Sentient Design and the Future of AI-Powered UX with Josh Clark and Veronika Kindred

What happens when AI becomes more than a tool and starts shaping the interface itself? Josh Clark and Veronika Kindred of Big Medium, co-authors of Sentient Design: Crafting Intelligent Interfaces with AI, join Karen Lynch to explore a new approach to designing AI-powered experiences. They discuss treating AI as a design material, moving beyond chat-based interfaces, and creating products that adapt to users’ context and intent in real time.

The conversation also examines one of AI design’s biggest challenges: trust. Josh and Veronika share approaches for building transparency, explainability, and healthy skepticism into intelligent systems. For UX researchers and insights professionals, they explain why user research becomes even more important as AI changes product development—and how researchers can take a more active role in shaping the products themselves.

Key Discussion Points

• Moving beyond chat with AI: How sentient design uses AI as a creative design material to build intelligent, adaptive interfaces rather than simply adding chatbots or AI tools to existing products.
• Designing experiences that adapt in real time: How interfaces can respond to user context and intent, including dynamically generated dashboards and AI agents that work alongside users within familiar environments.
• Building trust into AI-powered products: Why AI’s confidence can be misleading and how transparency, explainability, visible sources, and “productive humility” can help users evaluate AI-generated outputs.
• Balancing AI opportunity with risk: How designers can determine where AI is appropriate, identify unacceptable risks, and build safeguards into systems where failure has meaningful consequences.
• The evolving role of UX research and insights: Why researchers need to participate throughout design and development—and how research findings can become direct instructions that shape how AI-powered products behave.

Resources/Links

• ⁠Sentient Design: Crafting Intelligent Interfaces with AI - Use code SENTIENT-GREENBOOK for 20% off your purchase of the book at Rosenfeld Media through September 30, 2026.
• ⁠Big Medium

You can reach out to **Josh Clark** and Veronika Kindred on LinkedIn.

Many thanks to Josh Clark and Veronika Kindred for being our guest. Thanks also to our production team and our editor at Big Bad Audio.

Transcript

[Karen] Hello, everyone. Welcome to another episode of the Green Book Podcast. I'm hosting today. I'm Karen Lynch, and I'm excited about this episode because it doesn't always happen that I get to introduce an author to the podcast. I certainly don't always get to introduce two co-authors, but today that's what's happening. I'm speaking with two co-authors of a new book on Sentient Design: Crafting Intelligent Interfaces with AI. This one's going to be a great one, friends, because obviously AI is the talk of the town, but also I have so many listeners in our audience that are in the world of user interfaces and UX design, whether they're designers or product leaders or product designers, design-minded developers that I think really would appreciate learning from these two individuals. I have Josh Clark here, the principal of Big Medium, who's a digital agency, and his colleague Veronika Kindred is here. I think it's best for them to introduce themselves, so I'm going to very quickly turn it over to you, Josh, to introduce yourself. Welcome to the show. Please tell our audience a little bit more about yourself.

[Josh] Hello, Karen. Thanks so much for having us. It's a real treat to be here. As you mentioned, I lead a digital design agency called Big Medium. I've been doing product design and product strategy for 30 years now, 25 of those years running Big Medium, and so I've seen a lot of different shifts over the past few decades, AI the latest, but on the web, mobile, design systems, sort of like all these things, and I've written several books about this stuff. So I've thought over this time, as one does with experience, oh, I've kind of figured out this user experience thing and product design, but the more that I've worked with AI and machine learning over the last decade and with LLMs, especially in the last few years, I've realized that a lot of those closely held assumptions, hard-won best practices need revisiting. And that is where this one comes in. Veronika?

[Veronika] Thank you. Thank you so much for having us. I'm Veronika. I'm a designer and researcher at Big Medium, and I am co-author, along with Josh, of Sentient Design. And for those listening, maybe you could tell I'm a little bit younger than Josh. Haven't quite been at this for 30 years myself. But really, my career has focused first on design systems and now on designing with AI. I mean, really, I kind of grew up in this world beyond static websites, where it's like everything is kind of ruled by algorithms, thinking of TikTok. But really, my career has been AI-native. I graduated my senior year when OpenAI released ChatGPT. So my whole career has really been kind of in this post-LLM world and just like thinking of new things that we can do with this new technology.

[Josh] Yeah, I mean, I think Veronika doesn't yet have my experience, but neither is she burdened with it. And, you know, I mean, I think you're looking at two generations of designers here, and literally so. Veronika is not just my colleague, she's my daughter. I'm very proud of her.

[Karen] Wow, that is a bit of a spoiler that I did not see coming. Congratulations on the collaboration. So beautifully played there.

[Veronika] I always blush.

[Karen] That's great. That's great. And this is when we step off totally off the brief for a while. When I was in my late 20s, I was the research director at my father's consultancy. And yeah, and it was really wonderful. He was a quality control consultant in management consulting. And he, many of his clients were looking for most largely customer satisfaction work. Lots of, you know, kind of surveys, customer satisfaction surveys, but also some in depth qualitative interviewing going on in that space. And he really needed a pro. And that was the field that I had, you know, entered into. So it was probably some of the finest years of my career was kind of working alongside my father. I will never forget it.

[Josh] That is so cool.

[Karen] Yeah, yeah. So warms my heart right away. And I also think, and again, this is not what this podcast is about. So to all of the listeners listening, this is not what we're going to be talking about. I think that there's so much that you can learn from the different generations. We have lots of people here on my team. And you know, Veronika, I don't know if you can tell, but I also have clocked 30 plus years in this industry. You know, having started

[Veronika] Would never have guessed.

[Karen] Yeah, well, thank you for saying that. But I, I did start in the 90s. So it does add up. But there are lots of people on the Green Book team who are in their 20s right now. And I learn from them every day. And I think that I feel very lucky and very privileged to be working alongside of them. And I, I think of them really as peers who are teaching me and, and I don't, I never feel that that age is a barrier to our collaboration, because we are very, we just feel very fortunate, I think, to be working alongside each other. So I think that comes, that comes with mutual respect. So clearly, if you can co-author a book together, you can probably, you know, have a successful career, you know, ahead of you, Veronika, and also, you know, Josh for the years to come. So congratulations.

[Josh] Thank you for that. And I think you're saying something that's really true. And especially at sort of moments of change, where it's experience helps to weather shift and change. But often these things require fresh experience. And that old heads like mine, really benefit from the young heads that we're talking about. Yeah, I think it's more important than ever to get fresh perspective.

[Karen] Yeah, yeah.

[Veronika] Conversely, on the other side of that, I think it's been really great for me as someone like going through all of these kind of earthquake changes in the industry, and also like in academics, to have a collaboration with Josh and really have someone with so much perspective, somebody who's like been through AI boons and winters before and also like, was kind of in the industry for like dot-com and for the burgeoning of mobile and like all these things who can really help like, put everything into perspective. So yeah, it's been a great collaboration.

[Josh] Now I'm blushing.

[Karen] I love that. So this podcast is released in video. But for those of you who are just listening to the audio, see you're missing out, get on YouTube and check out the video as well. So let's dig into this book, because we will have some links in the show notes, but there are at the bottom of the kind of the Rosenfield media site for the book, which I was lucky enough to get a copy of to read before the episode. You have some praise from some great sources, and Notion among them, which I was really excited. I think it's the head of design of Notion. I don't have it all right in front of me, but we're Notion users. And I'm like, well, that is something. And thinking about the people that have read your book and commended you on your book. And part of what they're saying is this is a new look at the field based on the new tools of today. And I think that's really important because the field hasn't necessarily changed, right? But the tools that are available to us have. And the need to understand customers hasn't changed. That is still there. But AI has changed how we are considering that. So why don't you give kind of the high level overview? Like if somebody were to say to you, the premise of your book, which we can read on the site, but I'd love to just kind of start off, what's your high level premise for your book? What are you telling people?

[Veronika] So sentient design is really about designing intelligent interfaces that radically adapt to the user experience. And our core kind of belief, you know, this sentient design offers both a framework and a philosophy. And our core belief really centers around using AI as a design material, rather than using it kind of as just tooling or processing. We see so much focus right now in the industry of people integrating AI in innovative ways, but really as a tool, like a kind of engineering perspective of AI where it's very bolted on or bringing it into their process, but it's not necessarily making its way into the product yet. And so sentient design for us is really about how to think about using AI as a new material in design rather than just a tool.

[Josh] I think one of the things that's exciting about that is that it frames AI as a creative platform. But I think right now there's like a lot of fear and uncertainty about AI. Is this replacing what I do? It's certainly creating demands to do what I do faster. And what about quality? That this sort of push in this era of AI for grinding out efficiencies, for making what we already make or doing what we already do, but faster. You know what I mean? It makes sense as an ROI kind of perspective. Savings are important, but what about the opportunity to create new value of creating extraordinary new experiences? And I think one of the premises of sentient design, this idea of creating intelligent interfaces, is we can make something we couldn't make before. Things that are newly possible, that we've been talking about adaptive interfaces and personalization for decades. And now we've got a technology that can actually do it, or at least much better than it could before. What can we do with that? How can we answer customer needs and business needs in ways that just weren't possible? And so I think one of the things that is exciting about this is this isn't a dreary, diminishing, grinding out new efficiency stories. You've got a new material that make new things with that can really make a difference. And that's exciting to us. And I think it's a story that is weirdly not being told very much about AI.

[Karen] Yeah, Josh, one of the things I did want to talk about were some of those experiences. And I was wondering if you had examples of that, to bring it to life for our audience right away so they can get on board with us a little quicker. Some examples of those product experiences that are showing where the shift is happening. Do you have any up your sleeve?

[Josh] Yeah, we have a lot of them. Literally, there are hundreds in this book.

[Karen] All right.

[Josh] Well, we don't have to do hundreds. I promise we won't. I promise we won't. But no, this is not a speculative thing. This is something that some forward-looking companies are already beginning to do, and that we all have the tools and ability to sort of take advantage of this. So when we're talking about intelligent interfaces, what we really mean are systems that have awareness of context can infer intent in ways that we haven't been able to do before, and have the agency to adapt the interface in the moment. So a way to think of this is, what if the system was smart enough to make design decisions in the moment to adapt to the user's needs? So instead of creating sort of static views for a lowest common denominator view of the user, that we can instead make a change to adapt to the current need. So we do a lot of work with large enterprise companies. We're doing projects, sentient design projects, right now with like three Fortune 50 companies. And a lot of their products, right, it's like traditional enterprise software, SaaS views. It's like, oh, we've got a million different contexts. So our answer is, we're going to have a million different dashboards, an incredibly dense information architecture for people to navigate and find. One of the patterns that we have called bespoke UI is something where the interface can just assemble as you need. So Salesforce, for example, has something called the generative canvas, which is basically like, what do you need? Just ask for it. And instead of getting search results, you'll actually get like a dashboard that assembles the information that you need from trusted sources. The AI isn't making it up. It just knows where to look for the information. And sort of like a design system of widgets. So it's the kind of thing that instead of having to navigate to a view that does this, the system builds this bespoke interface on the fly. I would say that's maybe one of the most kind of graspable, easy concepts. It's like, oh, it's a dashboard that builds itself. But it can take a whole bunch of other sort of views, where, for example, the system can take the form of another user in a multi-user environment. So if you think of something like Figma, which has other people flying around with their cursors, what happens when AI is driving one of those cursors? Clearly labeled, clearly sort of scoped functions, but that it can add things to your Miro board, or it can do certain repetitive tasks for you, and you see it doing its work. So instead of going off to some sidebar chat or a different application entirely to talk to AI, it's participating within the established paradigm in a really sort of natural way. You can see its actions and its traces the same way that you would see it through one of your human colleagues. We call that the NPC patterns, like non-player character, like from gaming.

[Karen] Yeah. So if you've ever used a site like Miro or something where you're seeing the little circles, sometimes they're initials, sometimes they're little animal icons or something like that. This is just AI would be in there, and they'd be moving around as if they were a human being.

[Josh] That's it. That's right. And I think right now we start with this assumption that AI and intelligent interfaces is chat. And that's longstanding from Turing's imitation game notions of like 75 years ago, over a century, we've been talking about AI or systems that can talk. And it was a thought experiment that kind of turned into a design brief somehow. And there's a huge gravity around the idea of like, oh, it has to be chat. Sentient design sort of challenges us to say, what if we could weave that kind of responsive intelligence that we have in a conversation into any interface? And what becomes possible? Sometimes it elevates traditional interfaces. Other times it creates entirely new experiences.

[Karen] Yeah. So it's the world of what's possible. So with that in mind, what are you finding the response is? So now I'm kind of stepping outside of the book for a minute and just thinking about this type of practice, right? When we're thinking about intelligent design and we're thinking about, okay, we're doing this. And now in practice, when people are embracing this, right, this concept, what is able to change?

[Veronika] It's been really interesting to see how this kind of practice has actually been able to spur new types of products. So one company that we work with, they're data specialists. They do data security really, really well. And that's really their thing. And we came in and we started doing sentient design work with them. And by adding these kinds of layers of design that are able to not only grasp and contextually understand the data, but then also answer questions about it, it's been able to spur a whole new line of business for them. So that's been really, really cool to see.

[Josh] And actually, do you want to talk about the form of that? That it's like, this is sort of a new way to explore data because AI has enabled it.

[Veronika] Yeah, that's right. It's combining a couple of things that we've already touched on, including this NPC or this gaming pattern. It's really allowed users to be able to explore data on a dashboard. And you can kind of initiate the interaction with the AI, which will then start going through and going through the different steps that it takes to answer your request, whether that's like wrangling, understanding, maybe it's going to use this data to fill out a new form or answer other questions. But because it's happening visually rather than in chat, there's this really visually auditable trail that it's going through. Because it's showing up like another user, you're able to see what it's doing, where it's doing. And it's like, yeah, if you're working in Miro or Figma and you have a colleague and you want to see what they're doing, you just go click on their profile and then you can be like, okay, this is where they're at right now. That's great. Oh, maybe they're going the wrong direction. Or like, I love this, maybe I'll branch out. You can do the same thing with AI because you're able to interact with it visually. And because it's a client that's very interested in data security, it becomes this really transparent trail that's really auditable. So you can see exactly where the conclusions came from.

[Josh] Yeah. And sort of paint a picture of like what this is. It's like, imagine a visual canvas, like a Figma or a Canva. And you can ask a question of the canvas of like, we're using this for fairly serious things, like things like financial investigation, where it's just like triage the 3000 reports of suspicious activity for interesting things. And an agent will come in and we'll start by showing visualizations of like, oh, here's a beeswarm chart that shows kind of like the distribution of things. And it's like, oh, these 40 are interesting. It's like, tell me about those 40, show me a distribution of the recent deposits. And what you start to get is this bloom of data visualizations of AI and not only saying, here's, let me fetch that data for you. But what's entirely new to this company is choosing data visualizations to demonstrate it so that in the case of like a complex investigation, you're getting building on the fly that detective’s pinboard of strings to images and showing the train of thought and connections so that you're visually able to do this storytelling. And you're able to sort of see where AI got this stuff. So you can have confidence that it's working, which is such a big, important piece of this. Is like designing for trust.

[Karen] Yeah. Yes. And I want to get back to that designing for trust part that you just dropped. It's like, okay, let's sprinkle that over here for just one minute. Um, because I still, I want to go back to the bloom for just a minute, because in my head, what I'm picturing is, um, is again, something I had wanted to talk about, but I'm kind of caught up in what's happening in my mind is this blooming concept. So what's happening is we're kind of exploding the possibilities here of where we can go with the design elements when we step beyond chat, but also there's this new way of getting kind of user feedback also beyond chat, right? You're getting an interpretation of user needs that are changing moment by moment, right? Context is changing. Um, their intent, things are changing and this is very applicable to just user research in general, right. And understanding the usability part of all of this, right. And, and the user experience research and how it plays into design. And so that’s what's happening in my brain right now is these, the overlap of these two things, right. The design aspects of this and the UX researcher who might be listening and thinking about how this applies to their job. So when you mentioned bloom, I start to see, okay, it's about to get big in my head. That's where it's getting big. So I'd love for you to connect some dots for me about how, um, a researcher who might be having all of these big questions might be able to wrangle what you're saying right now.

[Josh] Well, the good news is that user research is needed more than ever right now. I mean, it's like if we are to create systems that are aware of specific context of what they're doing, we need to really understand what that context is and teach AI what to do it. So I want one thing I want to sort of say is like, Claude doesn't know how to do this stuff in really specific domain areas or the specifics of somebody's role. Like it might be able to make some okay guesses, which is not a bad place to start, but the work here in sentient design is really behavior design at its root, right. It's sort of like the things that we're talking are sort of familiar design patterns that humans have done, but it's like, Oh, wait a second. How do we teach AI to assemble the right dashboard on the fly, where to get the information, what the right presentation of that is? When does a financial crime investigator need to see this type of visualization, but somebody over here who is, you know, a landscape architect wants to see something completely different in a similar context. That is about understanding the user and translating that into instructions and context for AI. So, so much of this is behavior design of like telling it the context that it needs and giving it sort of procedures to follow in ways that are broad enough that we’re taking advantage of what's new about AI, which is to be able to handle fuzzy situations and apply sort of some reasoning. So it's not, here's all the specific rules, but it's like apply some reasoning to fit into this rules, the rules of this environment that we created. So this might sound a little unsettling, like some of the stuff we're talking about, it's like, wait a second, this is like AI designing the thing on the fly. How do I have any groundedness? And really the idea is like, you know, you design this universe of rules and interactions that are predictable in themselves that both people and AI will work within. And we're going to teach AI how to work the machinery within those rules. So there's like some predictability and groundedness, but we need the user research and that contextual domain knowledge to be able to teach AI to do the right thing. So it is more important than ever.

[Karen] So at what point, maybe Veronika, you can answer this. So at what point does that user come in or, or at what point does, you know, does the researcher or somebody on your team bring the user in to conduct that research, to help kind of guide the AI in the subsequent steps?

[Veronika] Yeah, great question. Like beginning, middle end. I feel like what's been so interesting in our own work and kind of like talking to other people in the industry as well, is there's kind of the, it's just been the collapse of like the waterfall sprint and the collapse of the idea of a project going through discipline phases. I feel like now everyone is working so closely together and things are so iterative. As a designer, I'm working in code, you know, devs are designing next to me. It's, it's everyone, this, the cycle is just so fast such that it's, it's really needed at every stage. Like, I mean, to start a design, the first question you should be asking is, okay, what's the problem we're trying to solve? You know, you need research for that and you need it all the way up to the end when you're like, does this accomplish what we set out to? So yeah, all over the place, all over.

[Josh] Yeah. I mean, in a perfect world, right, that's how every project works. I mean, Karen, we don't have to tell you or your audience, like research should be involved straight through. Through the validation, through the acceptance, the whole thing, right? And so often it's sort of assumed it's like, this is an upfront thing. And all right, see ya. Great. We've got your research. We'll take it from here. Like Veronika is saying, what we found is that there's this opportunity to work together through all phases of design and development and actually contribute, not just like review. I mean, this actually changes the role of user research a bit because user researchers should be part of writing that system prompt, that context, the procedures, like capturing and documenting. This is how people work. These are the outcomes that they want. That's stuff that needs to be mainlined to the AI to make this stuff work. So this actually brings user research into the production phase, not just in terms of advisory, but actually like contributing to the code, which is sort of an odd thing, but code here is a description, right? It's an AI, it's AI context that we're giving it.

[Karen] There's that feeling with AI, right? Where can we trust it? Is it leading us in the right direction? And yes, user research can inform it. So we have a little bit of trust when we know that we're bringing users into the process. So hopefully it has learned, hopefully it's going down the right path, but how do you build that trust? And when are you in danger of maybe putting too much trust? Not you, I don't mean you two. When our teams out there in the world, kind of in danger of either over-trusting it or not trusting it enough. How do you walk that line? Let's explore this space for a little bit because I think that's a big area.

[Veronika] Yeah.

[Josh] I want to suggest at some point you should tell your Leonardo DiCaprio story here. Sort of like what AI is good for and not.

[Veronika] All right. I like that. We're going to answer the question. We're going to take the long way. I think there's this movie starring Leonardo DiCaprio called Catch Me If You Can, where he plays a master impersonator and con man. And he plays this like 18 year old boy who runs away from home. And rather than doing what most runaways do, he starts to impersonate all these different people. So first he impersonates an airline pilot and he gets flown all around the world by finding this uniform and starts cashing checks, which creates his own problems. And then he...

[Josh] He's not doing the flying though, right? He's just like he's riding along.

[Veronika] He's not doing the flying, but he's like walked on and able to sit in like the pilot seat because of his great impersonation skills. And then he becomes a doctor, which like similar thing. He's not actually practicing medicine, but he's wearing the coat and people are talking to him like he's a doctor. And then in the end, he becomes a lawyer. And really, when you think about it, AI, LLMs are exactly like this character where they're really, really good at taking on the manner and the form of the answer, but they're not actually equipped to give you the answer. Like Leonardo DiCaprio should not be flying the plane. LLMs should not be practicing medicine, but they are really, really good at like giving that presentation layer of convincing. And really, it's what they're built to do. Like they're probabilistic models, and they're built to show you with 100% confidence what they think is true. And the miracle here, like what's really crazy is that oftentimes they are actually true. So it builds in that like incredible trust of like, where I'm like, okay, I need to go talk to my therapist, ChatGPT, what do you think? Not that I actually do that, but...

[Karen] No shade to using chat GPT to ask that sort of thing. I'm the first to admit, but I have often said, hey, chat, nickname chat, what do you think? But anyway, please continue.

[Veronika] Exactly. But really, it becomes this design problem in how we are building the systems for the users and how we're teaching the users to trust this system. You can't really rely on an LLM who's basically an impersonator to be like, by the way, I'm totally tricking you right now. Like, that's not what they're built for. That's really not at their core, their character.

[Josh] Always confident, right? They're always confident.

[Karen] Yeah.

[Veronika] 100% of the time, they're pretty sure, which is not great. So it really behooves us as designers and as builders of these systems to do two things. One is to encourage some productive humility on the part of the system. So leaning into designs that surface different kinds of confidence. There's a stat, Netflix used to surface a piece of data, which was like how likely you were to like something. So it'd be like, okay, you are 75% likely to like this movie. I was like, what does that mean? I don't know what that means. Like I said, what part of it, what are you basing this on? And there was some research done that showed that percentages don't really mean much to people. But if I were to say to you, like, I think you'd like this movie. You're getting from a lot of different things, not only my words, but also my tone that like, you might like this movie or you might not. Like, I don't know, 50-50, you know? And so in that way, systems can surface that kind of productive humility. And one of the ways that this is done in design is like Google, for example. It used to be that when you Googled are reptiles good pets, the answers would differ drastically from if you asked if reptiles are bad pets, because there's so much bias inherent in the question. And what they do now is, and we call this pattern spaghetti scenarios, spaghetti kind of like an overlapping weather map where there are multiple probabilities that could happen, is when you ask that question, it will surface similar questions. So if you ask are reptiles good pets, it will also surface that little design element. So it's saying, did you also want to ask are reptiles bad pets? Do you want to see how the answers differ? So kind of just like showing you the bias inherent in the system and also in your own question. And then the other thing I think you can do is, so in addition to encouraging this productive humility on the part of the system, I think is encouraging this healthy skepticism with the user. So doing things to really keep the user engaged so that they're not totally taking the backseat. And you can do things like surfacing contextual information that would help them make a different decision. There's this one example from several years ago where Google Maps offered an alternative route for a bunch of people. And what ended up happening was that in a small town, I think it was in Colorado, Josh, was it in Colorado?

[Josh] Yeah.

[Veronika] Maybe. Colorado, great. There ended up being hundreds of cars stuck in the middle of this field because what Google Maps didn't surface was the fact that it was a really muddy road. It wasn't an actual service road. It had just rained. And like all of these other risk factors that it didn't really show people. And if it had surfaced these things to people, the individual drivers might've been like, I'd rather just stay in traffic an extra 10 minutes, but it presented it as equal options. So yeah, long rant, but I think there's a lot of things that we can do to encourage both users and systems to be kind of humble in their outlooks.

[Josh] You know, Karen, there was a lot there, but just to pluck out, there's like a few things to underline just as bullet points. You can't trust AI for facts. It's sketchy on facts. It's fantastic at manner. And it's good at getting facts when you tell it where to get them. So it's like, oh, presentation. Leonardo DiCaprio should be the MC. Don't let him fly the plane. So actually this dovetails to sentient design. How shall we present information? Not what is the information? And two things I think that Veronika was underscoring there was how do we do transparency where we're saying these are signals, not facts? And how are we transparent about that and about bias, which is what she's talking about with the different Google questions? Like there are five different ways to ask what you're looking for and you're going to get different answers. And you see Google explicitly doing that now, right? It's like, here's five different ways to ask what you're looking for. And the answers are different. So it's sort of surfacing and suggesting people, helping people to navigate issues of data bias. So that's sort of transparency. And there's also then explainability, which is another piece of that. Why is this decision here? And giving some ability to kind of see what happened. That's what we were talking about with that data canvas of being able to see, oh, here's how it got here. That's really important. We did some work with a medical software maker for radiologists that helps them translate their dictation into the medical reports that you give to the physician. Don't screw that up, right? Don't make stuff up there.

[Karen] Yeah. Oh my gosh. Do you watch the Pitt? Do you know the scene I'm talking about? Oh my gosh. Anyway, for those of you who are listening, you either know or you don't know. And I apologize if you don't. So continue.

[Josh] No, just all to say that, like, you know, so we designed some patterns where it was really easy for the doctors to verify where in their dictation this passage came from and vice versa. Where is this in my dictation reflected in the report? So it's just a lot of these things of just recognizing and embracing that AI is uneven. I mean, that's the thing that's hard about it. It is brilliant at some things and so stupid at others. And it's not obvious which is going to happen when. How do you design for that cushion and that prompting the skepticism?

[Karen] That's a great question. How do you design for that? How does a team, I think this was a question I actually had. How does a team keep in mind these moments? How do they evaluate whether an experience, an AI powered experience, is based on a good decision or whether the AI is making a good decision? How do you put these kind of guardrails in place? It feels like the more I think about this, and this happened when I was at this book, the more questions I had about how do we know? It's like the more I think about this, the more questions that come up for me. How do we know that these are good decisions that AI is making beyond just whether it's easy to use?

[Josh] Right. I mean, one thing is we've got a long history of unreliable actors and they're called human beings. I think that we have sort of this assumption. There's often this thing of like, oh, we'll put humans in the loop. And humans are fallible and suggestible as we've seen how they use and interpret AI. So not always great. There's a large number of-

[Karen] And humans can't always give you the breadcrumbs of their thought process either, whereas AI can.

[Josh] Right. Well, sometimes. Some of it is a black box, but they can at least say out loud what that process is. That's right. And so I think two things. One is let's remember that humans are pretty loopy and unpredictable themselves, which is one of the hard things about this because often we're letting the humans and the systems create a journey that we as designers didn't specifically create. You see this web chat is a great example, right? The interfaces we're talking about go well beyond text-based chat and text-based agents, but they are incredibly open-ended and can kind of go all kinds of places. And you see open AI and Anthropic constantly scrambling with really unfortunate outcomes because the people took the system someplace crazy and vice versa. So unpredictable on both sides. Well, so all right. So this is even a worse problem than you were talking about before. So I think one is we've got hospitals like the Pitt have all kinds of safeguards in place for how do we keep processes on the rail? How do we have checklists? Who's reviewing those things? And so that's part of it that goes into this is when there are things that cannot fail, should AI even be involved at all? Or should we go with process that we know? And if there is some acceptable risk, how are we going to mitigate that through processes that we've already put in place? Can AI or other agents help to look at things and crazy check? So a lot of things as we look at systems, agents taking care of things that you can have systems that are responsible for critiquing, reviewing, clarifying won't be perfect just like teams of humans aren't. And where do people play into all of this? But I think a big part of this is with any software, but it really comes to the fore with AI is looking at where is risk acceptable? Where is it unacceptable? And how do we map that across the possible benefit of automating this thing? We think in general that a lot of these sort of presentation things, which is why we're drawn to it. It's like, oh, AI is good at presentation and manner, bad at facts, but wow, actually, if we can adjust the manner, the UI, the interaction to the moment, that feels relatively safe, not entirely. So that's an area that's, oh, that's interesting. What can we do with that? And what new experiences does that let us create? We'll leave the data and the answers to more trusted traditional systems.

[Karen] There's so much here. And I keep thinking over at this brief, and I'm like, we're not even skimming the surface of the things that I really wanted to dig deep into. But I'm also looking at the clock and here we are, we're at like 41 minutes already. Didn’t I tell you? So I know that I'm pretty close to wanting to wrap this up because that's our promise to our listeners. But I really do need to get to what all of this means for insights professionals, because our audience, primarily insights analytics professionals, UX designers, UX researchers. And so I know that we've talked about how we're talking about this is not making that role obsolete. This is actually calling for more research and the need for a stronger hand in the process. But I'm also curious as to how the role will evolve, because it does signal a change. Everything that you were just talking about to me, and one of the things I wanted to talk about was responsible design and how to some degree, the role of the researcher or the role of the designer is kind of this, just like we have responsible AI. Now we're talking about responsible design, and we're talking about kind of spearheading how we're using AI in this context. It's not dissimilar to how we use AI in other contexts, right? But we're making sure that we are proper caregivers of the entirety of the system and the ecosystem that we are putting together. I don't know if that makes sense to you, but is that a part of it? What is the evolving role of UX researchers or insights professionals looking like now that you've put this book into play and done this thinking about intelligent interfaces? Veronika, you have a long future ahead of you. What do you see?

[Veronika] Josh, do you want to start?

[Josh] Yeah, sure. We talked a little bit about how important it is to understand the people in the domain, for sure. The thing that I was just talking about, about risk, super important too, is what risks are the users unwilling to do? What is the risk profile of this domain? We need those kinds of insights in order to make good decisions about what is acceptable, where are the thresholds, what are the existing processes to prevent problems? There's that whole on the responsibility side. How do we prevent problems? I think something that's unusual about working with AI is that we typically, historically, in user experience have designed for the happy path, right? We're designing a path and we're going to take people through it. That part is still important. What is the common route? What are the common cowpaths that we want to pave for people? But here we're saying, there's not one path anymore. There are a million paths potentially. We're designing for failure here or anticipating failure instead of designing for success. There's a whole range of things where we really need those insights, risk areas, what could go wrong covered. That sounds a little dreary. On the upside of all of this, there's also getting in touch with the fundamental outcomes that people, both users and business, are looking for. Pulling back from known solutions and really getting in touch with those fundamental needs and desires. Then saying, hey, we have a whole new technology that can meet users where they are through these intelligent interfaces. Being partners and thinking through what we've identified these desires. How can we make the technology, what are the opportunities to invent something new or elevate existing things? That is super exciting because now we're saying, you know what? This is not just background research, which is unfortunately, I think where a lot of researchers are pigeonholed. This is critical visioning work that we're doing here of what's next, of what could be a new business model or even disrupt an industry if you're going to be super Silicon Valley-y about it.

[Karen] Yeah. We'd love to talk about that. Anything to add, Veronika? We are so out of time here.

[Veronika] I'll be quick. In addition to that, I think there's this kind of new role, almost a promotion for the designer, for the researcher, for the insight person, which is really creative director of the product. If you're willing to really engage and get into that prompt and really communicate to your team and to the system itself, your ideas, you really have this great opportunity to step in and direct the creation.

[Josh] Documentation becomes instruction, and that's like it becomes a really active production role. Yeah.

[Karen] Yeah. Super cool. There is so much to this book and there is so much to this topic. We skimmed the surface here. I will say that we are fresh out of time, but I know that you have a gift for our audience. I know that I want to be able to leave you with those final words to plug this book. How can they get it? How can they get their hands on it and learn more?

[Josh] Great. You can get the book. It's published by Rosenfeld Media. You can get it at any of your favorite booksellers, but if you get it at Rosenfeld Media, we would invite you to use coupon code sentient-greenbook. That will get you a 20% discount at Rosenfeld Media through September 2026.

[Karen] That's great. We'll have a link to that in the show notes. Also, both Josh and Veronika, how can our audience find you all if they want to connect and learn more about what you're doing and your company and your other books, Josh?

[Josh] You can find us at bigmedium.com, where we publish a lot of things as well as have a newsletter. Veronika and I are both reluctantly on LinkedIn, and we would love to have you join our professional network.

[Karen] Oh my gosh. Don't say reluctantly. I love LinkedIn. It's one of my favorite playgrounds right now. It's just a way for us all to stay in touch because we can't see each other every day.

[Josh] It's so important. It's a love-hate relationship.

[Karen] It is. Oh, good, good, good. Friends, I'm just so glad that we talked. Actually, I can believe it. It's super predictable that we've talked as long as we have. Here it is, the top of the hour, and it really is time to fly. But what a pleasure it was talking to you both.

[Josh] Thanks so much, Karen.

[Veronika] Thank you so much. Thank you, Karen.

[Josh] We're happy to come back next week.

[Karen] Oh, what a pleasure. Next week, we are recording this before it's launched, and next week, I am happy to say I will be on vacation.

[Josh] Hey, enjoy it. Well deserved.

[Karen] You can come back, but I will not be here. I will be on a beach, which is very on-brand for me. Anyway, thank you both so very much.

[Josh] Thank you.

[Veronika] Thank you so much, Karen.

[Karen] You're welcome. You're welcome. In addition to thanking you, I need to thank our editor, Big Bad Audio. I know what you do for us, and I know the parts that you take out. Thank you for making us sound the way you make us sound every single episode. We appreciate you. To all of our listeners, thank you for tuning in time and time again. We will see you next time on the Greenbook Podcast. Have a great day. Bye-bye, all.