Two ideas are dominating design conversations right now, and they can't both be true.
On one side, the argument that the best AI interface is no interface at all — buttons are legacy, menus are friction, and the future is seamless, intent-driven experiences the user never consciously sees. On the other, the EU has just introduced guidelines that say if AI is involved, users need to know about it. And no, the little sparkle icon doesn't count.
What happened
The "death of the button" argument has been building for a while, but it's gaining fresh momentum. The pitch to designers is that our role is shifting from crafting visible UI elements to orchestrating invisible interactions that anticipate user intent. Seamlessness is the goal.
Meanwhile, the EU has published new AI labelling guidelines that push in exactly the opposite direction. These aren't vague principles — they specify when AI labels are legally required, what meaningful disclosure looks like, and why generic indicators (like that ✨ we've been putting everywhere) fall short of genuine transparency. If you're building AI-powered products for European users, this isn't optional.
Two forces, one interface
Here's where it gets interesting for us as designers.
The invisible interface crowd wants AI to disappear into the experience. The regulators want AI to announce itself clearly. You can't fully do both. Something has to give.
But I don't think this is actually a contradiction we need to resolve — it's a tension we need to design for. And honestly, it's a more productive tension than either extreme on its own.
Complete invisibility was always a fantasy. When AI makes decisions that affect what you see, what you buy, or what information you receive, hiding that process isn't elegant — it's just opaque. But plastering a sparkle icon on every AI-touched element isn't great either. It's the design equivalent of a terms-and-conditions checkbox: technically there, practically useless.
The EU guidelines are pushing toward something more nuanced. Contextual disclosure, proportional to what's at stake. A chatbot needs different labelling than an AI-curated feed. A generated image needs different disclosure than an AI-suggested layout tweak.
Disclosure design is the new craft
This is the bit most designers haven't clocked yet: AI labelling is becoming a design discipline in its own right.
Think about it. How do you communicate "AI is involved here" in a way that's genuinely informative, appropriately prominent, and doesn't wreck the user experience? That's not a compliance task you hand to legal — it's an interaction design problem. A hard one.
The guidelines are essentially asking designers to answer three questions for every AI-powered feature:
- What is the AI doing? Not "AI-powered" — specifically, what role is AI playing?
- How much should the user care? An AI-sorted list and an AI-generated medical recommendation require very different levels of disclosure.
- What can the user do about it? Can they override it, adjust it, or opt out?
If those questions sound familiar, they should. They're the same questions we ask when designing any interaction — what's happening, why does it matter, what are the user's options. AI disclosure isn't some alien new discipline. It's existing design thinking applied to a genuinely new problem.
What to do with this
If you're working on anything AI-powered — and increasingly, that's everything — here's where I'd start.
Audit your sparkle icons. If every AI feature gets the same generic indicator, you're not communicating anything meaningful. Different features need different levels of disclosure.
Design disclosure as a spectrum. Low-stakes AI (sorting, suggestions) might need a subtle indicator. High-stakes AI (content generation, decision-making) needs explicit, contextual explanation. Map your features accordingly.
Treat this as a UX opportunity, not a compliance burden. The teams that bolt on AI labels as an afterthought will create clunky, legalistic interfaces. The teams that design disclosure thoughtfully will build more trust — and arguably better products.
Stop chasing total invisibility. Some friction is good. A moment where the user understands "AI helped with this" isn't a failure of seamlessness — it's an act of respect.
So basically: the invisible interface isn't dead, but it just acquired a very specific set of constraints. Designers who treat AI disclosure as a genuine craft — not a checkbox — will be the ones who navigate this well. The sparkle icon era is ending. What replaces it is a much more interesting design problem.