#101 Designing for AI Agents: The Future of UX Beyond Interfaces
13:40 Share

Welcome to the future of UX, the podcast where we explore the trends, the challenges and also some ethical dilemmas sometimes shaping the future of design. I'm Patricia Reiners and in each episode we dive into the intersection of UX, technology and the future of design. So you can stay ahead of the curve. Today's topic is one of the biggest shifts happening in UX right now.

If AI makes its own decisions, do designers still matter? For years, UX has been about designing interfaces, designing buttons, screens, navigation flows. But with AI agents, things are changing. AI doesn't respond to user input, it acts on its own. Instead of clicking through menus, users will talk to AI assistants or let them handle tasks automatically or autonomously.

And we are moving from user interfaces to user-agent collaboration. So the big questions for this episode are: How do we design for AI agents, not just static interfaces? How do we ensure AI is trustworthy, predictable and user-centered? And the big one: Is AI design shifting from interface design to behavior design for AI?

I would say let's dive right in. Imagine you're trying to book a flight and a hotel, maybe for your business trip. You open five tabs to compare the different prices on different websites. You manually enter all the dates and the locations and the preferences multiple times on each website. And then you basically shift through confirmation emails and then set reminders for your trip. Now imagine instead you say,

Hey AI, book me a flight to Berlin next Monday, find a hotel near the conference and add everything to my calendar. This is now possible with the new feature, the AI operator from ChatGPT. It's pretty interesting. Unfortunately, it's not included in the premium version, but you need, I forgot how it's called, but you need to pay around like $200 or so per month.

But this AI operator is connected basically to your different platforms, to the different services that you use, and it does things for you. So it books a flight for you. It searches for a recipe and then buys ingredients. And the interesting thing is that you can watch it doing all those things on the side. So you see it clicking, you see it scrolling, you see it searching, and you can just sit back and relax and relax.

You see that AI does things for you without you needing to interact with multiple interfaces. AI does that for you. And I think it's absolutely fascinating to see that. I'm also going to link a few references in the show notes. So feel free to check it out. Have a look at the videos. Maybe even test it yourself.

When we think about the traditional UX or like the UX or how users are still interacting with websites, it's the user clicks, the user searches and navigates through and interfaces. And when we think about how agents could change that,

uses delegate tasks to those agents that interact autonomously. JGPT's AI agent or the AI operator is not the only example. There are also from Google, the Google Gemini AI agents or the Salesforce agent force, which handle workflows without the human intervention.

And when I see these things, when I explore these things, when I try these things out, I always think, what does it mean for your ex-designers? And thinking a little bit ahead, a little bit into the future, it means that we are now just designing interfaces. We are designing behaviors. So users just, you know, or...

How do you say it? Users don't just need basic usability anymore. It's not just about usability. They need trust, they need predictability, and they need control over AI.

So the first really big key takeaway is: UX isn't just about where a user clicks, it's about how AI makes decisions on their behalf. And I know this sounds still so futuristic, but this is already what's happening following ChatGPT, the AI operator. Make sure to have a look. It's absolutely fascinating to see that from a UX perspective and it already shows what is possible.

By the way, if you already had one key light bulb moment or if you're really enjoying the podcast episode, I would be so happy if you could give us a five-star review on either Apple Podcasts or on Spotify. That would be such a huge support for me, for the team to find good people for to interview or yeah, to just make the podcast even better. So thank you so much for your support. I really, really appreciate that.

Okay, so what are some of the challenges of designing for AI agents? With AI making its own decisions, we UX designers face some challenges. What is that? The first and most important is trust and explainability. So how do users know what AI is actually doing? AI is powerful.

but it's also a black box more or less. So if users don't understand why I made the decisions, why won't trust it? And I think the ChatGPT AI operator does that very, very well because you can basically watch it booking that flight. You see the cursor scrolling, you see it clicking on something. And on the other side, like the little panel where you communicate with ChatGPT AI,

It also highlights what it's doing. For example, searching and then there's this little loading animation or clicking on something. So you can basically watch it and see what it's doing. So it's very helpful when a chatbot explains also why it recommends certain solutions and allows users input. What it definitely should not do is make decisions without context or leaving users frustrated.

So providing transparency, you know, here's why I made this choice or also helping users to see what it's doing. So helping users to understand how it actually came to a certain conclusion, you know, make users watch how the flight was booked or give users control. Would you like to adjust this decision? For example, when the AI operator, when the agent

you know, puts all the things in the shopping cart before ordering it, double checks with the user. Are you sure? You know, is this all? Is there something missing? You know, please double check and create these feedback loops. So let users refine AI behavior over time. Give feedback all the time. So AI is great at automation. But what if it oversteps? What if?

AI auto-pays for bills without asking. What if AI deletes an email if it thinks it's spam, but it wasn't? There's an example, the Gmail AI smart replies. So users love auto-replies like, oh, thank you, and these kind of things. But what if AI automatically responds to an important client email the wrong way without having the user double-check it?

So AI should assist but not replace your decisions. And I think this is super important. It's not about automating everything, but automating certain things and then leaving the last decision to the user. So the user should always have the final say in important actions. So the second key takeaway is UX for AI is about designing the right level of automation, not complete automation.

So it's too much and users lose control. So too little and the AI feels useless. So there's this balance between too much and too little. And as AI becomes more agentic, UX needs new principles and

It's very difficult to define these principles, but I thought a little bit about it. What would be helpful? Also thinking about the past, the future. So that might be a little bit of a guideline, but those are not picture perfect principles that you should use. Think about also your own principles when you are designing it. I think the first principle from my experience or from my perspective is predictability. AI should not feel random.

Users need to know what AI will do before it acts. So Google Maps reroutes you based on traffic, but it shows the new route before switching. So you need to say, okay, yes, I want the new route because of the traffic. The second thing is explainability. And we talk so much about that, I think, in the podcast at the moment. So AI decision should be clear.

Users should understand why AI made a choice. Super simple. ChatGPT, for example, Perplexity AI, they show the sources and the reasoning for its answers. And showing the sources where information is actually coming from is such, I would say, like a game changer for people to understand where it's coming from, to double check, to use their critical thinking, and to make sure that the information is right. And

AI should always allow corrections. So one feature that I really love is Gmail's undo send button for AI generated replies. Something like this. Or yeah, that you can basically undo something. And the last one is adaptability. So AI should learn from user preferences. Spotify's AI adapts to your music taste over time.

So it always adjusts. So the first thing predictability, AI should not feel random, but should ask the user before it does something. Second, explainability, AI decisions should be clear. Understanding why AI made a decision or made a choice, then AI should always allow corrections. Undo send, for example.

and adaptability, AI should learn from user preferences and adapt their decisions or their behavior. So the key takeaway here is that AI should feel like an assistant, not a syndictator. So now thinking about the future, will UX design stop being about screens and about interfaces and become about shaping AI behavior? I would say 100% yes.

AI eliminates traditional menus and traditional buttons. It also learns the context of preferences. So the manual navigation becomes sometimes unnecessary. For us, it's so difficult to get our head around that because we are so used to design a search bar, to design a menu, to design a button. But AI experiences are about conversation, about automation and not so much about clicking.

So we may shift from interface creator to behavior architect, teaching AI how to interact with humans. And to summarize this a little bit, AI agents won't replace UX designers, but they will change how we design. Curious to hear your thoughts. Do you think AI agents will reduce the need for traditional UX designs or do they make UX more important than ever?

Feel free to share your thoughts. Drop me a message on Instagram at ux.patricia. Feel free to say hi. Let's connect. Or let's connect on LinkedIn at ux, no, at Patricia Reines. Or follow the podcast on LinkedIn. I'm going to share everything in the show notes. And yes, thank you so much for joining, for supporting the podcast. And I would say thank you so much for listening. See you in the next episode.