It’s time to start designing for agents, not humans

For as long as software has existed, the user has been a person. Someone sitting at a desk, poking at a phone, or calling an API. That assumption was so obvious it was never really a design principle, it was just common sense. Every decision about hierarchy, color, button placement, and error messaging was downstream of a single fact: a human being is going to see this.

With agents, that assumption is starting to fail.

In a matter of weeks, @OpenClaw has been used to spawn something like 1 million agents. The reason it went viral isn’t that it’s a novel idea (autonomous agents have been hyped for years now). It’s that it finally feels real. People are setting agents loose on tasks like managing inboxes, booking flights, ordering flowers, and they’re just... doing them. Autonomously, persistently, without someone clicking through a UI.

OpenClaw is interesting not just as a product but as a signal. It marks the moment when AI agents stopped being a concept people talked about, and started being a thing that actually exists in the world. Running on servers, taking actions, consuming software products the same way humans do.

When an agent like OpenClaw goes to accomplish a task, it doesn't open a browser and look for the sexiest landing page. It looks for an API it can call, a CLI it can run, an MCP it can connect to. It wants structured responses it can parse, error messages that give it enough context to recover, a surface that makes capabilities discoverable without requiring a tutorial. The things that make a product good for an agent are almost orthogonal to the things that make it good for a human.

I recently rebuilt our marketing site for @Basedash, and as part of that I wanted to start tracking more events in Google Analytics. Seems like a straightforward thing to do. Except you can't do it through code alone. You have to open the Google Analytics dashboard, click through pages, fill out forms, and create an event type manually in the UI. There's no API for it. No CLI. No way to do it in a way that an agent can easily handle.

This is not a knock on Google Analytics specifically (though let’s be real, it sucks). The product was designed for a particular user: a non-technical marketer who doesn't know what an API is and wants a visual interface with clear affordances. That was the right call for a long time. The design optimized around the actual user.

But now people are working with Claude Code, Cursor, or OpenClaw, and the natural way to handle this task is to describe what you want and let the agent do it for you. But it can't. Not because the agent isn't capable, but because the product's interface was never designed with that in mind. The product is optimized for a user that is no longer the user.

Good UX isn't going away. It's shifting medium and audience. The principles are still the same: make things easy to discover, give useful feedback, handle failure gracefully, don't make the user do unnecessary work. But the user is different now, which means that those principles in practice are different.

For a human user, good UX might mean a clean visual hierarchy and an intuitive flow. For an agent, good UX means a well-structured API with predictable behavior. It means response formats that are easy to parse. It means error messages that carry enough context for the agent to figure out what to do next. It means capabilities that are exposed somewhere the agent can find them, not buried behind a login flow and a sequence of modals.

These are still design problems. They just look more like DX than traditional UI/UX.

The user has always been the organizing principle of product design. Everything else follows from who you're designing for. Visual design, information architecture, onboarding, error handling, documentation. All of it is downstream of a mental model of the person on the other end.

For the first time, that person is sometimes not a person.

I think the best product designers in the next few years are going to be the ones who internalize this early. Not because human users are going away, but because the products that only optimize for human users are going to feel increasingly like Google Analytics feels to me right now. Like a product designed for someone who isn't really there anymore.

The good news is that the underlying craft is the same. You're still trying to understand your user, reduce friction, and make capabilities accessible. You just have to ask yourself a question that used to have an obvious answer:

Who is the user?