People will reach institutions through their own agents, which pull data from those institutions and generate interfaces built for one person. Organizations will influence what their customers see but won’t control it, so design shifts from crafting screens to shaping the data, constraints, and suggestions an agent works from.

The power of experience: The Google parable

Consumer preferences shift alongside their experiences, and in digital products, those changes often follow major technological shifts. Consider the early days of Google. Once people experienced fast, accurate, and relevant search (for the time), simple searches built into sites suddenly seemed unacceptably bad in comparison.

Behaviors follow preference. To avoid using those in-site searches, people started going to Google first, not individual sites. Even when they knew they wanted to buy from a particular brand (say, L.L.Bean), they would start by googling “L.L.Bean sweater” to speed up their experience.

These behavior shifts cause market and economic shifts. People who googled “L.L.Bean sweater” might end up buying from L.L.Bean. Or they might notice an advertisement for a competitor and buy the sweater there instead. Google became an information broker that could put its thumb on the scale for an advertising fee. And to survive, L.L.Bean needed to pay.

Experience is extremely powerful. When new technology enables substantially better experiences, that becomes the new standard. When consumers find the old offerings unacceptable, companies must adapt or be left behind. Even government agencies, nonprofits, and enterprise products eventually feel the pressure of new standards, though they adapt much more slowly.

Right now, we’re at the beginning of a new experiential shift that will redefine what it means to design and deliver experiences.

Personal agentic AI is finally ready for consumers

Within the past six months, the landscape of personal agents has changed substantially. Yes, it’s been possible to have your own personal agent: an AI system or service that stores information about your life, preferences, and history, and can take digital actions on your behalf. However, these were only available to the relatively small segment of people (like us) who had the time, interest, and ability to research, experiment, and hack together what they wanted—as well as maintain that system over time.

But now, personal agents for consumers have finally arrived: offerings like Meta’s Muse that simply work, without asking normal people to spend hours learning the difference between a prompt, plugin, and skill. Depending on the product, they may not even have to use a new app. Instinct just lets people text their personal agent, as they would a human.

We’re a long way off from everyone having a personal agent in their pocket. More than 10 million people in the U.S. and Canada downloaded Muse within six weeks of its launch. That’s impressive growth, but it’s only 3.4% of the adult population.

But this is where the power of a positive experience will kick in. The more people use personal agents, the more accustomed they’ll get. The more they’ll show and recommend them to their friends. Until eventually, the idea of manually scheduling appointments or ordering dish soap online will feel tedious and archaic to many people.

As companies like Apple roll out products like Siri AI into products people already use, exposure to the capabilities of modern personal agents will grow. Trust will come easier as well—Siri already sits inside people’s messages, calendars, and email, and Apple has built its brand on privacy. It may well become the personal agent many people use first.

We have a multitude of examples in the history of experience design that show what happens next. Like the introduction of Google Search, personal agent adoption will change users’ preferences and behaviors, and businesses will need to adapt. But its impact on how we design experiences and conduct business will be much greater.

Outcome-oriented design for personal agents

While working at NN/G, we were among the first people to define the terms generative UI and outcome-oriented design (March 2024), building on Jakob Nielsen’s earlier work on intent-based outcome specification (June 2023). In April of this year, we argued that digital products now need to be designed for users’ agents as well as the users themselves.

At the time of writing those pieces, we could trace these undercurrents but not quite feel how they fit together in a new experiential reality. We now see that the rise of personal agents will soon combine these elements into a new experience dynamic.

In February 2025, Jakob predicted that agents would become the primary users of many websites and that the traditional user interface would vanish. But he was largely describing agents visiting AI-optimized websites on behalf of their users. We see the interaction between agents and institutions happening in the data layer instead, and then building a custom interface around that data when, or if at all, needed.

Personal agents will become data brokers. They’ll be the primary way that people interact with the internet. Most users will grant their own agent access to their private and sensitive data (communication, events, preferences, personal details, etc.).

Institutions (businesses, governments, retailers, etc.) will become data sources. They’ll pass their data, content, and interface suggestions to an individual’s personal agent through protocols like MCP (and, on the iPhone, Apple’s App Intents). In this way, they will influence but not control the end-user experience.

Those agents will spin up custom, personal, generated interfaces built in real time and fed by data-layer connections to institutions. Those personal interfaces will be perfectly tailored to their owner’s data, preferences, and needs.

The personal agent = a single, private data broker

Enter the truly personal agent. By our definition, a personal agent is an AI system deeply personalized to the individual who owns and uses it. It empowers its owner to supply as much personal information and context as they feel comfortable with.

Ideally, this personal agent also protects and defends that owner’s data and interests against outside influence, manipulation, and data abuse. There are obviously many ethical, privacy, and security risks with this future. (Many of which were predicted by Don Norman all the way back in 1997.) We’ll explore those in depth in another piece—along with our vision for how this might work safely and ethically.

These agents will have so much personal context that using third-party AI services that lack this context will feel tedious by comparison. (As people who have built our own capable assistants, we’ve felt this personally already.)

Consider: Why would you bother to manually enter your information in a long prompt into a company’s AI chat, when you can just have your own agent accomplish the task for you? It already has all the context it needs. You don’t need to expose potentially sensitive details of your life to companies you may not fully trust.

No other organization or entity has the full scope of access and authority into your life. They’re on a need-to-know basis.

In this role, the personal agent acts as the primary way that the individual interfaces with the rest of the digital world.

Institutions = data sources

The personal agent controls what information of yours is shared with institutions, requests data from them that you need, and then determines how that data is reformatted and presented to you.

This happens in two directions in the data layer:

• Agent-to-institution: The personal agent can send requests for data, content, and design. It can also share the minimum amount of personal information needed to process transactions on behalf of the customer.

• Institution-to-agent: The institution sends the requested data back. This can include any form of multimodal content that will be processed by the personal agent and may (or may not) be shown to the actual customer.

Imagine it’s time to replace your running shoes. Your agent tells a shoe retailer only what it needs to know: your size and the model you’re replacing. The retailer sends back comparable shoes with prices, stock in your size, delivery dates, and photography, along with suggestions for the interface should look, according to its design branding.

This is already beginning. On Nov. 3, Apple opened its new Siri AI to all app developers in a preview. Apps expose their actions and content, Siri decides which ones a request needs, and the app supplies a branded card for displaying its content.

Generated UI = the primary UI

When the agent receives that data and content from the institution, it processes it. This includes:

• Considering the range of products, services, or information the institution sent

• Conducting additional research or comparing against other institutions’ data

• Filtering out any options or information that isn’t relevant to the consumer’s task at hand

• Generating a custom interface that is perfectly tailored to the user’s exact preferences and needs (including accessibility needs)

For those running shoes, your agent compares the retailer’s options with other stores’ and drops anything that isn’t in stock in your size. Then it shows you a few finalists, from across retailers, side-by-side with the pair you’re replacing, so you can see what’s different, borrowing the retailer’s photography and styling where they help.

Proponents of conversational design may wonder why the agent will render an interface at all. Why not have the agent handle everything, getting the user’s input and decisions as needed via voice or chat interface?

We don’t believe a conversational interface is the best for every task. Even when the content and information is tailored to the user, they will still want to browse, compare, scan, and assess things for themselves. They won’t want to delegate every single decision to their agent, even if that agent controls what options the user sees.

All digital design will become outcome-oriented design

This is an absolutely massive change in the very definition of an experience, and in the relationship between a user and an organization. It means that design will not be about precise control of what the user sees.

Instead, design will become more amorphous and intangible. It’ll be about:

• Setting up the data and content layers so users and their agents have the information they need

• Establishing constraints and guidance about what is shared in response to agentic requests

• Providing that data in the format the agents require

• Sharing suggestions (in the form of branding and design tokens) that may influence the final output that’s shared with users… but never control it fully

Crucially, we’ll have substantially less control over what gets shown to users and how. There’s a possibility that this is immensely beneficial for people. As long as they truly own their personal agents, their behaviors and preferences are not collected, inferred, stored, and sometimes abused or sold by these various institutions and organizations they interact with. Their patience and time are not wasted by having pop-ups and promotions and intrusive notifications shoved down their throats. They’re not constantly being nudged to take an action that they don’t actually want to take or that may not be in their best interest.

But that assumes that these agents can and will act in their owners’ best interests (and that the agents can even understand what those interests are). If that is not the case, the potential negative impacts for people are huge as well. There’s a strong risk that businesses will not lose their ability to manipulate customers but will simply shift to manipulating the agent, which would much harder for the user to notice and avoid.

If these agents are provided as a free service (as Muse is), then the companies that own them will have the incentive and ability to sell that manipulation as a service, and mine the wealth of data to find new ways to manipulate consumers themselves. An agent that decides what information you see, what information you don’t see, and how that information is presented will have immense power over your perception and interaction with the world.

Again, we’ll unpack and explore these themes in future articles.

In-product agents vs. personal agents

This thesis will scare businesses and designers. And it should. Many of the tactics and strategies that have worked for the past decade will stop working as this shift takes hold. How companies adapt will make or break them.

As Sarah shared in her keynote at the Service Design Global Conference, food delivery offers a preview. A restaurant used to control the whole experience, from the plate to the atmosphere to whether the food arrived hot. Now it hands the meal to a courier and gives up the last mile.

Personal agents will take over the last mile of many digital experiences. The organizations that do well will be the ones whose data, content, and services are strong enough to stand on their own, whoever presents them.

Of course, this won’t be a switch that flips overnight. Consumers will sometimes want to visit the institution’s site or app directly. There will be many people who will never be comfortable sharing their lives with any agent, even if it’s a private, personal one.

Recently, Cory Doctorow made the argument that ecommerce will never adapt to agentic AI, because “these tools are a frontal assault on all the most exciting (and sleaziest) ways to make money in the waning hours of end-stage capitalism. Far from welcoming bots, ecommerce sites were always going to block them.” We think he’s right that many companies will resist this change, and that will significantly impact the timeline.

We’re believers in the power of experience to drive change, though, and the difference between personal agentic and non-agentic or non-personal experiences will become very hard to resist. The power of convenience will become the power of preference and behavior, which will increasingly drive institutions to adjust.

At first, the shift will happen in simple consumer services that are highly competitive and interchangeable. For example, Amazon currently works to block third-party agentic shopping on its site. It started blocking Muse shopping almost immediately after the Meta product launched. Amazon would prefer customers use its own in-house AI shopping agent, formerly known as Rufus but now Alexa for Shopping. That way, it collects all the data about what people are shopping for. It can control how results are presented. It can put its thumb on the scale and promote specific products without users even realizing they’re being influenced.

But if, for example, you’re shopping for a birthday gift for your niece, why would you use Alexa for Shopping? Your agent already has access to your calendar. It knows your niece’s birthday is coming up and how old she’ll be. You don’t have to explain what you’re looking for the way you would to Amazon’s assistant. And if Amazon isn’t playing nicely, your agent will surface gifts from retailers that are, like Walmart and Etsy.

Amazon won’t hold that line forever. On the company’s third-quarter earnings call in late October, CEO Andy Jassy said Alexa for Shopping will be “the primary shopping agent customers use to shop Amazon.” But he also hinted that the block on outside agents could change and said he expects Amazon to work with general-purpose agents.

Over time, the sheer pressure of consumer preference will influence institutional change. And from there, that preference will spread to other types of institutions. Probably not every institution, and certainly not all at once. This will take a long time to reach every corner of the economy. Plenty of organizations haven’t caught up to today’s experience standards, let alone the next ones.

Some organizations will have regulatory and legal obstacles, and others will have no competitive pressure to make them change. But even government agencies, which are arguably the least easily interchangeable service, may be motivated to embrace agentic transactions if it furthers their own goals (like encouraging citizens to take desired actions).

What we’re keeping an eye on

This shift won’t happen instantly. Even if the technology is capable, humans and organizations will move at a much slower pace. (This is what the AI hype advocates of 2024 forgot about with their timeline predictions.) It’ll show up first in a handful of signals we’ll be monitoring, and the organizations watching for them will have time to adapt.

What to watch

• Whether people stick with personal agents: Retention for personal agents like Muse, Instinct, OpenAI’s Dots, and Anthropic’s Claude Duet will start to matter more than download counts.

• Whether the walls come down: Amazon lifting its block on outside agents would be a clear sign that consumer preference is winning.

• Whether agents complete transactions: Today, AI-driven shopping still mostly ends with a person checking out themselves on-site.

• Whether anyone makes agents trustworthy by design. A major agent maker committing to protect customers’ privacy and autonomy would make personal agents more safely adoptable.

What companies can start doing now

• Find out what an agent can learn about you today. Ask a consumer agent to compare your product with a competitor’s, and note what it gets right, what it gets wrong, and where it gives up.

• Keep building AI features and optimizing products for agents. Most people don’t have a personal agent yet, and won’t for a while. The AI optimization you do today serves those customers now, and the work behind it (structured data, well-organized content, clear APIs) is the same foundation personal agents will need.

• Treat agent requests as research data. A surge of agent requests for one product signals interest, the same way page views do today.

• Translate your brand into suggestions. Design tokens, tone of voice, and imagery are how you’ll influence an interface you’ll never render yourself. (Many product teams have already done this for better collaboration with AI design tools.)

Up next in the series

This piece has been dense. If you’ve made it this far, we thank you for your time (and your choice not to simply have your personal agent summarize it, but no judgment either way).

There’s a lot more to unpack and explore, and we plan to do so in future articles. In the next few weeks, we’ll cover the implications this future could have for individuals’ privacy and independence of thought. We’ll also imagine and illustrate what this dynamic would feel like for a user completing everyday tasks through a personal agent and custom generated interfaces.

References

Donald A. Norman, “How Might People Interact with Agents,” in Software Agents, ed. Jeffrey M. Bradshaw, AAAI Press/MIT Press, 1997.

Jakob Nielsen, “AI: First New UI Paradigm in 60 Years,” Nielsen Norman Group, June 18, 2023.

Kate Moran and Sarah Gibbons, “Generative UI and Outcome-Oriented Design,” Nielsen Norman Group, March 22, 2024.

Jakob Nielsen, “Hello AI Agents: Goodbye UI Design, RIP Accessibility,” Substack, Feb. 21, 2025.

Kate Moran and Feifei Liu, “Discoverability of AI Features: Learn from Amazon’s Mistakes,” Nielsen Norman Group, March 21, 2025.

Ana Boyer, “Design systems and AI: Why MCP servers are the unlock,” Figma, Aug. 6, 2025.

Sarah Gibbons and Kate Moran, “AI Agents as Users,” Nielsen Norman Group, April 10, 2026.

Kate Moran, “Agentic AI Is Not Ready for Public Consumption,” Future Problems, April 24, 2026.

Barbara Ortutay, “Meta launches personal AI agent, Muse, emphasizes safety and privacy,” The Associated Press, Sept. 8, 2026.

Sarah Perez, “Viral AI agent Instinct raises $1B Series C at a $10B valuation,” TechCrunch, Sept. 28, 2026.

Lucas Ropek, “OpenAI launches dots, its bubbly agentic avatar,” TechCrunch, Sept. 29, 2026.

Slobodan Manic, “Amazon Blocked Meta’s Muse, And Robots.txt Had Nothing To Say,” Search Engine Journal, Sept. 29, 2026.

Caleb Sponheim, “The New Siri’s Biggest Strength Is Not Intelligence,” Nielsen Norman Group, Oct. 9, 2026.

Alex Veiga, “Meta Q3 revenue tops estimates and Zuckerberg says Muse accounts have topped 10 million,” The Associated Press, Oct. 21, 2026.

Michelle Hawley, “Claude Duet: Anthropic Previews an AI Agent That Messages You First,” VKTR, Oct. 22, 2026.

Cory Doctorow, “Agents vs ecommerce,” Pluralistic, Oct. 28, 2026.

Anna Hensel, “Amazon says more than 400M customers have used Alexa for Shopping in the last 12 months,” Modern Retail, Oct. 29, 2026.

Apple Developer, “Siri AI is now open to all developers,” Nov. 3, 2026.

That was the main point of the closing keynote I gave at the Service Design Global Conference in Wiesbaden last week. The room was full of service designers, people who already do this work, but my argument is for a broader audience.

Whether it’s a product team putting an agent in front of customers or a founder setting up an assistant to run the back office, anyone building an AI system is designing a service, even if nobody on the team would call it that.

When service designers explain what a service is, we almost always start with the restaurant example. The diner sits in the dining room, which is the frontstage (the host, the server, the menu, and the plate arriving at the table, ideally still hot).

Behind the swinging door is the backstage, the kitchen, where most of the work happens out of the diner’s sight. An order travels from the table into the kitchen and comes back out as a meal. Every part of a service is there, which is why the example has lasted, and why it maps so neatly onto the service blueprint Lynn Shostack introduced in Harvard Business Review in 1984.

In my keynote, I used that same (metaphorical) restaurant to show where AI agents are starting to appear. Some agents work in the kitchen alongside the team, doing the prep the diner never sees. Some carry the plate out and deliver the service to the customer directly (think AI support bots, like Cursor’s Sam). And some wait at the door like a delivery courier, working for the diner rather than the restaurant and picking up an order on someone else’s behalf (think Meta’s Muse or Instinct AI).

Kate Moran and I run Gibbons & Moran on a system built this way. Kate and I are the diners. Each of us has an assistant who works with us directly, Fern for Kate and Iris for me. They are waiters who can also work in the kitchen. Behind them is Rosemary, a shared agent who works almost entirely in the kitchen: she files the notes from our calls, writes a call sheet before the next one, and brings the company record up to date every morning before either of us is awake.

Work passes between them the way an order passes through the kitchen door. For example, take a single client call. Rosemary preps a call sheet from our calendar, email, and past notes, files what we promised afterward, and by seven the next morning Iris or Fern is picking up the next step with each of us.

All three read from one company record, and we wrote down rules about who may read it, who may change it, and who may send anything on our behalf, the same way a restaurant decides who handles the money and who talks to the guests.

Those are the parts of a service. Getting them right is system design work, regardless of who does it.

• Who does the customer actually work with?

• Who works out of sight, and what do they need from other actors or systems?

• What context and responsibility passes from one to the next, and what could get lost on the way?

• Which record does everyone trust, and who is allowed to change it?

• When does an agent need to stop and bring in a person for approval?

• When does an agent find someone who knows something it doesn’t?

Service designers and system designers have been asking these questions for their whole careers. Many teams building with AI are answering them for the first time.

recently wrote that he was wrong to think people would need to organize agents, because the agents turned out to be good at organizing themselves. I think he’s mostly right.

However the harder decisions are at the metaphorical kitchen door, where agents hand work to humans, act on someone’s behalf, or change a single source of truth. Deciding what happens at that door is a separate job, and in a lot of companies right now nobody owns it.

I think this is where service and system designers come in. Most AI systems in use today are still fairly simple, with a few agents and a few handoffs between them. They won’t stay that way.

Before long, a client’s own agent will be engaging with products and services on that client’s behalf. Every new agent adds interactions and every new tool adds rules about who may do what. Soon the hardest part of the system will be managing everything that passes between agents and people in a single day.

Service designers have been practicing two skills that matter here for decades: mapping handoffs and mapping context.

When work passes from one actor to another, so does responsibility. The goal is when something breaks, you know where it broke and whose responsibility it was. When we first set up our system, Rosemary booked client calls without telling Iris and Fern, who run our personal calendars, and I ended up double-booked. This was one problem for one person (me) and easy to troubleshoot. The same gap is a bigger deal at scale and within more complex systems, where a dropped handoff has serious repercussions.

Context is what each actor needs to do its job well, and it’s one of the biggest factors in how good an AI system’s work is. AI engineers call this context engineering, and it’s the same problem service designers have been mapping for years.

Teams building AI systems need both, and service designers already know how to do them. I’m looking forward to seeing what happens as those two groups find each other. Helping businesses with the strategy and talent at this intersection is part of what we’re doing at Gibbons & Moran.

References

• G. Lynn Shostack, “Designing Services That Deliver,” Harvard Business Review, January 1984.

• Anthropic, “Effective context engineering for AI agents,” September 29, 2025.

• Ethan Mollick, “The Dot and the Swarm: Benefitting from the Bitter Lesson,” One Useful Thing, October 1, 2026.

• Benj Edwards, “Company apologizes after AI support agent invents policy that causes user uproar,” Ars Technica, April 17, 2025.

• Meta, “Introducing Muse: The World’s First Personal AI Agent Built for Everyone,” September 8, 2026.

• Instinct, personal AI agent.