What happens to your product when the person who wants it never opens a browser?
In Gabriel Benmergui's conversation with Kent C. Dodds, they got to one of Kent's boldest predictions. He thinks most of the traffic your product gets in the next few years won't come from people at all:
"It's going to be 99.99% of your traffic in the next three years will be just agents. My people talk to your people… my humans don't talk to your humans anymore. It's just agents."
That number is his bet, not a measurement. But the reasoning behind it is concrete, and it matters to any founder deciding what to build next and who to hire to build it.
The pizza test
Kent made the point with an example. You tell your agent you want pepperoni pizza at 7:00. You'd prefer a specific chain, but the agent finds it hard to order from them, so you shrug and say another one is fine. The order shows up at your door, and from then on the agent just uses the place that worked.
The first chain didn't lose on price or on taste. It lost because the agent couldn't use it. As Kent put it, "the user won't even find out about you anymore because they aren't going to go to their browser. They're going to just go to their agent."
He compares this to the move to mobile: "the companies that don't adapt to that will disappear." The companies that treated mobile as an afterthought didn't get a second chance once their customers' habits had changed.
"Agents can just use the browser" is a weak defense
The obvious reply is that agents can already drive a browser, so a good website is enough. Kent doesn't buy it. "Browser use is expensive," he said. Driving a page built for human eyes burns tokens, so agents will learn to prefer whatever is cheaper to call: "it's more optimal for the agent to use an API or an MCP server."
Here is a simple way to think about it. Your website is your storefront window, and your API is the loading dock. For a decade we put all our effort into the window. If Kent is right, a growing share of your buyers will only ever use the dock.
The comparison only goes so far. People still walk past windows, and Kent is clear that human connection still matters. His point is narrower. For errands, nobody needs it: "it's not like I need to go buy a bar of soap. I need human connection to do that… I'm going to have you buy me a bar of soap and I'm going to go hang out with my family."
Even the pessimistic case is worth building for
Gabriel pushed on this during the interview. Even if you think agent adoption will be slow, say only 5% of your users show up through an agent, it is still a distribution channel you probably need. If your customers are developers, it's already happening.
Kent's answer was that it won't stay at 5%. You don't need to agree with his exact number to act on it. A clean, well-documented API and an MCP server, a standard way for agents to discover and call your product's actions, are cheap insurance against a shift that is already showing up in developer tools.
What this means for who you hire
This changes what "frontend" means on your team. The interface is no longer only screens. It's also the set of actions, data, and permissions an agent can use safely and cheaply. The engineers who will do best here:
- Think in capabilities, not pages. They can describe what your product does as clear actions with clear inputs and outputs, so they work for a person in a UI and for an agent in a tool call.
- Care about the agent's experience. Good errors, predictable auth, and docs a model can read are now product work, not cleanup.
- Ship with agents themselves. People who already use coding agents every day know firsthand what makes a tool easy or painful for one.
At Silver.dev we screen for this. Our AI adoption interview looks at how engineers actually work with agents, not whether they can recite a framework's API. You can read how we prepare candidates in the AI adoption kickstart and how the AI adoption interview works (both in Spanish).
The takeaway
You don't have to believe agents will be 99.99% of your traffic to see where this goes. If someone's agent can't use your product, that person may never find out you exist. Building for that is a product decision and a hiring decision, and the teams that make both early won't need to catch up later.
Watch the full conversation with Kent C. Dodds on YouTube. If you're hiring engineers in Latin America who already build this way, talk to us.