The Interface Will Be Back

I’ve had a soft spot for user experience since the days when “interactive” meant a CD-ROM, the PC was cream, and it arrived on a trolley that required planning permission. If you wanted video, you compressed it until everyone looked mildly unwell. If you wanted delight, you hid it behind a button and hoped someone found it.

Those constraints taught a simple lesson early. The interface is not decoration. It is the promise we make to ensure the job is as hassle free as possible.

Which is why I’m sceptical when people say large language models have “changed the interface”.

They haven’t.

They’ve removed it, and in doing so they’ve exposed something business leaders tend to miss. Interfaces exist to absorb complexity. When you remove them, the complexity does not disappear. It moves onto the user.

For a while, that felt like freedom. Typing a sentence into a blinking cursor. No menus. No training. No onboarding tour with a cartoon hand pointing at things. Just ask the system what you want and it does something vaguely helpful.

Then the cost shows up.

Because once the novelty wears off, you notice what’s really happening. The system doesn’t understand intent. It recognises patterns. It doesn’t know constraints, downstream effects, or what happens next. So the user compensates. They over-explain. They rewrite. They prompt, tweak, paste, retry. They learn the system’s preferences and work around its blind spots.

At scale, that is not efficiency. It is friction redistributed.

It also feels oddly familiar.

In the late 90s, home computing looked much the same. Nerds like me sat in bedrooms with beige towers humming under the desk, a three-inch-thick book on Java applets open on our knees, trying to build a website. There were no patterns. No shared conventions. No friendly tutorials. You learned by breaking things, mis-typing things, and slowly internalising a private syntax that only worked if you respected it.

The barrier was not intelligence. It was tolerance for friction.

Only people willing to absorb that friction got anything done.

Prompt-driven AI is in that phase now. Powerful, expressive, and quietly exclusionary. If you know how to talk to it, you look productive. If you don’t, you’re punished with vague outputs and endless retries. The interface is technically “natural language”, but the behaviour is anything but forgiving.

That phase never lasts.

Eventually, the interface arrives. Not to limit capability, but to remove the need for everyone to specify everything in exhausting detail. The work stops being about remembering incantations and starts being about outcomes. That is not dumbing down. It is how tools grow up.

Which is why traditional UX is re-emerging, without fanfare.

Not the decorative sort. The practical kind. Clear states. Named actions. Visible consequences. Design that reduces decision-making rather than pretending choice is a virtue. Design that makes the correct action easier than the clever one.

You can already see the shift. Language models are being wrapped in products. Prompts disappear behind buttons. “Summarise this” becomes an action. “Generate a report” becomes a flow with inputs, limits, and review points.

The model becomes an engine, not a conversation partner.

This is the uncommon sense business leaders need to sit with. The advantage is not who gives people the most expressive tools. It is who removes the most unnecessary thinking from everyday work without increasing risk.

The mistake is to ask where AI can be added. The better question is where certainty is required.

Because organisations do not scale on intelligence. They scale on behaviour.