Gen AI UX

Chat is not good UX

Why the proliferation of AI chat interfaces bothers me

Everything is a chat now ☹️

Since AI and LLMs have become household names, they’ve been crammed into every piece of software. It’s understandable: chat is a simple, obvious medium that makes interacting with large amounts of information simpler.

But that’s missing the bigger picture.

Chat demands effort. Users have to think of the right question, type it out, then parse through the answer. Most scenarios don’t warrant that effort. Simple affordances beat typing almost every time. Make use of buttons, dropdowns, and sliders. Minimal reading + clear actions = good UX.

To be clear, there is a place for chat interfaces and this isn’t a hit on generative AI as a whole. It’s to point out the awful UX design choices that plague today’s software and make me cringe whenever I see “Now With AI ✨.”

Users shouldn’t always have to consciously engage with AI; they should just notice things getting easier, quicker, better. AI works best when it disappears into the experience, removing friction rather than creating it.

Generative UIs, like Perplexity Labs, are impressive tech- but only a tiny audience knows how to prompt it to get working results. These UIs appeal to tech-savvy users who already know what they want. To everyone else, it’s disorienting and hostile to UX. Unlike the software interfaces of the last several decades that promote memory efficiency gains the more they’re used, Gen UI confuses users with nondeterministic layouts.

Choice overload is a well known phenomenon. In an generative AI context, chat interfaces provide infinite choices. It’s up to us as engineers and designers to decrease cognitive load on users and make generative AI interfaces that mirror existing (efficient) software usage patterns.

TLDR: design for lazy users; think of how to abstract AI chat interfaces into simple, obvious affordances.

00:00