Here's a question worth sitting with: what if the biggest problem with AI in design isn't how we build it — but that we're building it at all?

Not all of it, obviously. But a lot of it. More than most product teams want to admit.

The disconnect is getting louder

A few threads have converged recently that paint an uncomfortable picture. The first is straightforward: most people genuinely don't want more AI in their lives. Not in the breathless, future-is-here way the tech industry keeps pushing. There's a growing gap between what companies are shipping and what users actually need.

The second is an argument I've been mulling over — that designers should stop building new standalone tools and start embedding useful functionality into the workflows people already have. People have established mental models. New features work best when they slot into those models, not when they demand users learn an entirely new way of operating.

And a third thread, focused on designing mental health apps, makes the sharpest point of all: when you're designing for vulnerable users, following trendy UI patterns can actively cause harm.

Put these together and you get a theme that's hard to ignore. We're so busy adding AI to everything that we've stopped asking whether anyone wanted it in the first place.

Why "add AI" isn't a product strategy

There's an assumption baked into most product roadmaps right now: AI features equal progress. Ship an AI summary. Add an AI assistant. Sprinkle in some generative magic. Tick the box.

But users don't think in terms of AI. They think in terms of problems. "I need to find that email from last week." "I want to resize this image without it looking awful." "I need to understand what this data means." They don't care whether the solution involves machine learning, a clever algorithm, or a well-designed filter dropdown.

When we lead with "let's add AI" instead of "let's solve this problem," we end up building features that impress demo audiences but frustrate real users. And I think a lot of design teams know this — even if they're not saying it out loud in planning meetings.

The integration argument is the one that really lands for me. We don't need another standalone AI tool. We've got dozens. What we need is for existing tools to work better together — for AI to quietly handle the tedious bits without demanding attention, without requiring a new interface, without asking users to change their behaviour.

That's boring. It's unsexy. It doesn't make for a good keynote demo.

But it's what actually helps people.

When trend-chasing causes real harm

The mental health app conversation takes this further than most of us are comfortable with. When designers blindly adopt the latest interaction patterns — the flashiest animations, the most novel layouts — in contexts where users are distressed or in crisis, the cost isn't just a clunky experience. It's genuine harm. Increased cognitive load on someone who's already overwhelmed. Eroded trust in a moment that demands safety.

This isn't only a mental health problem, though. It's a principle that applies everywhere we're designing with AI. Every time we add a feature because it's expected rather than because it's needed, we're prioritising our own excitement over the person on the other end.

Tool spotlight: a framework for questioning AI features

One of the more useful ideas I've seen recently is an evaluation framework originally designed for assessing whether trendy UI patterns belong in mental health apps. But it transfers beautifully to any AI feature decision. Before you commit to building, ask:

  1. What problem does this solve? Not "what capability does this add" — what specific user problem does it address?
  2. Does this reduce or increase cognitive load? AI features that demand new mental models are usually a red flag.
  3. What happens when it fails? Every AI feature has failure modes. How bad are yours?
  4. Would a simpler solution work? If a dropdown, a filter, or a well-organised settings page would do the job, you probably don't need AI.

These aren't revolutionary questions. But I'm consistently surprised by how many teams skip them entirely when "AI" appears in the brief.

So what do we actually do?

Next time someone on your team suggests adding an AI feature, ask one thing before anything else: "What's the user problem we're solving?"

If the answer starts with "competitors are doing it" or "leadership wants AI in the product" — that's your signal to push back. Hard.

The best AI features are the ones users never notice. They just make the product work better. That should be the bar. Not "does it use AI?" but "does it help?"