The 10 UI, UX and Product Shifts That Will Shape 2027

The attention economy optimised for time spent. The intention economy optimises for tasks finished. Ten shifts, three pillars, and what to actually do about each.

Trend lists age badly because they confuse what is photogenic with what is structural. So here is the structural read: the attention economy optimised for time spent in the product, and it is giving way to an intention economy that optimises for tasks finished. Every shift below is a consequence of that one inversion. We have grouped the ten into three pillars - interfaces that act, interfaces you can feel, products you can defend - and for each pillar we say what to do and what to ignore.

Pillar one: interfaces that act

1. Agentic UX. The interface stops being a place you navigate and becomes a counterpart you delegate to: state the outcome, review the result. The design problem inverts from findability to trust - showing what the agent did, what it will do next, and where the undo lives.
2. Generative UI. Screens assembled at runtime for this user, this context, this task, instead of one layout for everyone. The craft moves from drawing screens to designing the system of rules and components the screens are generated from.
3. Anticipatory design. The product resolves the problem before the request: the invoice chased before it is late, the anomaly flagged before the complaint. Anticipation done well feels like service; done greedily it feels like surveillance, and the line is consent.

What to do: pick one workflow your users repeat weekly and make an agent finish it end to end, with a visible log and a hard undo. What to ignore: bolting a chat box onto an unchanged product and calling it agentic. A text field is not an agent; finished work is. This is the exact boundary we work on in AI integration engagements.

Pillar two: interfaces you can feel

1. The return of depth. Glass, light, refraction, physicality - flat design solved 2014's rendering constraints, and those constraints are gone. Interfaces are reclaiming the visual cues that tell fingers and eyes what is touchable.
2. The human signature. As AI floods the web with competent sameness, deliberate roughness - visible grids, real photography, opinionated typography - becomes proof a person was here. Imperfection is turning into a trust signal.
3. Calm technology. Attention treated as a budget the product spends, not a resource it mines. Fewer notifications that interrupt, more states that wait to be looked at. The products people keep for years are the ones that respect the meantime.
4. Multimodal by default. Voice, camera and gesture stop being accessibility extras and become the primary path for a growing share of tasks - you photograph the meal, you do not type it. Keyboards remain; their monopoly does not.

What to do: rebuild one high-traffic flow around the sensor that fits it - camera for capture, voice for hands-busy contexts - and measure completion, not engagement. What to ignore: cosmetic texture slapped on an unchanged information architecture. Depth that signifies nothing is decoration with a GPU bill.

Pillar three: products you can defend

1. Accessibility as law, not virtue. The European Accessibility Act moved WCAG from best practice to legal exposure for consumer-facing products, and enforcement is starting to produce case law. The economics were always in favour; now the calendar is too.
2. Local-first as a feature you can say out loud. Processing on the device - journals that never leave the phone, models that run offline - stops being an architecture choice and becomes a selling line, because 'your data never leaves' is the one privacy claim a user can verify.
3. Provenance UI. Products that show what was generated, what was measured and when it was last true. As synthetic content becomes the default, the interface that discloses its sources earns the trust the others borrow.

What to do: audit against WCAG 2.2 AA now, before a complaint schedules it for you - the business case is already written - and put a visible date on everything your product asserts. What to ignore: trust theatre - badges and seals that disclose nothing. Provenance is a data model, not a sticker.

• 10 shifts, three pillars, one inversion behind them all
• 2 questions per trend: does it finish tasks, can you defend it
• 0 of these require a redesign to start; all require a decision

The products that will feel current in 2027 will not be the ones that added the most AI. They will be the ones where someone decided what the AI was for.

More on this topic: Web & Product.

Want this built for your business? See what we do.