Vercel: Human in the loop for the agentic web

Tom Occhino helped create React at Facebook. Now, as Chief Product Officer at Vercel, he's thinking about how to do for backends what React did for frontends. At Lift Off, Vercel announced the integration of World's proof of human, powered by World ID, into Vercel Workflows, making it possible for developers to add verifiable proof of human to any agent workflow in a single line of code. The team sat down with Tom to discuss Workflows, why proof of human is a natural fit, and why verifiable human intent is poised to become standard in production software.

You just launched Workflows at Vercel. What is it, and why does it matter for the agent ecosystem?

Workflows is the easiest way to build long-running, durable TypeScript apps. I'm biased because I helped create React at Facebook, but I think about Workflows as the analog to React for backends.

One of the things React does is enable us to encapsulate sophisticated frontend business logic into reusable components. I can publish a component to npm, import it anywhere, and it just works. That's how UI component libraries work. Workflows does the same thing for backend logic. And backend logic is harder. It's more complicated because you have asynchronous processing, long-running tasks, long sleeps between steps. Imagine a new user signs up and you want to send an email campaign in 30 days. Before Workflows, that was hard. You had to implement retries, durable state restoration, all of it. Workflows makes it dead simple. One line of code.

What's exciting about this architecture is that you can start thinking about sophisticated backend capabilities the same way you used to think about UI components. Package it up, publish it to npm, and anyone can pull those capabilities into their app. 

What makes proof of human such a natural fit for what Vercel is building with Workflows?

Unlike other types of verification where you can synchronously perform an operation on the client or server and report back immediately, proof of human requires the human to actually do something. Scan a QR code, take action. That's an asynchronous task. Building that manually means implementing retries, sleeps, and sophisticated orchestration.

Workflows is perfect for this because we turn the entire process of proving a person is real, or proving an agent is acting on behalf of a verified person, into a single line of code. One function call. You import a package from npm and invoke it. That's it.

As a developer, that's liberating. You're not thinking about durability, retries, state management, and restoration. You just think: prove it's a human. Done. It's a match made in heaven, and it's one of the reasons we were so excited about this use case. This can become the canonical pattern for how you should author software going forward.

When a developer adds a human-in-the-loop step to a workflow, what does it actually look like in practice?

You import an npm package, and inside one of your workflow steps, you write a single function call: await proof of human. The step suspends until that proof comes back, no matter how long it takes. No setting up timeouts. No retry loops. You just write await, and the function is reentrant. As soon as the proof of human comes back, you continue with your idiomatic TypeScript code. Python support is currently in beta.

As agent-driven development accelerates, do you see verifiable human intent becoming a standard part of production workflows?

I think so. We're going to start with the really high-value workloads that are sensitive to abuse and where you need to control who your audience is and who the actor is. We're already seeing a proliferation of agent traffic across the internet, and certainly across Vercel.

But I think most developers, even those using coding agents, are going to want a way to say: this piece of my experience, I need to verify it's a human. For the rest, like reads or other lower-risk actions, you can reduce friction as much as possible. But for high-value, high-cost, high-trust actions, this is going to have to become ubiquitous for us to feel that the internet is still ours and that we can verify who we're talking to. Vercel Workflows and proof of human is how we get there. Explore Vercel Workflows and the proof‑of‑human integration in the human-in-the-loop documentation.

Get started with Vercel Workflows and World ID

Ready to add human‑in‑the‑loop verification to your agents? Get your free World ID at world.org/world-id. For integration details, visit docs.world.org.

Learn more

To stay informed about World, visit the World website, join the daily conversation on World social media channels, or get additional important information concerning the project by reading the World protocol whitepaper.