How do we introduce agentic interfaces without abandoning the interaction patterns people already know?

It’s not just the brand of milk, it’s the preference.

That sounds like it was lifted straight out of a textbook on consumer behavior. We could spend hours digging into the cognitive science of why people prefer what they prefer, but the truth is usually simpler and more environmental than that.

Take coffee. I remember a university professor sharing a case study about how, decades ago, companies had to systematically teach Japanese consumers to drink coffee. They had to cultivate a taste from absolute zero. Growing up in Colombia, I didn’t need a case study. I grew up with the physical smell of coffee permeating the morning air. You eventually acquire a taste for something that, on paper, isn’t actually that tasty. It’s bitter. It’s learned.

We can extend that exact thought to product design.

In 2005, the internet forced us to learn digital menus and navbars. It felt clunky then, but over two decades that structural scaffolding became a universal taste. We acquired it. It became our spatial memory.

Right now we’re being told to acquire a new taste: artificial intelligence. But look at most websites trying to integrate AI today, and interacting with them is a real pain. Designers are frantically stripping away the scrolling, the visual cards, the navigation, and replacing all of it with a stark, empty chatbox.

If I did that on my own portfolio, it would be a nightmare. It demands too much cognitive load from a human who just wants to see what I build.

So when I launched a new portfolio domain eight days ago, I deliberately kept both. You still get the navbar, the cards, the standard scrolling layout, but an AI agent is tightly integrated right alongside them. I kept the scaffolding because humans rely on what they’ve already learned to love. Balance, harmony, logos, none of this is innate. It was all learned.

I’ve watched Contra do the same thing. They added an agent that helps you draft your profile, build out services, even maintain your site for you. But they didn’t touch the rest of the site. The navbar’s still there, the post feed is still there, the whole scaffolding is untouched. They didn’t drop it for the agent. They put the agent next to it.

Right now, the way we connect with our projects is fracturing because we’re forcing users to drink a bitter interface with no training wheels.

Agentic Websites: The Mechanics of the Funnel

A product can’t live purely on the builder’s desire. Any milk brand has its name because of a thousand different environmental factors, and it takes real, consistent effort to deliver content that appeals to the exact people you want to reach. If I want to reach UX/UI designers, I have to talk about the real, messy problems they face.

But there’s a deeper mechanical issue underneath that: websites don’t exist alone in a vacuum. They only exist because marketing funnels drive users there. Once a user arrives, it’s the designer’s job to make sure they interact with the things we want them to interact with, smoothly, without friction.

I’ve been tracking this closely. Over the eight days since launching my new site, when I check the telemetry, the behavioral pattern is plain. Some visitors arrive, show deep curiosity, and stay. Others click, hit a micro-moment of friction, and leave immediately.

That immediate bounce is the breaking point of the modern funnel. When you replace an intuitive, visual layout with a blank text input, you break the user’s momentum. If someone has to guess what to type just to figure out what a company or a product does, the funnel fails. The taste is too bitter, and they walk away.

This is the core challenge of agentic websites: how do you keep the user engaged without demanding they acquire a taste on the spot?

The old way left users entirely on their own, clicking through static pages. The immediate next step is an AI that optimizes for and actively frontruns the user’s wants, predicting the interface changes they need before they realize they need them.

The future of product design isn’t about choosing between human familiarity and machine intelligence. It’s about building hybrid architectures that respect the tastes we’ve already acquired, while figuring out what we’re going to learn to love next.

That question is bigger than interface design though. If the taste itself is learned, what happens once the thing doing the teaching is the website itself, not a designer, not a marketer, but the site watching its own visitors and deciding what to say next?