Why products built with AI start to feel the same

Chris Halaska

Observation

More and more of the products I'm asked to look at were built mostly with AI tools. They've got something in common, and it took me a while to put a finger on it.

Every screen is reasonable. The navigation is the navigation you'd expect. The settings page looks like a settings page. Buttons, colours, type and copy are all fine.

Put the screens together and the product feels heavy. There are more steps than there need to be. The main thing you came to do is no easier to reach than the tenth thing. And nothing about it tells you whose product it is.

I think it comes down to how the tools make choices. Asked for a screen, they reach for the most common pattern for that screen. That's a sensible default. But a product is a set of decisions about what matters most, and those only show up when you look at the whole thing and start taking parts away.

That reducing is most of what I do. After twenty years of this, including six at Google, the hard part has always been deciding what the main path is and getting everything else out of its way. Drawing the screen was the easy bit.

It matters even more where I work. Crypto had no good patterns when I started designing for it, and AI products are in the same place now. Conversation, canvas, agents doing work on your behalf: the standard answer is often somebody's first attempt. Best practice is where I start. It's rarely where the good version ends up.

I'm not against the tools. We use them every day, and they've made exploring ideas far cheaper. What they need is better instructions than "make a settings page". They need to know what this product is for, what it leaves out, and why.

That's what we're working on now. More on it soon.