For 40 years, your users have spoken the machine's language. They hunt through menus, learn your navigation, and memorise where you hid the export button. That's been the deal since the Macintosh shipped in 1984: the human adapts to the computer.

That deal is breaking.

We're entering the third paradigm of computing. The interface you have to master is giving way to systems that read what the user wants. The current scramble for "prompt engineering" talent is a sign of the shift, not the destination. It's a temporary skill, a bridge while the interface learns to disappear.

For anyone who owns a digital product, that raises a sharper question than "which AI feature do we ship next?" It's this: what is your product when there's no screen left to design?

Five things worth knowing.

1. From Command To Intent

Today you tell the computer how to do a job. Click here, then here, fill this field, confirm. Jakob Nielsen calls this Paradigm 2: command-based, the GUI we've lived in for four decades. The computer is an obedient servant that needs every step spelled out.

Paradigm 3 flips it. The user says what they want. The system works out how. Nielsen calls it intent-based outcome specification, and it reverses who's in control.

That reversal carries a cost. As Nielsen warns:

"Do what I mean, not what I say" is a seductive UI paradigm –– as mentioned, users often order the computer to do the wrong thing.

Intent-based systems are a big leap. They're also not the finish line. A true noncommand system, an idea Nielsen first floated in 1993, works as a side effect of normal action. Think of a car door that unlocks because you pulled the handle with the key in your pocket. You issued no command. The system just read the context.

Intent-based AI still takes the lead in execution. So it produces wrong answers your user now has to audit rather than simply run.

‍

2. Liquid UX: Built For One Person At A Time

Most design still aims at the "average user." One screen, built once, shown to everyone. Generative UI (GenUI) changes the target. Interfaces get generated in real time, shaped to a single person's context. Some call it "liquid UX."

You can see early versions already. Adobe's Contextual Task Bar in Photoshop predicts your next move from what you're doing right now. In the enterprise, SAP's Joule and its Joule Agents push it further.

Picture a supply chain manager chasing a disruption. They used to bounce between five apps. Now they ask Joule, and it builds a temporary control centre on the spot: interactive maps, live supply chain diagrams, scoped to that one crisis. When the crisis ends, so does the interface. Project Neo, Adobe's browser-based tool that lets 2D designers build and edit 3D graphics without any specialist 3D skills, points the same way, with the interface flowing across modes and environments around the user.

3. Designers Move From Creating Interfaces To Defining Systems

When AI generates the interface, the designer's job moves up a level. Instead of creating each interface directly, you define the design system the AI works inside: the parameters, the constraints, the standards it has to follow. Closer to system architecture than hands-on craft.

That needs what people are calling agent-aware artifacts. Design documentation is moving from visual style guides to machine-readable files like design.md. Encode your standards once, upstream, and agents produce on-brand work by default. Your brand colours, spacing, and components come out right the first time, instead of a reviewer catching every deviation by hand.

John Moriarty describes the change with a food analogy:

"Instead of handing customers a pantry full of raw ingredients (components and frameworks), we now provide something closer to 'HelloFresh' meal kits: pre-scaffolded agent and application templates with prepped components and proven recipes that work out of the box."

4. The Agent Is The New User

Here's the part that surprises most teams: agents are becoming users of your enterprise platform themselves. Now you're designing for people and agents working side by side. That brings a control paradox, and we've seen how it can go wrong.

AutoML automated the fun part of data science, picking the algorithm, and left humans with the tedious infrastructure they'd hoped to offload. GenUI can repeat that mistake. The fix is hybrid interfaces: AI-generated content sitting next to the familiar controls people already trust, sliders, filters, buttons. People keep the power to step in.

Getting agents and tools to talk needs a standard, and the Model Context Protocol (MCP) is becoming it. Think of it as a standard port: one way for an agent to plug into Google Calendar, Notion, local files, a database, whatever the task needs. The "USB-C for AI" tag is doing a lot of work, but it's accurate.

5. Craft In A Probabilistic World

Software used to be deterministic. Predictable steps, the same path every time. We're moving into probabilistic territory, where a system takes an unexpected, non-linear route to the goal you set. In the "vibe coding" era, your job moves up a level. You architect the system that makes the product, not the product itself.

When AI handles implementation at scale, craft is the differentiator left standing. Krithika Shankarraman of OpenAI puts it well:

"The companies that are going to distinguish themselves are the ones that show their craft. That they show their true understanding of the product, the true understanding of their customer, and connect the two in meaningful ways."

The buttons we've designed around for decades, physical and digital, are fading into the background. Which leaves one question on the table for anyone who owns a brand: if the interface disappears, what's left of your relationship with the people who use your product?

What now?

Working out where AI fits into your product is the hard part. We've done it before, and we stay involved from the first sketch to what ships. If you're figuring out what the next generation of your user experience should look like, get in touch. Let's talk through what's possible for your product.