Agent Experience in Wearables: Designing for the Screenless User

What happened to wearables in the agent era

Wearables are electronic systems people carry on their bodies that perform a technological function: smart watches, fitness bands, smart glasses, earbuds, medical monitors, AR and VR headsets. That definition has not changed. What changed is what sits inside the device.

First-generation wearables were sensor and screen: collect data, show it, notify. The second generation, agentic wearables, collect the data, interpret it, decide on the user's behalf and transact with the outside world when needed. Smart glasses ask an AI what the camera is seeing, watches carry an assistant on-device, earbuds answer the question directly. What these devices share is that the user reaches an outcome without opening an app or looking at a screen, simply by speaking or looking.

This shift has two sides and the rest of this article follows both. For whoever designs the device, the interface is no longer the screen but the agent's behaviour. For whoever wants their brand to appear on that device, the interface is no longer the website but the data the agent reads.

Simplicity gives way to intent

The foundational principle of wearable experience was simplicity: no complex menus on a small screen, one-tap actions, generous tap targets. That principle still holds but is no longer sufficient, because the best wearable interface has no menu at all.

On an agentic device the user does not walk a menu; they state an intent. They say "I am going for a run" and the device starts the workout, opens the music and dims notifications. Nielsen Norman Group calls this outcome-oriented design: the user states the outcome rather than the steps, and the agent completes the steps.

For a designer that means three new questions.

• Did the agent read the intent correctly? Can the device tell "I am going for a run" apart from "I want to buy running shoes"?
• Does the agent show its work? It has to report the steps it took, briefly and plainly. The first two of Microsoft's Human-AI Interaction guidelines say exactly this: make clear what the system can do and how well it can do it.
• Can the user undo it? A one-tap action needs a one-tap undo. When the agent misreads, the correction must not sit deep inside a menu.

The simplicity principle updates like this: the user says one sentence, the agent produces one outcome, the device shows that outcome at a glance and makes it reversible with one tap. The menu was removed; transparency and control were added.

Ergonomics and aesthetics, plus trust

Nobody wants to carry a bulky, crude or uncomfortable device. Weight, form, material and fit with the user's identity remain the physical foundation of wearable design.

The agent era adds a third leg to that balance: trust. An earbud that always listens, glasses that always see, a watch that always measures. The user carries this device not only because it is comfortable but because they trust it to decide correctly on their behalf and to protect their data. Trust is the digital counterpart of ergonomics; without it the device stays in a drawer.

• Visible state. When is the device listening, recording, or talking to an outside service? The recording light on smart glasses is the simplest example, for the wearer and for the people around them.
• On-device processing. Which data stays on the device and which goes to the cloud? The user should know and be able to choose.
• Supervised autonomy. The agent handles the routine alone and asks for approval on critical decisions such as money or health. Not full automation, supervised autonomy.

There is an aesthetic shift too: the greatest design achievement of agentic glasses is not looking like technology. The less the screen shows, the better the agent is working.

The agent learns habits, the designer sets limits

In the old approach the designer ran user research, analysed routines and built the interface around them. That work is now largely done by the agent itself, in real time. The device learns when the user runs, when they sleep, which notification they open at once and which they never open, and shapes the interface accordingly. The industry calls this adaptive UI.

The designer's job does not disappear; it moves. They are no longer the person analysing behaviour but the person setting the limits of what the agent may learn from behaviour and what it may do with what it learned.

• Which patterns may the agent learn and which must it never learn? Sleep pattern yes; who the user met, no.
• How does it show what it learned? "You usually run at this hour, shall I start the workout" is a suggestion; starting it silently is crossing a line.
• Can the user reset the learning? Habits change, and an agent stuck on the old one ruins the experience.
• How is mis-learning corrected? Sleeping in once on holiday must not permanently move the agent's sleep target.

Good design makes the user feel natural without thinking about it. On an agentic wearable that naturalness comes not from the agent learning but from it knowing when to stop learning.

From notification management to decision management

The critical topic of wearable experience used to be notification frequency and content: too much disturbs, too little weakens engagement; build a priority system, offer haptics, light and sound options, silence during meetings.

In the agent era a notification is a decision the agent takes. The device does not say "a message arrived"; it decides "this message matters, show it now" or "this can wait". Notification design becomes decision design, and what the designer builds is not a settings menu but the agent's decision policy.

What does it surface? A health alert always; a social media notification almost never; the wide space between the two is learned and corrected by the user.

What does it resolve itself? Action instead of notification. Rather than "your meeting moved by 10 minutes", the agent updates the calendar, refreshes the route and simply says "you gained 10 minutes".

What does it never do alone? Sending money, sharing health data, granting third-party access. Human approval for these is not a design preference but a safety requirement. Two principles of the HAX toolkit apply exactly here: control and traceability.

The goal has not changed: let the device integrate into the user's life without disturbing them. What changed is that not disturbing now means taking a decision rather than reducing alerts.

Different environments: the agent reads context too

A screen readable in sunlight, water resistance, long battery life. The physical durability criteria stand as they were, and on agentic devices battery life matters even more, because the agent is always working.

What is new: an agentic wearable uses its environment not only to withstand it but to read it. Glasses see that the user is in a shop, earbuds hear that the space is noisy, the watch feels that the user is running. That context shapes the agent's decisions.

• Context changes the interface. While running the screen shows a single large number and audio feedback leads; at a desk the detailed chart opens.
• Context must not change permissions. A user standing in a shop does not mean the agent may buy there automatically. Context produces suggestions, not authority.
• Context reading must be visible. If the glasses recognise a window display and show a price, the user must know the glasses are looking. Silent context reading erodes trust.
• There must be an offline plan. If most of the agent runs in the cloud, what does the device do in a tunnel or on a plane? Keeping core functions on-device is a resilience question, not only a privacy one.

The principle of outdoor testing updates as well: test the device not only in the rain but in situations where the agent misreads context. If the agent silences notifications because it thinks a passenger is driving, that is a design fault.

Integration with the real world: how does your brand look through glasses?

A user stands in front of a shop wearing glasses and asks "is this open now, do they have this product, what does it cost". What do the glasses do? They do not show the shop's website. The agent reads the site's structured data, the business profile and the API if there is one, and answers in a single sentence. Your site's design, colours and hero image play no part in that experience. What plays a part is whether your data is machine readable.

That is the real meaning of wearable experience for brands: how do you look through the agent's eyes?

Your answer is in structured data. Current hours in LocalBusiness, correct price and stock in Product and Offer, the questions an agent will ask in FAQPage. Schema must not contradict the visible HTML; most agents read both.

Your answer is in freshness. If the agent quotes a price that went stale on your site, the user walks in and meets a different one, and the lost trust lands on you. Feed and site consistency is not advice in the wearable era, it is a requirement.

Your answer is in action. Without an API or an MCP tool definition the agent can call for "book a table", "place an order" or "take an appointment", the glasses only inform; the user goes to the competitor who can transact.

Your answer is in measurement. An agent query coming from glasses never appears in analytics. Without separating agent user agents in your server logs and tracking your visibility inside AI answers, you will not even know the wearable channel exists. You can start that measurement with an agent-readiness scan.

The medical device example widens on the same logic: a device that alerts on a change in heart rhythm can now find a nearby clinic and book an appointment. Whether that clinic appears in the scenario depends not on how beautiful its website is but on whether an agent can read it and act.

Voice, agents and the screenless interface

Voice is now the default interface of the agentic wearable. The user talks to the assistant and the screen exists only to confirm the outcome. For a designer that is a new space beyond the visual interface, and it has its own rules.

Conversation, not a one-shot command. The old voice command was "set an alarm for seven". Agentic voice becomes "I have an early meeting tomorrow, set things up for it". The agent must hold context, ask a single question when uncertain, and not ask too many. In a voice interface every extra question costs more than an extra click on screen.

Confirm the action out loud. On a transaction involving money or critical data the agent states what it is about to do and waits. Skipping that step buys speed and spends trust.

A fallback for places without voice. Voice fails in a meeting, a library or a noisy street. Touch, gaze and a minimal screen should be designed as the fallback, not as the primary interface.

Accessibility moves to the centre. Voice and agents are the strongest side of wearables for users with physical impairments. Designing for that audience means designing better for everyone.

One last note for brands: when a user talks to their glasses, the agent is in a sense talking to your site too. Your headings, definition sentences and structured data are the raw material of the answer the agent reads aloud. "Discover our products" says nothing in a spoken answer; "open weekdays 9 to 19 in Kartal, Istanbul, same-day delivery" is an answer. We covered the button-scale version of the same logic in our CTA article.

User experience on wearables is therefore defined on two sides at once. Whoever designs the device is designing agent behaviour rather than a screen. Whoever wants their brand on that device is designing the data the agent reads rather than a website. The shared name for both is Agent Experience.