← Thinking
Essay20254 min read

What’s wrong with AI UX

Five-or-so thoughts on AI interfaces at the start of 2025, from chat as the default to work you can’t audit.

That AI is so hot right now. I just worry maybe people aren’t talking about it enough, you know? Anyway, I wanted to start 2025 by sharing some thoughts on what’s good and bad about AI UX as we head into a new year. A toddler and a four-month-old made that apparently impossible until now, but here they are anyway.

1. AI should not have main character energy

With the exception of the AI hype crowd, nobody opens an AI tool thinking, ‘I can’t wait to have a deep, introspective dialogue with this machine.’ They do it because they have something to get done. AI should be the supporting cast, not the lead: the stagehand, not the star.

It should enhance workflows, not hijack them, adapting to what people need instead of forcing them into a rigid and predefined interaction model. In fact, AI should be as unopinionated as possible about its interaction model. The more it gets out of the way, instead of defaulting to a generic chat interface that tries to be a Swiss Army knife but ends up perfect for almost nothing, the better.

Don’t make the tool the focus. Make it disappear into the flow of the work. The best AI is the one you barely notice at all.

2. Mental models are everything

It’s about persistent information, trackable iteration, and the ability to make specific, targeted adjustments with repeatable results. Stability in variation is crucial for professional workflows: especially in creative fields, but also in finance, research, engineering, pretty much anywhere that matters.

The next generation of AI tools will be defined by how well they fit human mental models for how tasks should be accomplished. And guess what? That almost never means a linear chat thread.

Too much context, information and detail gets lost in a sea of text. Recall is hard. We shouldn’t be making users juggle context in their heads; AI should be handling it for them.

3. Chat ain’t it, chief

We’ve entered the era of intent-based interaction, yet somehow most AI tools still feel like Clippy on steroids.

Jakob Nielsen, legend that he is, is bang on when he says this isn’t just an upgrade to old UI models. It’s a fundamentally new interaction paradigm. But right now, that paradigm is still mostly reduced to a single blinking cursor in a chat box.

Where are the spatial UIs, multimodal tools and interactive canvases that actually fit the way people work? The good news is we’re starting to see experimentation in this space. The bad news? We’re only just scratching the surface.

Chat has its place as a natural-language input pattern, but let’s be real: it’s just one tool in the box.

4. Show your working

AI should expose its chain of logic. I want to see why it spat out that result, how I can nudge it in the right direction, and where I can adjust its decision-making.

Explainability isn’t just about trust; it’s about making AI useful. (See also explainable AI, a key principle in unlocking value, usability and adoption in creative industries and beyond.)

Oh, and let’s not forget the sexiest of topics: auditability. In a world of stolen IP, generative content and fake news, being able to track provenance is going to be table stakes for intellectual property, accountability, and avoiding straight-up propaganda.

5. Context is king

AI that understands what you’re doing, why you’re doing it, and how to make itself actually useful? That’s the dream.

Nobody needs AI that feels like an overeager intern who just skimmed the first page of a report and thinks they’ve nailed it.

Sure, one day AI may be sufficiently advanced that all we need is an adequate description of intent, and the machine will figure out the rest. But today is not that day. Until then, while we still need to trust and effectively steer AI tools, we need a lot more power packed into the user interface and interaction model.

The inevitable conclusion

We’re at the start of what could be a golden age of creativity, productivity and all-round brilliance, but we’re not there yet. And there are so many ways we can still get in our own way.

As my first creative director used to say before a pitch (lovingly):

Don’t fuck it up, kids.

Next essayGenerative AI in design →