This is so much bigger than bot protection. It's the opposite, right? It inverts the way you think about this to say, my software is human only, and then I can decide how I want bots that are acting with a human backing them to behave on my platform and what access I want to give them.

SPEAKER_01 0:16

It gives humans a way to be first-class citizens of the internet again. Welcome back. Our next guest is someone I'm really excited to chat with. We have Tom O, who is the chief product officer at Vercell. Tomo, welcome to the podcast. Mason, thank you for having me, man. I think the most helpful thing from the listener perspective would be to give a quick overview of Vercel today and the agentic landscape in terms of how your product is being leveraged.

SPEAKER_00 0:43

Yeah, so we describe Vercel as a gentic infrastructure. It's kind of three parts to that. Um, the first is that we're a great place for agents to deploy software. Uh that goes for you know websites, web apps, and other agents. Uh exceedingly agents are building other agents. Um, number two is uh we're a great place for you to build and run agents. Um and number three is uh Vercel itself, the platform, is agentic. So uh I like to describe it as sort of self-driving, but you know, you used to take your software and put it into production and then sit there until you take new software and put it into production. But our goal is to make it so that whatever you put into production, it actually can self-improve. Um, something happens, there's an increase in errors or a spike in something, and we can automatically run an investigation, tell you what's going on, and even send pull requests to fix it for you. So those are kind of the three parts of uh agentic infrastructure is Vercell. Where we came from, you know, we started out as the front-end cloud and um wanted to make it really, really easy for teams to be able to focus on what they were building and not how it gets deployed and scaled and secured and how they observe it. Um we just kind of made this sort of batteries-included control plane over all of your infrastructure, auto-scaling. Uh that was something we we called uh framework-defined infrastructure. And now we're kind of moving into the agentic infrastructure era that I described.

SPEAKER_01 2:04

And as you enable all of this agentic infrastructure, we increasingly move to a world where agents are doing more tasks on behalf of humans and sometimes on behalf of other agents. And so there's this core question of where does identity play into you know how humans interact with agents? Uh and so it'd be great if you could frame up kind of the the the crux of the critical issue of why this is important and and maybe how that led you to your partnership with world.

SPEAKER_00 2:29

Yeah, it sounds good. Um, if you'll humor me for a few minutes, I want to build up some context first and kind of get there to the identity piece.

SPEAKER_01 2:36

Suspense us.

SPEAKER_00 2:37

Yeah. So um, you know, as the front-end cloud, we started building um more and more apps that were more and more sophisticated and leveraged AI. One of those apps we built was VZero. And in building v0, we needed to build a bunch of new tools. Um, we needed an AI SDK that abstracted away the differences between models because we were using new models are being released all the time, and we wanted to use them for different parts of our app without changing product code. We needed the AI gateway because we needed to centralize how everybody was accessing those models. Um, we needed sandbox for untrusted code execution and a handful of other things. We recently released something called the chat SDK. But the most recent addition is a primitive that we call workflows, uh, versel workflows, and then uh the open source workflow SDK. And this will connect to identity in a minute. Um, the the the key claim to fame here for workflows is we're kind of doing for backends, so sophisticated backend uh any type of work that you need to do on the server. We're doing the same thing for backends that React did on the front end. Uh I I use this analogy a lot. I helped create React, so I'm a little bit biased, but um, React made it so you could take really sophisticated UI front-end business logic, and you could encapsulate it into a single component. And then that component could be reused across your app, it could be used across many apps, um, but that component boundary made it really composable and really modular and reusable. What workflows is doing is the same thing for really sophisticated back-end logic, including logic that is asynchronous or requires a long-running job, or critically requires a human in the loop. And so now we've sort of arrived at uh you know world ID and this proof of human. In order to build a system that leverages uh the back end associated with proving that someone is human, think about everything that has to happen here. Um, the request comes in, we need to notify the human, they need to do something with their phone, pull it out, it needs to go talk to the sort of secure you know, on device, you know, then it needs to call back to the server, provide some proof. Like this is a whole asynchronous workflow. And if you're manually building a service that does this, you have to implement things like retries, you have to implement observability for each of these steps. You have to handle like re-entrance and and uh durable state and all these other things. And so if we wanted to introduce something like proof of human um to agents without something like workflow, there's just a lot of error-prone work to do. Um but the workflow SDK makes this dead simple. It's literally one line of code. You just await and then call the function. So we really think that the combination of all of these things together are the sweet spot for developing and running agents. And like Vercel is the center of gravity of a lot of this that's happening. So the partnership with World ID was so, so exciting for us because this is the perfect poster child use case for workflows with this asynchronous, sophisticated back-end tasks. Um and uh, you know, I'm I'm very excited to roll out more of this type of I want to create an entire ecosystem really of like these back-end um uh concerns that now can be really high quality and reused across apps, just like proof of human.

SPEAKER_01 5:55

Right. And help us imagine the future for a bit. So we're moving towards a world where you're gonna have more agent actions, but like you said, there may be certain actions where you want a human in a loop or an agent that you know has the backing of a human to kind of like approve that task. And so as you think towards this world where you know you want a human in a loop, or you at least want a verified human, what are like the types of of use cases that are you're excited about, or or how do you think that evolves uh, you know, in the in the next couple of years?

SPEAKER_00 6:26

So when we first started talking to the team, the the the obvious one was I hate captchas. We did a whole thing at our conference last year about how. I'm so bad at them nowadays. I feel like they get me. It's like, come on, like I, you know, computers can't do this for sure, but I can barely even do this. This is the stupidest way to verify that I'm a human. It just feels archaic, right? So that was the obvious use case. But I think as we've been seeing more and more sophisticated approaches to this, like everybody's trying to build some version of bot detection, bot scoring, bot ID, all these things. I think the easier route is to say, look, let me just default to I only allow humans. And so there's a whole host of things where I really only want to allow humans. And we saw the concert ticket purchase, we saw sneaker drops, we saw all of these, like the use cases around I'm building software where I need to verify that I'm talking to a human on the other end. And like, as we started talking to the team, I'm like, oh my gosh, this is so much bigger than bot protection. It's the opposite, right? It inverts the way you think about this to say, my software is human only, and then I can decide how I want bots that are acting with a human backing them to behave on my platform and what access I want to give them.

SPEAKER_01 7:34

It it gives humans a way to be first class citizens of the internet again.

SPEAKER_00 7:37

That's right. Yeah, that's a that's a really good way of framing it. And what's really interesting is like over time, we're gonna want to actually differentiate the experiences that we offer to humans versus bots, and we're gonna want to provide trust scores based on you know what some human that's responsible for a hundred agents is doing with those other bots, right? And so um I think this proof of human gives us this opportunity to say, like, yeah, okay, I have five bots doing things and they're all doing good things, these are all good actors, but there's this one thing over here, it's like, should I trust this person? So I was trying to just I was like, what metaphors do you guys use? And like the Uber star rating thing kind of comes up for certain actions that are really high cost actions that you want to really protect. Having proof of humans is going to be just a paramount uh capability.

SPEAKER_01 8:21

Yeah. And if you're a developer and you want to leverage this, you know, proof of personhood, how can you do that today?

SPEAKER_00 8:27

Yeah, I mean, I think the easiest way to do it is there's an NPM package that you can just import into a workflow and it's one line of code. Um, so we'll share more about that on the workflow's website and on the world ID uh developer docs and things. I think that's the easiest way to get started, and there's a bunch of other ways to leverage it as well, but um certainly unbiased workflow SDK is is the best way.

SPEAKER_01 8:47

Amazing. Well, Tomo, thank you for joining us and uh sharing more about this partnership.

SPEAKER_00 8:51

Absolutely. Mason, thank you for having me.