Imagine sketching a blueprint for a house and before you even finish drawing the front door. The plumbing is already being tested. Exactly. It's testing the plumbing, ordering the bricks, grading the soil. Welcome to this deep dive. You, our listener, have handed us an absolute mountain of phenomenal sources today. Yeah. And what they prove is that the traditional, safe, you know, sequential assembly line of product design is just completely dead. Totally dead. Like the old way where you sketch something, hand it off, and just kind of hope for the best. Right. It is a fundamental collapse of those old boundaries. We're looking at a total re-engineering of the workflow, which, I mean, is frankly terrifying for some people. Oh, absolutely terrifying. But it's also incredibly empowering if you actually understand the mechanics of what's happening here. And that is our exact mission for you today. We're looking at these extensive study guides from the Nielsen-Norman group, the NNG, on designing AI products. Plus, we've got those dense, highly technical research reports on tools like Claude Code and the Model Context Protocol. Right. And this rising concept of the new quote-unquote 10x designer. But our real anchor today, the specific lens we are using to make sense of all this, is a manifesto from Chris Mullins. Yeah. He's a principal product designer who is actively using these new techniques out in the wild. So our goal is to cut right through the marketing hype. We want to look at exactly how AI acts as a force multiplier and the real, messy trenches of product design. Which is honestly such a refreshing perspective. I mean, it is so easy to get caught up in the theoretical promises of artificial intelligence. Tell me about it. But seeing how it actually changes the day-to-day reality of someone whose job is to actually ship working software, that's where the real value lies for anyone trying to navigate this space. Okay. So let's unpack this because usually we hear about AI in one of two very extreme ways. Yeah. It's either a magical cure-all. Right. Like it's going to do everyone's job while we just sit back and sit margaritas. Or on the flip side, it's completely dismissed as this useless hallucinating gimmick. But Mullins' perspective shows it's really neither of those things. No. For him, it's an embedded thinking partner. It completely changes the entire workflow all the way from that terrifying blank page to the final shipped code. And you know that workflow starts long before a single pixel is placed on a screen. Right. The discovery phase. Exactly. Before a designer can even begin to build a solution, they have to figure out what the actual problem is. This is where AI's role fundamentally shifts from a creator to an analyzer. Using AI in this early-stage discovery phase honestly reminds me of having a digital detective's corkboard. Oh, like the ones from the movies with the red string. Yes. With all the photos and the red string connecting the clues. Mullins talks about feeding the AI, his raw interview notes, early hypotheses, and massive amounts of dense domain research. We're talking complex healthcare regulations or heavy supply chain data. Right. And the AI acts like that red string. It instantly synthesizes patterns from raw messy data to help him compress discovery. He's using it to move rapidly from a very vague, you know, hey, we think this might be a problem to a sharply defined, testable problem space. Especially in those incubation environments he works in. Yeah, those fast-paced, high-pressure, startup-like teams within larger companies. They are under so much pressure to just build something, anything, to show progress. Let's get something out the door quickly. Exactly. And AI synthesis highlights the logical gaps in their understanding that those fast-moving teams might otherwise just breeze right past. Wow, so it actually slows them down in a good way. Yes. What's fascinating here is how this practical application directly answers a major warning from the NNG sources. Back in 2021, well before the current AI explosion, Jacob Nielsen explicitly warned against treating AI as a hammer looking for a nail. Which is the ultimate trap of any new technology. You see it everywhere. The company realizes they need an AI strategy, so they lazily slap a conversational chat bot onto a website where it absolutely doesn't belong. Like a water utility bill payment portal? Exactly. It doesn't actually help the user pay their bill faster, it just gets in the way. That is the perfect example of a useless hammer. The NNG research is blunt about this. Implementations created just for novelty rarely produce actual value. But Mullins avoids this trap by using AI to dig deeper into the nail, to deeply understand the user's friction points, rather than just swinging the hammer to see what breaks. Exactly right. Let me push back on this a little bit though. If we are outsourcing our deep analysis to an AI, aren't we just risking hallucinated insights? How do you mean? Like if I'm a designer and I just have an AI summarize 50 hours of user interviews, aren't I losing touch with the real human pain points? Doesn't the designer lose their empathy if they aren't the one agonizing over the transcripts? That risk is exactly why Mullins and the research sources emphasize strict methodologies. The rebuttal to that fear is understanding that AI isn't making the final call on the product direction. It doesn't own the final decision. Right. The research explicitly points out a harsh reality here. AI doesn't control the customer's purse strings, real people do. Real people decide to buy or use the product. So the AI is just synthesizing the data faster so the human can make an informed decision. Precisely. And the sources point to a specific framework for this called the CARE framework for prompting. CARE stands for context, ask, rules and examples. So the designer is setting incredibly strict boundaries for the AI to follow. Yes. You aren't asking the AI what should we build. You are saying here are 50 transcripts based only on this text, identify where users expressed frustration with the checkout process and provide direct quotes. Forcing it to cite its sources basically. Exactly. The designer acts as a strategist. The NNG refers to AI in this context as a cybernetic teammate. It does the heavy lifting of sorting the data so the human can focus on preserving empathy. A cybernetic teammate. I love that framing. So once the problem is actually sharply defined, once that detectives corkboard has given us a clear target, the next step is ideation. Building potential solutions. Right. And this brings up another massive shift. Traditionally, a designer might sketch out maybe three different mock-ups for a new feature, mostly because drawing them by hand takes days. Three variations is standard. If they even have the time for that, often they just go with their first good idea. Yeah. But Mullins points out that AI allows designers to explore dozens of directions instantly. He calls it scaling divergence without losing structure. Instead of just generating pretty pictures, he uses it to generate complex user flows and to aggressively stress test those concepts against weird edge cases. And real-world constraints, which makes sense when you look at the specific types of products Mullins is building. Right. Incredibly complex systems. He mentions working on AI voice agents, like one named Lorena, and intricate care coordination tools for healthcare. And for those types of products, the surface-level UI, the buttons on the screen, that is only a tiny fraction of the actual user experience. Because a voice agent doesn't even have a traditional screen most of the time. What does it matter what color the microphone icon is if the agent doesn't know how to handle a patient stuttering? You've hit the nail on the head. What happens if a patient coughs while Lorena is asking for their date of birth? A static design file can't design for a cough. Exactly, but Mullins uses AI to map out that exact conversational failure state, instantly generating the logic for how Lorena should politely ask the user to repeat themselves. Without sounding like a broken robot. Right. AI is helping him think through the underlying system behavior. How does the system respond when the user says something unexpected? How does it handle a dropped internet connection? So it's not about the screens. It's about the logic between the screens. Yes. And this ties into a massive concept from the Deep Research Reports called experience engineering. Experience engineering. Let's dig into that because it sounds like corporate jargon, but the mechanism behind it is really fascinating. It is a major paradigm shift. Product design is moving away from just producing static artifacts, like a beautifully rendered screen in a Figma file. Or a written specification document. Exactly. Instead, the focus is shifting to designing measurable end-to-end systems. It's about how the entire experience operates across time, across different channels, and under different constraints. To make that concrete for you listening, think about a failure of experience engineering. Imagine an airline app. A traditional designer makes it look beautiful. Stunning typography. Great layout. Exactly. Yeah. The boarding pass looks great. But then you are at the gate, your cell service drops. And because the app requires an active internet connection to load the boarding pass, the app crashes. And you can't board your flight. Right. That is a failure of experience engineering. A static screen mock-up looks perfect, but the system failed under a real-world constraint. And AI helps map out those offline failure states before a single pixel is even drawn. Which means if you are listening to this and you manage a product team or you're an entrepreneur trying to build something new, your definition of done has to completely change. Done is no longer handing off a pretty design file to the engineering team. Done is about measurable impact under those real-world constraints. Here's where it gets really interesting. Because you have the logic, you have the system mapped out, but eventually someone has to actually build the thing. The code has to be written. And this barrier, that massive historically frustrating wall between design and engineering, is completely collapsing. It really is. Mullins explains that he doesn't just stop at wireframes or logic maps. He uses AI to generate working, interactive front-end prototypes. He's entirely bypassing the static mock phase. Completely. We're talking actual HTML and Tailwind CSS, which is just a wildly popular way to write code that styles web pages quickly. He is having the AI write the component structures and the interaction logic. He is building real interactive things that he can instantly put in front of users and engineers to test. If you want to see if a button feels right when you click it, you don't show a picture of a button. You build a button. Yes. This kills the ambiguity of a static picture. And the tooling that makes this possible is undergoing an absolute revolution right now. The research reports dive deep into a tool called Claude Code. And we need to be very clear about what this is. This is not just a slightly smarter autocomplete that suggests the next line of code. No. Claude Code is defined as an agentic coding tool. Agentic, meaning it has agency. Yeah. It operates independently to take actions. It does. It can independently read your entire code base, edit multiple files at once, run commands in your terminal, and even run your software tests to see if the code works. Teams aren't using it just to type faster. They are delegating entire chunks of the software making process to it. But if the AI is writing code, there has to be a bridge between the visual design and the actual terminal. If it's just looking at code, how does it understand the designer's visual intent? Right, which brings us to a concept in the sources called MCP, or the Model Context Protocol. Anthropic, the company behind Claude coined this phrase and I think is a perfect analogy. They call MCP a USB-C cable for AI apps. That analogy does a lot of heavy lifting because it perfectly explains the mechanism. A USB-C cable doesn't care if it's plugged into a camera, a monitor, or a hard drive. It just standardizes how the data flows between two devices. Exactly. MCP standardizes how different applications can feed their unique proprietary context directly into an AI's brain. So how does that actually look in a design tool like Figma? Well, if we connect this to the bigger picture, the impact of this USB-C cable for design is staggering. Figma is the industry standard where designers build their component libraries and color variables. And Figma has built an MCP server. Right. This means they can take all of that incredibly specific customized design context and pipe it directly into an AI coding agent like Claude code. So instead of the AI looking at a picture of a button and guessing it's light blue, it reads what's called a design token. A design token is essentially a line of code that acts as the company's single source of truth for, say, brand blue. And because the AI and the engineers are reading from the exact same dictionary, the final product perfectly matches the design. That completely eliminates the old screenshot guessing game. The AI agent actually reads those underlying design tokens. It knows the exact text codes, the exact padding required between text boxes. So when a designer like Mullins uses AI to build a prototype, the AI is generating code that inherently respects the company's design system. It scales perfectly, and it drastically reduces UI drift. That frustrating phenomenon where the coded product slowly drifts away from the original beautiful design over months of updates. And the efficiencies we are seeing from this kind of agentic tooling are wild. The research reports highlight some incredible case studies, like Spotify, for example. They are using these agents for fleet-wide code-based migrations. They have agents generating and successfully merging hundreds of pull requests every single month. The AI is proposing the change, testing it, and pushing it live. Hundreds of times a month. And RAM, the financial platform, reported dropping their incident triage time the time it takes to figure out why a piece of software broke by up to 80 percent. It's not just making the typing faster. It's parallelizing the work. The AI is doing things simultaneously. At a scale, humans simply cannot match. But there's a harsh reality to these massive efficiency gains. If Spotify's agents are merging hundreds of code changes a month and tools like Clogcode are bypassing the static design phase entirely, what is the designer actually getting paid to do? Are they just sitting around writing text prompts all day? Right. That is the existential fear for a lot of people in the industry right now. It feels like the ground is shrinking beneath their feet. But Chris Mullins has a very clear philosophy on this. He says explicitly that AI is a force multiplier and a thinking partner. It is absolutely not a replacement for design judgment. He talks about using it heavily in product definition, generating product requirement documents, structuring features, and aligning stakeholders. His key point is that AI acts as a counterbalance. In those incubation environments where teams default to urgency and just want to build something quickly based on their gut, he uses the AI's synthesize evidence to ground decisions in actual user needs. Which perfectly aligns with the research reports definition of the new 10x designer. For decades, the mythical 10x designer was thought of as the lone genius in the corner with their headphones on. Someone who just drew 10 times as many screens. Exactly. But I have to play devil's advocate here. Calling someone a 10x designer just because they build systems sounds great in a research paper, but doesn't that just turn designers into systems administrators? It's a valid concern. Right. If they aren't agonizing over the exact curvature of a button, aren't we losing the actual craft of making things beautiful? That is a very common pushback, but the data suggests a different reality. A 10x designer isn't abandoning craft. They are applying their craft to the leverage rather than the artifact. Applying it to the leverage. They are the ones building the reusable systems. They set up the token pipelines we just talked about, ensuring that brand blue is mathematically perfect everywhere it appears. They create the standardized prompts for the team to use. They collapse the handoffs. Their impact is multiplicative because they elevate the productivity and the aesthetic baseline of the entire organization, not just their own individual output. Okay. I see where you're going. But if the AI is doing the heavy lifting of drawing the screens and writing the code, how do you even measure a designer's productivity now? We clearly can't just count the number of screens they designed this week. We need entirely new key performance indicators. The sources suggest we have to look at metrics that measure systemic impact, things like lead time for change. Which is how fast a brilliant idea moves from the initial request to actually being a live feature in a user's hands. Right. We look at PR throughput and importantly rework rate. How often do we have to go back and fix a design because it failed in development? But as the metrics change, the risks change too because AI isn't perfect. If we give it agency, it has the agency to make really confident mistakes. That's far from perfect. The research points out massive new risks that require human oversight. AI can produce what they call hallucinated UX. Let's explain what that actually looks like mechanically. Mechanically, hallucinated UX is when an AI generates an interface that looks visually coherent. Maybe it perfectly matches your design tokens, but it behaves in a way that is logically impossible or inconsistent. For instance, it might generate a beautiful submit button on a form, but forget to wire up the logic that actually saves the user's data. Or it creates a navigation menu that leads to dead ends. And the NNG also points out that generative AI often violates core webwriting principles. It tends to be overly wordy, excessively polite, or it uses dense jargon instead of plain language. Oh, and we have to mention the sparkles icon. It is everywhere right now. Yes. The sparkles are taking over everything. The NNG guide specifically calls out the proliferation of the sparkles icon. It has become the universal symbol in software for AIs doing something here. But it is inherently ambiguous. Users often have absolutely no idea what clicking the sparkles will actually do. Will rewrite my email? Will it generate an image? Will it delete my history? It's lazy design. It's a failure to communicate function. It is entirely lazy. And all of these risks, the hallucinations, the bad copy, the ambiguous UI sparkles, they mean the designer has to undergo a fundamental role shift. They have to pivot from being primarily a creator of artifacts to an auditor and a strategist. So what does this all mean for you listening? It means the real value of your work in the future isn't going to be your ability to push pixels around a screen faster than anyone else. Your ultimate value will be your taste, your discernment. Yes. It will be your ability to look at 30 interactive prototypes the AI generated, instantly identify which one actually solves the human problem without hallucinating a broken menu and then enforce strict constraints. Like accessibility standards and safety guardrails. To ensure the final product doesn't just work, but works responsibly for everyone. Exactly. The human is the ultimate arbiter of quality. The AI can generate a thousand variations, but it cannot feel the frustration of a confusing checkout process. To wrap this all up, integrating AI into product design as Chris Mullins so brilliantly proves from his experience in the trenches isn't about replacing human empathy. Or turning designers into prompt writing robots. It is about utilizing a brilliant, high-speed, inexhaustible intern. An intern that can clear away the massive piles of busy work, synthesize dense research, and write repetitive code. So that you can actually focus your energy on building systems that matter to real users. This raises an important question though. One that builds on everything we've discussed about AI generating code and perfectly applying design tokens. Okay, let's hear it. If these AI systems and model context protocols eventually standardize all visual interfaces, if every app inevitably converges into perfectly optimized, mathematically proven, highly accessible layouts because the AI just knows the best way to display a button, does the future of competitive product design shift entirely away from how a product looks? Oh, wow. Does the battleground become entirely about how it behaves and its ability to anticipate your needs before you even click? That is a fascinating thought to leave on. If the aesthetics become a mathematically solved problem, then the behavior, the underlying empathy of the system becomes the entire product. Thank you so much to you, our listener, for providing such a phenomenal stack of sources today. Keep exploring, keep questioning, and remember, just because the blueprint can draw itself and test its own plumbing doesn't mean it knows where you actually want to live. We'll catch you on the next deep dive.