In the first episode of Patterns, Chris Strahl sits down with Dave Brown, design leader at Qualtrics, to explore what modern systems thinking looks like in an AI-driven product landscape. Moving beyond traditional notions of software design, the conversation reframes product creation as a shift from a single golden path toward a world where every experience is effectively an edge case. Together, they unpack why context, not features, is becoming the primary design material and how AI is forcing teams to rethink how systems are structured, constrained, and evolved.
Drawing on his experience leading AI and ML initiatives at AWS and now at Qualtrics, Dave explores how designers and builders can shape better outcomes by designing for context, learning loops, and adaptability. The discussion spans designing for AI versus designing with AI, the rise of compound engineering, and the collapse of rigid boundaries between design, product, and engineering. Rather than shipping static features, the future points toward systems that learn continuously, respond in real time, and improve through every interaction.
Key takeaways
- Context is the core design challenge of 2026, shaping how AI systems behave, adapt, and deliver value.
- Product systems are moving from golden paths to infinite edge cases, driven by personalization and real-time decision making.
- Designing for AI means creating learning loops, where systems improve through continuous feedback rather than static rules.
- Compound engineering reframes software creation around systems that get smarter over time, collapsing traditional role boundaries.
Guest
Dave Brown is a design leader at Qualtrics, where he focuses on AI initiatives and the evolution of the company’s design system. Previously, he spent nearly a decade at Amazon, including six years leading design for AI and machine learning services at AWS. His work centers on building adaptive, scalable product systems, with a particular interest in context, learning loops, and how teams can design systems that get smarter over time.
Transcript
Robin Cannon:
This podcast is brought to you by Knapsack, the Intelligent Product Engine, helping teams design, build, and deliver digital products at the pace of ideas, Knapsack creates a system of record built for both humans and AI, giving product teams the data structure and alignment they need to deliver with speed, scale, and confidence. Learn more at Knapsack.cloud.
Chris Strahl:
Hey, everyone. Welcome to the Patterns Podcast. Each episode we sit down with the leaders and builders defining how modern digital products come to life. From systems and tools to culture and decision making, we dig into what's driving real impact today and shaping the future of how teams build.
Hey, everybody. Welcome to the Patterns Podcast. If you're looking for the Design Systems Podcast, you happen to be in the right place. That's right. We've changed the name. We're talking a lot more about the way that people think about systems and product in general. And so we've decided to kind of move the podcast along past design systems and into this idea of how systems and product interact to make the things that we use every day. With that, we're going to be having our first guest, Dave Brown from Qualtrics. Dave, why don't you introduce yourself, say hey.
Dave Brown:
Super happy to be here. I'm one of our design leaders at Qualtrics. I'm focusing on our AI initiatives and our design system. I've been at Qualtrics for about six months, and before that came from Amazon. I was at Amazon for around nine years and led the AI and ML services at AWS for the last six.
Chris Strahl:
We had the opportunity to meet at a Patterns event in Seattle, and it was super great. One of our marketing folks, Kim, she used to work at Qualtrics and recommended that you end up being there. And so this is our second time getting to have sort of a formal sit and speak situation. And I'm really excited for this because at that event, I felt like you were saying the things that were, like, floating around in my brain in a way that was way better than I could say them. And so I've been looking forward to this conversation for a couple months. And actually, just so our listeners know, I'm still a little bit sick. I'm getting over the flu, and so I'm. I've got a little gravel to my voice today, but I was so excited to do this that I couldn't delay it any longer.
Chris Strahl:
And so with that buildup, why don't you tell us a little bit about what Qualtrics does?
Dave Brown:
Sure. Well, Qualtrics is really defined category, which is the experience management category. And so we have a number of types of software that helps different types of customer across different parts of the customer experience journey. So we have software for customer experience, for employee experience, and also around strategy and research. What probably a lot of your listeners know Qualtrics for is for surveys, because we send many surveys every year. But I think even beyond just a single survey, what Qualtrics is great at is helping brands and researchers build an omnichannel understanding of their customers and on their users. Right. So a survey is a specific channel and a specific moment in time.
Dave Brown:
But we also have connections into lots of other channels, from social listening to data for users and customers calling in through call centers. And Qualtrics is a platform that helps bring together a more unified understanding across all those different channels. And so a brand can rely on Qualtrics to build connection into listening across all those different channels and have a platform that brings together in a way that's really coherent and actionable for those brands.
Chris Strahl:
That's great. So you all have a bunch of different data points or moments that you collect data and information from a business's customers, and then you can use that data to understand more about how those customers are behaving, the experience that they're having, and then recommend or take action on that based on what that data is telling you.
Dave Brown:
Exactly. We kind of talk about this listen, understand and act journey, you know, where we can listen across all these different channels. And because we're integrated across all these different moments of an end customer journey, we have a much richer and deeper understanding of how customers experience frustration or delight across different moments of that customer journey. And so you could think about that applying across lots of different types of industries and verticals, from hospitality to travel, consumer experiences, and the way customers are interacting with those brands. Right. Ultimately, we have the opportunity to help brands better understand their customers. And then because we're at those moments of friction, really intercede and help resolve that friction. Right.
Dave Brown:
So it's not just about listening or understanding about what's going on, but more and more, how can we intercede and resolve customer complaints or frustrations and make more delightful experiences?
Chris Strahl:
I think that's a wonderful idea. The idea of a data point being somebody just like rage clicking on a buy button and it's grayed out for some reason. I love the idea of that interaction pattern getting solved for me. I actually had that the other day. I just bought a 3D printer that I've been kind of like going nuts with, and my inability to, like, send a print because of some configuration error was driving me crazy. And I was like, man, I bet Qualtrics could help with this.
Dave Brown:
Any single moment of customer frustration or friction. It doesn't live in isolation. You know, if you have a bad travel experience, you know, understanding where you've traveled before, maybe what tier of traveler you are for a specific airline, the way that a brand may intercede on behalf of that poor friction is going to change based off of the history and your context. Context is kind of the name of the game of 2025, as I know we're going to talk about AI, but there's so many things beyond that individual single interaction that you need in order to help resolve and resolve a frustration.
Chris Strahl:
This is getting to the meat of it. Right before the show, you said to me, like, hey, are we talking about design systems or are we talking about systems as a whole? And I promise you, we were talking about systems as a whole. I think that there is an intersection here with design systems, but a part also of rebranding the podcast around patterns has been the idea of trying to look at this like a lot more holistically. And so this is going to be a holistic conversation about how we think about systems and structures, for how we think about experiences across big companies. And a lot of those experiences are being driven by AI. One of the things that you brought up is how do brands respond or how do they make decisions based on actionable insights about their brand? And that brand response differs wildly depending on what type of consumer you are, what type of brand you're interacting with. And I think that this leads us to this really interesting idea of how do we design constraint models for this, these sorts of interactions and these sorts of experiences based on the platform, based on the interaction patterns, based on the context that you mentioned. Because I think this becomes this really interesting forefront for systems and for design and for how we think about the future of the experiences we make.
Chris Strahl:
And I'd love to just get a primer from you around when you think about what modern systems really mean to design, not design systems per se, but what are you thinking about? What are you talking about?
Dave Brown:
I think we're all in the midst of this transition from whether it's building software or building something that scales. Kind of thinking about it as there's a set of a few paths that you go down for designers. We thought about it as there's a single golden path and a few edge cases you might move through. And then the far end of that continuum, you can imagine this hyper personalized future. You know, that's very AI driven.
Chris Strahl:
Everyone has their own path.
Dave Brown:
Exactly right. And so this interesting part is, what is the journey? How do we get from here to there, from this place where things are a single golden path with a few edge cases, to this world where every experience is an edge case.
Chris Strahl:
I love that. It's funny, my co founder, Evan, he often talks about how he likes to think about design from the perspective of the miserable path. And that's his favorite path because it's intended to be more encompassing of those edge cases. But I think you're right. There's this idea now of the software we write. We write it for a handful of different use cases and a handful of different pathways that people can take to some outcome or some goal. And as those goals become many and more varied, you end up in a situation where those paths are expanding. And oftentimes for big customers at scale, they're expanding exponentially.
Chris Strahl:
And you throw AI and agents in the mix, and now all of a sudden you have all of these things that are third party that are also exponentially expanding. And to give you just one example of where this is the case, if you look at like what an owned experience is now for an online retailer, for example, unless you're one of the maybe top five or ten retailers that are out there, most of your owned experience is probably driving at best, parity, which is with something like an Amazon experience or something like a third party shopping experience. And so as our way of connecting and getting to our products or to solving the problems that we have as users gets more broad and more varied, there's a lot of things that are actionable, bits of data that we just can't make happen ourselves. And I'm kind of curious, when you look at that problem space of the construction of all of these different pathways, what do you see as a solution? What's your roadmap or your guidepost for this journey?
Dave Brown:
There's a technical angle into this as to how do we build these systems. And then there's the design or product angle of what's the outcome that we really want to achieve. You know, what kind of system do we want to build? If you look at the investment in AI, you see the investment going at the very lower levels of infrastructure, right? Everything from building the data centers to building the new chips and the hardware layer.
Chris Strahl:
The price of RAM is quintupled in like three months, so.
Dave Brown:
Exactly. And you can start working up that stack, the database layers, and how we're moving into vector systems and rag and you have the application Layer, how are you going to connect with different agents? I think at that top presentation layer, whether that's a UI or other modalities and other ways that you're interacting. The opportunity for designers and design systems is, you know, context is king. We will spend billions of dollars, maybe trillions of dollars building out this infrastructure so that we think that the model has the right context to intercede for that end user. But the other opportunity is, as a designer, how do you create the conditions so that the end user is providing the right context to that AI system? There's a really wide range between an empty input box that says put in your goal and a single output that that AI is going to give. This is the opportunity for designers to invent new patterns that help those systems be able to adapt in real time. They're personalized, they're resilient, you know, when things go wrong. But also design system builders and owners, what are the right reusable patterns or components so that we've got that intelligence coming into the system in a repeatable, reusable way.
Chris Strahl:
So we're talking about context. We're going to be saying that word a lot, and I wanted to make sure that we have a common understanding of what the definition of that really means.
Dave Brown:
You know, on the engineering side, we talk about context engineering, and that's setting up a rag system or the right system prompt, or your evals and your orchestration across different models. I think when I'm talking about context, it's about the end user providing the system with the relevant information based off of what's, whatever is appropriate for that use case or task or job to be done. Sometimes you want that context to include past purchase history or previous interactions, and sometimes you don't want that, you know, very intentionally. It's the right information for the given task at hand. Yeah.
Chris Strahl:
And you can't really talk about context without mentioning, like MCP and a bunch of the other protocol stuff that has happened in the AI landscape right now, where the whole idea is like, hey, I have some sort of model and I want to connect that model to a bunch of different sources of context. And those different sources of context represent a huge amount of things. There's context all over the place. Like you said, 2025 is maybe the year of context in the greater idea of AI. And so as these models suck up all this different context, one of the things that is emerging as a system's need is the ability to curate that context across a different archetype or a different dimension of what you're actually trying to use that context for, like you said, purchase history, if you're looking up what somebody is most likely to purchase next is probably really great. If you're trying to do it in terms of like, hey, I have a sign up for holiday discount, it probably matters less. And so the idea of how do you use that context in the right way is something I think is very emergent from a systems perspective.
Dave Brown:
Yeah. And often that context will exist in the system, you know, it will be connected to the right database or you've got a rag system set up for designers. The other opportunity is how do you, at that top UI layer, at that presentation layer, at the interaction layer, how do you nudge the user to be specific enough and relevant enough and accurate and precise enough to give the system the right context if it doesn't exist?
Chris Strahl:
You can think about like a generation success rate going from like the mid-40s to the mid-80s based on just one small piece of information. But that piece of information isn't known intrinsically. You have to, you know, evoke that from a user. And in that evocation, like how is it that you get that user to want to give you that data?
Dave Brown:
That's right. What's the value for them and how do you bring them along on that journey? On one end of that spectrum there might be the very classical way of here's a wizard and we're going to walk you through these steps. And on the far other end of this continuum there's this open input box. Like these chat experiences that we have where we just expect the user to put in the right information. The AI system is going to know that. I think those ends of the spectrum are pretty wide and we haven't yet built the right interaction patterns and connections between that presentation layer connecting to the underlying AI infrastructure. Right. And so this is the opportunity for designers to do that.
Dave Brown:
Yeah.
Chris Strahl:
And it's honestly hasn't evolved much. I mean there's definitely been evolution. Right. It's way better than like here's a multi line text field. But it is also still fairly rudimentary, I think in terms of the design patterns that exist in how we interface with models and LLMs. And I'm really curious to see what this looks like in five years from now where we've refined these interaction patterns a lot. I think all of this brings about a really interesting question though. We kind of talked about like what I see is two different things.
Chris Strahl:
There's the idea of design and AI, but when you think about that, that's divided into designing for AI and then also designing with AI. And I think it's worthwhile to sort of tease those things apart for a second. Those different models for design, how do you view them and how do you look at the differences system?
Dave Brown:
I think of designing for AI as a new material that we get to use. These are new paints and paint brushes. So we've got new patterns to build, new technology to interact with. And because of that, we have new experiences that we can build while we're doing that. At the same time, we're challenging ourselves to design with AI. And this is the tooling that helps us do it right, Whether this is Figma, make and Cursor and whatever other tool that you want to use. So you're really challenging the way that you build and ultimately the value that your discipline or your craft delivers to the organization. It's tough because these things do go hand in hand.
Dave Brown:
Or you could choose to focus on one versus the other in any given day.
Chris Strahl:
I was looking at a video that we'll throw in the show notes that Brad Frost and Dominic from Storybook put out last week. There was also something that TJ did as well, was kind of all about the idea of agentic design systems. And this notion that while design systems are still going to have this thing that you can browse and touch and see, that probably looks very different than the artifact that an AI would use in order to want to consume information as context from a design system. And there's this whole sort of idea out there that is this notion of, hey, look, our traditional ideas of artifacts just changing. And my opinion is very strongly. Yes, and that like, something that looks like context for an AI is very different than something a human would browse. But ultimately those sources can still be the same. And I think this is a bit of a petty thing, right? It kind of messes with people's minds a little bit when we say, like, your design system is made with Figma and it's made with Git and it's made with all these other different things, but you never actually really see it.
Chris Strahl:
It's just this, like, context that floats. And the background as the traditional idea of a design system or like any type of system is like, hey, I can touch this, see it, browse it, experience it. And the thing that I've been encouraging people to kind of try to like, wrap their head around with this is like, if you never had to visit your design system and you immediately knew everything that was in it, would you still want to Go to it. And that I think is this like interesting interaction pattern for how we think about designing for AI that really hits at that point. Like, designing for AI means letting go of some of these artifacts that we've held fairly sacred for a really long time.
Dave Brown:
If you were designing purely for AI, there's some of those steps that you might just skip altogether of going from vectors into code, for example. Right? You might want to go just straight into code. The designing for AI piece is interesting now because in most of these experiences we were designing for humans and we're designing for AI. What's unique about some of these design systems is there's gotta be a feedback loop down to the broader system, right? So I think a lot about designing AI learning loops where you're not just trying to capture information or maybe even provide an affordance or a nudge or something at the UI later, help a user go through a flow and complete that task. But ideally there's some broader AI learning loop that's happening as well. And this is really interesting area of design patterns where maybe what a component looks like isn't purely a front end thing or a UI thing anymore. Maybe it's connected to some of those rag system that's underneath or some other piece of your Agentix stack so that the more customers interact with that component, the smarter it gets, the more, you know, relevant that is. Right? So you could think of maybe the future of design systems is going to automatically like, of course it's going to include your recommendation component, of course it's going to include your summarization component, your recommended next best action component, whatever these things are.
Dave Brown:
You could absolutely imagine building that natively into your system. And so the value to the business is like we've got a way through the design system that's connected to the underlying AI infrastructure that makes the whole applications smarter over time.
Chris Strahl:
I try to think about it in three different buckets, if you will. I kind of look at it as like, hey, here's a bunch of first party data, right? It's the history of design at your company. It's every figma file, every git repo, every commit you've ever made to code design, maybe even every JIRA ticket you've ever entered. And so you have this wonderful history of like, what is your brand, how have your products grown, changed, evolved? And then you have the secondary side of it which is like, what is all the data that I know about what works and what doesn't. And this like sort of second party stuff is kind of like what you get from Qualtrics, right? Like, what are the things that are working, not working, what are the places of friction and frustration? What are the places of like, delight and amazement? And how do I go about examining those in user land and giving feedback on that? And then there's a bunch of third party stuff which is like, do I meet accessibility standards, Do I have a performance budgeting tool that allows me to understand within my performance budgets, et cetera? And the amalgamation of those things feels like something a lot more than just a design system. It feels like a product context engine. And that context engine that you can build behind it all of a sudden has a tremendous amount more power than I think that we've ever really thought about it before. But it also requires us to be thoughtful about where we use that different bits of context for various different tasks that we're going to go have an AI work on.
Chris Strahl:
I think that in this new pattern or new way of working, where you're talking about this idea of like, hey, we have all this need to control all these different things. I'm curious what you think a modern stack looks like, because I have a hard time seeing exactly what the tool set lines up to look like. And I think it maybe is very unlike how we work today.
Dave Brown:
Today, the way that we build software, it's largely has gone unchanged for a lot of decades. There's the opportunity through some tooling and better processes to maybe kind of collapse the speed at which some of these things work. Going from design into dev handoff, which we've been talking about forever. And we could talk about that. I think there's going to be a lot of new tools that help improve those workflows. There's going to be people who develop the skills so they could more naturally cover a broader range of those. Sort of beginning to shift code. And that's really interesting.
Dave Brown:
We could absolutely explore that. There's a ton going on there. I also think there's going to be a whole new set of roles such that the makeup of what a team looks like to build an AI system that's running in production, that is, you know, the goal isn't to launch a feature. The goal is to build something that really works for customers and works for users. Right. And so in the future it might be more of, you know, a lot of the way these systems are being deployed today is you've got some version of a forward deployed engineer that's really embedded with the customer and understanding what's working and how that system needs to be Tweaked or tuned, it might be that that role is a core part of how we build software in the future. There might be the builder who builds before we deploy and then the forward deployed engineer that's using that system and tweaking it.
Chris Strahl:
I love this. And this gets to like this really fun topic of is the traditional idea of engineering product and design just going away. And when I hear you talk about the way that we'll build in the future being a lot more problem driven instead of feature driven, that's really exciting because I do think that one of the problems with the tangled pathway that you were talking about that's sort of existed forever in the way we make product is, is that the people that are really close to the problem are very rarely the people that are shipping the software. And the ability to draw a straighter line between this is the problem we're trying to solve and this is the solution for it, I think is really exciting when we look at that. Yeah, of course, that's speed, that's faster to value, that's faster to market. There's all these like really good business ROI reasons. But I think you're also going to get better software. And the reason why you're going to get better software is to be able to best understand the problem, are going to be the ones that can create it.
Chris Strahl:
Because you're not going to be reliant upon this system of at least three like highly specialized expert roles where you're going to have the engineer that knows exactly how to code that thing, the designer that knows exactly how to express the brand or the perceptual patterns, and the product person that understands exactly to articulate that into a feature set. You're going to have this proxy for skill that is maybe not perfect but at the same time is going to let people get a very, very long way sort of on their own, maybe in this builder role.
Dave Brown:
A lot of building enterprise software today is a big game of telephone. You know, we've got a hypothesis around something that we've gotten from research and PM might write a spec around it and design will create some mocks and engineer will build it and we'll deploy it and test it. And I think it's a very lossy process. We're losing information at every stage of that handoff and the handoff itself can take months to launch a feature. And so if the person building it is closer to the end user and closer to their customer and they can iterate very rapidly. A you're going to have fewer of those handoff points and B that process is going to just collapse and move much more rapidly where you can iterate and change things by the day, by the week, rather than over months.
Chris Strahl:
Hey everyone. I'm taking a quick break to tell you about Knapsack's Pattern Summits. If you've never been. These gatherings are for senior pro, product design and engineering leaders navigating the complexity of modern digital production work. We bring folks together for thoughtful discussion based sessions where you can share what's top of mind, learn from peers and leave feeling renewed. Pattern Summits are invitation only and intentionally small so the conversations stay meaningful. If you'd like to join us, visit Knapsack.cloud/events to request an invitation.
Chris Strahl:
You introduced me to this concept that I think is really interesting to explore in this, this idea of like, what is compound engineering? And I hadn't heard about this before our last conversation and I read a ton about it and it seems really interesting because from what I understand and help me if I get this right or wrong, is when you think about the philosophy of compound engineering, the idea is, is that you're using AI instead of something that's making like one offs, you're using it as a learning system for every interaction that you have with it. And in that learning system, the idea is it's getting better and better at solving the problems that you're actually trying to go and deliver on instead of just trying to make that feature or make that concept.
Dave Brown:
I read about this from Every, if you ever read Every Newsletter and they got lots of great AI content. But this concept of compound engineering is that as the engineer is building the software, they're using agents to build the software. And so along the way, from every commit, and even after you've deployed something, every bug that happens, you're not trying to just fix that bug, you're trying to fix the underlying issue in a way that makes the system smarter. Right? So I think there's some cohort of engineers that just have accepted and I think they're right. The future of software engineering is not writing code line by line, but managing a system that writes code. And if that's the case, your job as the engineer is kind of the air traffic controller to help make the system smarter over time. And so when you get into compound engineering, the idea is as you're building some bug that you just fixed, the agent itself learns and we'll write the own tests to catch that next time. And the engineer is going to review that output and then feed that back into the system so the next loop gets better.
Dave Brown:
Right. So this is similar to how we were talking about AI learning loops and designing for those. As a designer, how do you think about building these learning loops in the product? Maybe a core area to compound engineering would be these designing for AI learning loops where the end users are feeding information back into that system that makes it smarter over time.
Chris Strahl:
So that's amazing. And like, this is why I think that there is such this upending that we were talking about of traditional roles and tool stacks. The mystery of it all. To me, that's fascinating is this idea of, like, what this is all going to look like, right? Because if I'm reading your mind or if I'm trying to like, walk down the hallways of it, when I look around, I see a bunch of people that are really, really skilled at making systems and really, really good at understanding the problems that create the most value for the consumers of the product they're creating. And that ability to have that taste or that understanding of value for that user ultimately becomes a proxy for the skill of traditional ideas of engineering, product and design. Design in such a way that you should be able to have a bunch of people working with agents to make software in ways that have these learning loops that you affiliated with them that continually improve the software that we're creating. So, like, version one of something comes out and it works and it solves a problem. But then as the system learns more and more about how that organization solves problems, it gets better at solving them and it's better at solving them earlier and faster than it has ever before.
Dave Brown:
Exactly. And then how does that feed the next iteration, the next evolution? And this is where that design and PM role is probably going to kind of collapse a bit, where it's not what's possible, but what do we want to happen? Do you want that system to remember that interaction? What's the default state? ChatGPT default state? Do you want it to remember my context from every interaction, or do you want it to not. But I toggle it on. Right. And how am I going to manage that across lots of different interactions and across different projects? And we haven't even gotten into how we're going to collaborate together.
Chris Strahl:
So that's also really fun. And I think that's a huge part of the tool stack that is like, we have some idea of how we want to think about collaboration. Not we at Knapsack, which is like the royal we of the industry. Right. But collaboration with AI is still really hard right now. And there's Nothing that's really done this solve at team scale. Well, you have cursor codecs, cloud code, et cetera. All these things are out there that are working frantically on how do we make an individual or a small team of developers more productive.
Chris Strahl:
But there's very little that takes into account how does the system become greater than the sum of its parts. And I think that's where a lot of this is really interesting because if you think about what people are specialized into right now, you have a very like skill based specialization. I think that we're probably going to reshuffle the deck into a problem based specialization where we're going to start to look at people of like, oh, that person is really good at solving card abandonment problems, or that person is really good at solving catalogs. And it's going to be a lot less about how do I actually know how to code this particular thing or design this particular thing so much as how do I best understand that need and the ability to solve it. And that shuffling to the deck is going to require collaboration at a scale that I don't think we've ever really attempted before in product. And I can't wait to see what software solutions people come up with for this.
Dave Brown:
Don't you think though, that kind of gets us back to where we were 15, 20 years ago in the industry? When we think of design, I think we've gotten a little too specialized and a little too precious about the specific boundary conditions of where we work within design. I worked at an agency where we talked about everyone was a thinker and a maker, and the idea was everyone's a thinker. You can concept, you can understand the problem, but we're all makers and we have different skill sets for how we make. Some might edit a video, some might do 2D animation or 3D graphics, or copywriting, but it was just the idea that everyone's a thinker and a maker and we've gotten a little too specialized. I think within design where, you know, there's UX designer and visual designer and interaction designer and you know, some of these other roles like information architect and things like that have gone by the wayside, largely speaking, they've kind of collapsed to become a skill. And you think of that, it's gone from a role to a skill. And I think what we have today as different roles are going to collapse to become skills. And the AI builder of the future is just someone who has lots and lots of different skills or knows how to work with the right tools when they don't have that deep expertise.
Chris Strahl:
I really like that take. I've heard so much froth from folks about exactly this topic, like are we going to be more specialized or more generalist in our approach to product? And I think that the answer is of course, kind of both or it depends. You know, that frustratingly vague thing where we are going to need a lot more specialization into how we think about our interactions with AI. But there is this idea of maybe there's a little less specialization in the role that we keep and maybe that also widens to encompass things that would have had very different job descriptions. The obvious thing is the collapse of the great wall between design and engineering. I think that you're going to have a lot of people that all of a sudden can make things in code very, very easily. They might not be able to make them perform at enterprise scale, but you likewise are going to have a bunch of people that could never express a brand properly before, be able to create really interesting and unique brand expressions. All this comes down to this idea of innately AI is enabling us to do a lot of things that we haven't been able to do before.
Chris Strahl:
And that's exciting, especially as we're able to explore a bunch of design patterns, interaction patterns that we've never had to think about.
Dave Brown:
So many different angles to take this, I think on the design to engineering handoff, really what we're talking about is design to front end engineering. In the industry that has already collapsed over the last 10 to 15 years. There are incredible front end engineers that I've worked with. I'd also say most companies, they've moved to the quote unquote full stack engineer model, which is often, you know, code for. We're going to prioritize someone that has maybe some skills around building highly performant and scalable systems. And because they can do that, they can also build the front end. What if the designer who has the vision of what that end experience should be, not just from the static design, but through the interaction, could actually go and fix all the changes and maintain the usability backlog and redirect the calories that are spent building those lists of usability issues into actually resolving them. That's a future that I'd like to be in where we have a lot more ownership and control over what that end experience is and then let's work with our engineering partners to build the highly performant and reliable system.
Dave Brown:
So that's one that you just can sort of squint and see and know that it's going to happen, what the exact tooling looks like and how long it's going to take to really scale out, we'll see. But I think that's one area where it's pretty obvious that those areas are going to collapse. There's some other areas of being a broader AI builder, soup to nuts, solo entrepreneur.
Chris Strahl:
I think it is interesting that you're watching all these people crank out billion dollar companies in their basement. We'll see how long that lasts. But I think that there is this interesting conceptual idea of like someone with a really great idea can go a lot further than they've ever been able to go before. And that's amazing. Much more than it is necessarily the valuation of the companies being in the land of startups. But one of the things that I also think is this sort of unexplored territory that we are going to see that is going to create like new emergent roles is agents are innately this multimodal experience. You have agents that are responsive to voice, they're responsive to context, they're responsive to visual environments. All these different things that are going to be a part of that interaction pattern that is as much off screen as it is on screen.
Chris Strahl:
And when we think about the idea of designing for these experiences, it's going to really push us to expand our tool set to include a lot more of the gamut of perception than we've really had to do it before. I'd be curious your thoughts on that specifically as it relates to what are the emergent interaction patterns that you see with AI.
Dave Brown:
Yeah, I think some of the more interesting startups and companies in this space, they don't think about it as building for a specific interaction or specific surface alone. There's multiple examples of companies kind of building an agent, which is the idea you build at once. And you can deploy this across all these different types of channel and it can maybe adapt in real time if a customer wants it to be a more traditional chat based experience. Or it's got a voice mode but under the hood. The way that you would think about that is there's a core system that's driving that and then there's a set of interaction patterns that are specific to that modality. And you could even break that down further where you have interaction patterns to receive information, whether that's reading text or hearing audio, and then inputting that again, clicking, tapping, speaking, typing, motion. So I think you can deconstruct it. But what's fun when you have multimodal models that are powering this and you have systems that are driven by inference.
Dave Brown:
You can build experiences that can rapidly move across those different interaction types and modalities and just offers a whole level of personalization and adaptability that doesn't exist today.
Chris Strahl:
Yeah. And when we think about that idea of personalization and that idea about like all these different experiences and all these different types of experiences that are around, I think that there's also this really incredible thing of we get at some level as users the ability to choose our own context for how we want to consume or interact. And this idea of hyper personalization is really interesting. It's been widely talked about in AI circles and beyond. This idea about like, hey look, you know, I have my pathway, you have your pathway. And those pathways don't overlap very much. Whether that's because of our individual experience in a particular domain or particular piece of software, whether it's because of the outcome or goal that we're trying to achieve, whether it's because I have a different sensory set than you do, like maybe I'm vision impaired or something like that. The idea of how you can construct these individualistic pathways, or at the very least very segmented pathways through an application to user value has been this really hot topic.
Chris Strahl:
And I think that it's for the first time becoming like reasonably feasible. I want to be clear, like we're not inventing something wholly new here, right? Like having dark mode is kind of like a version of this. Right. And we all know how much it costs to implement dark mode at scale across a big enterprise application ecosystem. And so we're not talking about anything we couldn't do today, but we are talking about being able to do it on a scale that we've never been able to do before. And I think that's a really important distinction is that while there may be new ways of doing this kind of hyper personalized context, I think that what we're really talking about is the ability to finally use the tooling that are in front of us to be able to actually make an experience more unique and personalized than we've ever been able to do it before for.
Dave Brown:
And we'll be able to do it at this kind of micro interaction level. I don't know if you're using any kind of text to speech tools in your day to day, but I write.
Chris Strahl:
My email with my voice now and I love it. Yeah, never going back.
Dave Brown:
This is what I hear from everybody, right? Like I use whisper flow on little personal projects that I'm building in cursor and I would say 90% of the time I'm just talking to it, you know, and then when you have to jump back into typing because maybe you're around people or for whatever reason, it feels a hundred times slower. Right. And it's this. I'm never going to go back. But that's at the creation level. I also think at the consumption level. It's not hard to imagine that we're going to be able to do that with all the experiences that we have today, from the point of input to the way that you receive information in a single interaction, move between speech and clicking and typing and dragging and manipulating all that information. A lot of these agents, they're the canonical examples, are very operational type of tasks.
Dave Brown:
What's the most canonical example we've heard over the last couple of years of what an agent is going to do?
Chris Strahl:
I mean, book a flight, it's going.
Dave Brown:
To book my trip, right? Well, what is that actually doing to do that? Well, it probably needs to know that you like Avis, you're a Hilton Honors member, that you fly Delta, and you have this tier status. It knows which hotel room you usually book and what seat you like in the flight. Like, so what do you imagine that the user is going to provide all that information into this AI assistant back and forth, or that this AI assistant, this agent, is connected to all those different operational systems already and it's going to present that back to you as the customer and you'll confirm it to me. It's much more expected that that would be the latter. And so you can have this operational system, but that alone isn't going to lead to a great experience. Chris, what were you saying? It was like the difference between flavor and nutrition. Flavor and nutrition, right?
Chris Strahl:
Yeah. Like you can eat paste, but you know it tastes way better if you're having mashed potatoes with gravy.
Dave Brown:
The operational system will know that the last time I flew, I was put in this hotel room or I had this flight or I had this rental car. But what it really needs to know is, did I enjoy that experience? Was I frustrated? Would I be willing to have a later check in to get a different room rather than an earlier check in to get a room at a lower level where it was a little bit noisy. And so what you really need is you need this marriage between that operational data and that experience data. And it may be that that is going to lead to a much better outcome than the operational data alone.
Chris Strahl:
My point on this was I think that if you are looking at just the operational data, you're going to get a lot of very efficient Spartan systems. And those highly spartan systems are going to be based around the idea of, like, what's the quickest route to value? Because that's the thing that you're able to most easily measure. And because of that, like, a lot of your goal metrics for an agent are going to be around speed or efficiency. I think that if you have an experience index that is tied to it, that experience index is going to be something that is going to give that extra coloring to what the outcomes are going to be. And that experience waiting is something that is really interesting because I was saying, like, maybe this is a pipe dream, but there is the potential to actually be able to understand the value of specific experiences. And the opportunity there is really cool, because if you think about it, I could say, all right, I have some amount of value I place on my comfort. I have some amount of value placed on my time. I have some amount of value placed on my frustration.
Chris Strahl:
And those are all very different things to me. Sometimes they're interrelated, sometimes they're not. It sucks to wait in line. There is nothing more that I hate than waiting in line. And so if I can figure out a way to pay more, because I value that ability to not wait in line, if I can change a plan to not wait in line. But I'll only do that to a point, right at some point, waiting in line, either the cost benefit isn't there, or there's an efficiency aspect to it that changes. But this whole idea of, like, that's a shifting goal, and that's a very experiential goal. Maybe waiting in line isn't the best example in the world, but it's very similar, in my opinion, to the idea of, like, when a user complains about performance and the notion that how you solve most complaints about performance that aren't like egregious complaints about performance, performance is you add a spinner.
Chris Strahl:
The perceived performance is actually as valuable as the real performance itself.
Dave Brown:
Earlier, we were talking about, you know, the role of the designers to help provide the right affordances and nudges so that that end user provides the right context into the system. I actually don't want this operational AI agent to make decisions around which flight I want or which hotel room I want or which rental car I want, because I might be taking a business trip versus traveling with my family. What I really want is I want that AI system to understand the context, either implicitly or explicitly, through the way that I'm interacting with it and once you do that, it allows you to move from types of experiences that aren't just around planning something that's about to happen or looking back on an experience that did happen to inform the next trip. And you actually can start to move into the moment where you're mid stay on that trip and you can provide some feedback into that system. And these are some of the things that we're working on at Qualtrics right now around experience agents where you can take a survey mid stay, we'll pulse check, survey and we can connect that into other systems. I might say, hey, I'm loving my stay. Great hotel room, room, great stay here. But the air conditioner is a little noisy right behind the scenes.
Dave Brown:
We can take that information and route it and create a ticket into the maintenance team and so the maintenance team can come up. Or I might say, hey, I'm having a great time. You know, I didn't have a chance to talk to the concierge, but really wish I could find a great Thai food place nearby. Right? And you could have a human concierge reach out to you. Right? And so moving from these kind of post types of experiences to end the moment where you can not just measure what happened or try to even predict what might happen the next time, but to intercede and change that experience real time, that's really, to me, a good use of what it would be to be personalized and adaptive in real time.
Chris Strahl:
As the chronic cruise director of my friend and family group, by the way, I don't actually go on cruises is not my thing, but like I'm constantly the person that's got like the big like multitab spreadsheet three months before we go on a big vacation that has everything like all mapped out and stuff like that. I would love an AI that could help me create that based on my feedback in the moment. Because I always find myself with a bunch of stress and anxiety, which are also experiential things I would love to avoid. Largely because I want to make sure the people around me are having fun and the ability for something to help me with that or to take that off my plate. Sounds wonderful.
Dave Brown:
And we've got to solve this because the other version of this that we've all lived is kind of the classic machine learning version of this. And you get this through, you know, I've got my Spotify account and Dang it, my 10 year old daughter grabbed the phone, took away into the other room and now my playlist recommendation is broken, right?
Chris Strahl:
Dude, all I have is K Pop Demon Hunters. That's all I have.
Dave Brown:
Exactly.
Chris Strahl:
With two children, if I click my playlist, all I listen to is K Pop now.
Dave Brown:
Or my Netflix recommendation feed is broken or my Amazon shopping recommendation list is broken. Right. It's not personalized. And so for designing for those experience, when I'm talking about how do you make something that's really resilient under stress, when do you want that agent to remember for the next interaction and when do you not? And how do you manage that across a whole different set of agents? Is there going to be one uber agent that has to then coordinate your individual preferences across all of them? And what does that look like?
Chris Strahl:
I mean, do we each get our own. Right. That all of a sudden becomes our mirror image of ourselves? There's all these sorts of interesting ideas about, like, identity and stuff that play into that too, that are challenging because to have of omniscient access to all the places where my travel stuff is stored, like, I mean, what do I give an agent my 1Password account and just be like, go nuts? It's really interesting. There's a lot at stake around how we think about how we make these experiences come to life. There's a lot of excitement and enthusiasm about it. There's still so many unanswered questions that we have, in my opinion, an increasing amount of pressure to answer quickly.
Dave Brown:
And this is where I think there's just going to be a whole new set of roles and types of things that we're solving for that we haven't even begun to experience. Right. And so, yeah, maybe the way that I work today is going to change. You know what? Maybe that's okay. I didn't come into this field because I wanted to do the same thing every day for my whole career. I'm not so precious about the way that I work or the individual thing that I work on. I want to learn and grow. And so I think this is a really interesting moment to kind of come back to the root of why we came designers, why we became builders to experiment, try new things and, you know, enjoy the ride.
Chris Strahl:
This has been so wonderful. Thank you so much for sharing your knowledge, your wisdom, your insights. I've really enjoyed the conversation. I just want to say I really appreciate you taking the time to be on the program. I hope we didn't make it too design systemy for you. And just thank you so much. This has been really wonderful.
Dave Brown:
Thanks so much.
Chris Strahl:
This has been the Patterns podcast, everybody. I'm your host, Chris Strahl. Have a great day. Thanks for listening to the Patterns Podcast. If you joined us from the Design System Podcast, we're glad you made the move with us. You can connect with us on LinkedIn using the link in the show notes. The Patterns Podcast is brought to you by Knapsack, the Intelligent Product Engine, helping teams design, build and deliver digital products at the pace of ideas. Learn more at Knapsack.cloud.
