Designing products in the age of AI

AI lets a product do things that used to be impossible — summarise, generate, decide, act. But it also makes products probabilistic: they can be confidently wrong, slow, and opaque. Designing with AI is mostly the work of making that uncertainty legible, controllable and trustworthy. The model is the easy part; the experience around it is the product.

Why does AI raise the bar on design?

A deterministic button either works or it doesn't. An AI feature might be right, nearly right, or wrong in a plausible-sounding way — and the person can't always tell which. That shifts the design job from 'show the result' to 'help the person judge the result', which is a harder and more important problem.

What do well-designed AI features have in common?

• They set expectations — showing what the feature can and can't do before it's used.
• They make uncertainty visible — confidence, sources, and easy ways to verify or correct.
• They keep a human in control — preview, undo, and clear override of anything automated.
• They respect the wait — streaming, progress and partial results so the product never feels frozen.
• They fail gracefully — a wrong answer should be recoverable, not a dead end.

What's the most common mistake?

Bolting a chat box onto a product and calling it AI. The interesting work is rarely a prompt field — it's weaving intelligence into the existing flow so it feels like the product got smarter, not like a second, stranger app got stapled on. The second most common mistake is adding AI where a plain, predictable control would serve the person better.

How do you avoid the 'AI sameness' problem?

As more products ship the same generic AI patterns, the differentiator stops being the model and becomes the design — the taste, the restraint, the moments of genuine usefulness. The teams that win treat AI as a material to be shaped by craft, not a feature to be announced.

How we use it

We design the experience around the model first — trust, control, speed and graceful failure — and only then the surface. It's the same design-led instinct we bring to everything: start from how a person will actually perceive and judge what's in front of them.