Nik (00:05)
Welcome to AI and Design, where we explore how artificial intelligence is reshaping the world of design. I'm Nik Martelaro.

Dan (00:11)
And I'm Dan Saffer and we're faculty at Carnegie Mellon's Human Computer Interaction Institute. Each week, we break down the latest AI developments, dive deep into topics that matter to designers, and talk with fascinating guests who are right at the intersection of these fields.

Nik (00:28)
Whether you're a designer working with AI or an AI practitioner interested in design, we're glad you're here. On today's episode, we'll be discussing Google released a major update to Stitch, its AI augmented infinite canvas design tool, and Figma's stock price plummets in response. Is vibe designing now a thing?

Dan (00:45)
Design influencer, John Maeda dropped his 2026 Design in Tech report at South by Southwest.

He offers six skills for designers making the transition to what he calls AX.

Nik (01:01)
And finally, MC Dean has published dozens of design skills for AI agents, especially in the area of accessibility.

Dan (01:08)
let's start with the biggest news of this week, which was Google Stitch. I got to say, I've been looking forward to and dreading this podcast. Even though I didn't know when it was going to come, I knew it was going to come. And now we're here. Nik some designers are a bit worried. Some are excited. What's going on with this?

Nik (01:28)
Yeah, so Google dropped Stitch this is their tool that allowed you to create apps and websites that they've updated now to have an infinite canvas layout, la Figma and other infinite canvas design tools. And they dropped a ton of new features that I think speak to a lot of what designers like about working in a highly visual

open space tools. So one, you've got the infinite canvas. Two, you can have lots of design options and alternatives. Actually, it will start to create them as you start to prompt and ask for different things. It'll create different versions and variations of a design system and then show you what the app might look like. It actually creates a design system. it shows, say, here's what your fonts are going to be. Here's your primary, secondary, tertiary colors.

Here's your heading sizes. So it's actually putting together all the kinds of things that you would expect to see on say your Figma project that you can then use. And in addition, it adds some cool features that are quite interesting. Like for example, it has a voice-based mode where you basically, truly just talk to the AI agent and it starts adding things to the canvas and

updating parts of it. I can see why the world responded the way it may have because it is doing what Figma actually said is probably very important for designers that you want to have a space that is expansive. You want to have a space where you can see the overview. You can think about the overall system and you're not just working in code within a representation of

the functional app where you have to click through everything to actually see all the screens. And again, this is something that I think designers want to work this way.

Now, yeah, I tried it. And so I think that a lot of people are, talking about how amazing it is. I tried it and I thought it was, it was cool. And it is really nice to be able to say, hey, come up with different options. The voice mode is pretty nifty. I'm not sure that if entirely that it kills Figma. Actually, if anything, right, I think this is sort of Google's answer to

Dan (03:10)
Have you tried? Did you try it?

Nik (03:34)
and other AI agents having access to the Figma MCP and then being able to generate Figma, I think that if you're working with those two tools, you could also do many of the same things. And so, yeah, and when I tried it, I mean, I actually, what I had to do was I had to try to come up with an app where people could explore our podcast episodes, but specifically like the ideas from the episodes rather than just a list of our podcasts like we have on our website.

Yeah, and I don't know it just it just had mostly stuff that looked like templatized versions

It wasn't bad. will say it really didn't understand the design sense. I had a hard time getting it to understand kind of the design style and sense that I wanted, like the color palette. I just couldn't get working well for me. It kind of put in its own what it thought. And that's probably a lot to do with the fact that it has an opinionated default, which is the design.md file. Now that is a file that has the design rules, a la impeccable and other design skills.

Dan (04:18)
Mm-hmm.

Nik (04:33)
But it's built in and you have to edit it to get it to update to be in the flavor of style that you want to have. And I'll say I haven't gone and updated that or had a conversation to try to update it.

Dan (04:45)
Yeah, I tried it too. And some of the same things as you, I was like frustrated that I couldn't get it to do the things that I wanted it to do. maybe because visually there wasn't all the tools that I was expecting or needing and trying to describe kind of what I wanted. I was felt myself really stumbling and could never get to kind of where I wanted to be. It felt very like blunt.

Cool ish to me. I do to say like when it started up and I gave it kind of the initial prompt and it says like building design system. I had this like who like just feeling of like God and then when the actual results came up I was. I famously hate the color purple, not the book or the movie and this has nothing to with Prince who I like.

but I just don't like the color purple. Of course it was all the purple gradients and the purple style. that was so, then I was like, I gotta get rid of this. And so I went and was trying to do that. what I had to do was my take your newsletter, turn it into a podcast episode because I had vibe coded that.

A couple of weeks ago, I was like, well, let's see what this does vibe designing it from the other side. And the vibe design was considerably nicer and better than what I had gotten through vibe code. No, I didn't use anything like impeccable or anything like that, which I probably should have. My other big thing was that. So it.

generates this thing and then it just says, well, what do you think about the flow and visual style? That's literally the direct quote in the side. And I'm like, okay, I mean, and you and I here have mostly been talking about visual style because yes, you can immediately be like, I don't like that. I don't like that purple. But flow is very hard to see.

just from static mocks and a flow of any kind of complexity, it's really hard to determine if it's any good or not, even for an experienced designer like myself, without having a prototype or some kind of person actually walking me through all the pieces of it and how they all connect. So being able to guess from a whole bunch of screens how the flow is,

And this was not a big app. This was only a couple of different screens, but I was still like, well, I, I really can't tell how, the flow is. And so that's where I'm like, well, this is not quite ready for prime time. I don't think figment has much to worry about in the near term. and I've been looking at some of the reviews of this and a lot of them had.

echoed what we're talking about here that somebody called it the design microwave, you know, it was perfect for quick results, but definitely you're not cooking a Thanksgiving dinner on it, that it just was not ready for real world professional use cases that you are probably better off starting with something like a Figma and working your way up to this.

Now some people did say, that it is great for things like just kind of these fast prototypes, fast thinking through things for idea generation. And I agree. I actually think it was pretty cool to see things come up so quickly and be like, well, this is one direction I could go in. The problem, as we've talked about before on this show, is that

that way is pretty generic and for certain things that's fine but for other things that's gonna be subpar and so a mixed review on it I really like the direction and as we've talked about when Figma and Anthropic did their deal a couple weeks back

I think we both agree this is going to be the future is this kind of blend of code design, code design. But what is definitely missing from this is that ability to have flow, discuss flow, easily change flow. And then really the kind of sophistication of the direct manipulation tools that Figma has right now. I think

Both of those things are where this is still a little behind, but definitely enough for a lot of people to have gotten really excited and really spooked by it.

Nik (09:08)
Yeah, I think that one of the things is how quickly they've been able to develop a pretty full featured design tool that is nearing some of the capabilities of what Figma can do and what took years. Right now, of course, Figma also has the benefit of it's got AI.

software development and they might be pushing really hard and I mean they developed make pretty quickly and so they're also pushing but I just I will say it's amazing how quickly they've been able to do so much tooling and I will say I feel like the folks at Google listen to designers and worked with designers to figure out what mattered because I think there's so many features here which are very critical.

Maybe the generation's not totally there yet, maybe certain features, but like the fact that I've got my infinite canvas, I've got multiple generations, I can explore multiple ideas. The fact that it's got an annotation mode, I can go in, I can annotate and say, hey, update this, update this. I mean, those things are really core to, think, how designers like to work. And so I have to believe that the team there,

either the designers on the team or their user research and testing team did a pretty good job. I'm pretty impressed with what they've been able to do here. I actually, yeah.

Dan (10:22)
Oh, yeah, I

definitely I don't want to I don't want to certainly be poo pooing their work because it is shocking how good it is. They don't have the 10 years that figma has been around.

Nik (10:36)
the original startup was founded in 2022 by designers and builders. It was acquired in May, 2025 and relaunched as Stitch.

Google acquires AI-driven UI startup Galileo AI, and that was in May of 2025. So Galileo was started in 2022.

Dan (10:55)
and this is still a lab project to be clear. This is not something that they are saying this is out of beta. It's completely launched, go do it. It is, this is still considered a,

a beta product and they're still testing it. And for that alone to have it knock 10 % off of Figma's valuation in like a day is pretty impressive.

Nik (11:23)
It's interesting that a lot of people are, of course, doing the comparison to Figma and it makes sense, but I actually wonder here if the potential competitor is actually Canva. Because if you're a pro designer, you've been working with Figma for years, you've got all your tooling, your company is there, like switching over to new systems is hard. And Figma is innovating pretty quickly too. However, if you're not.

utilizing a tool like Figma already, but you happen to have a Google workspace. You got your Google Docs. You got your Gemini. You want to build some stuff. You want to vibe code a little bit with anti-gravity or Google AI Studio. I mean, this as a product within that.

I wonder if it's competing more with the Canva market of like, I don't need, I'm okay with mostly somewhat templatized stuff. I don't know a whole lot about this stuff. So having that design sense kind of baked into the product and it does pretty good on its own.

maybe that's really where the competition is. Cause I could imagine a lot of small businesses, companies that are like, you know, we really can't hire a full on design team, but we have Google stuff and now we can make things with this. People can build up our own stuff. could totally imagine that being a potential place where this fits into the sort of design tool ecosystem.

Dan (12:36)
is there a large number of very small businesses that are rolling their own apps? Not yet. we've talked about this before, the distribution problem, you know, are there going to be 10 million apps with three people using each one of them? Like,

Nik (12:44)
Not yet.

Dan (12:56)
But you know, maybe some of them will have been created via Google Stitch. Now there was this whole thing of people being like, designers are finished. And it's like, yeah, I don't think so. I think that this can do a lot of things. I think it can do some really

cool ideation and starter stuff. I don't know. don't see it yet making that leap to changing the career directory of designers. certainly not as much as I felt like when Claude and Figma were having that partnership and being like, ooh, these things are

going to be really merging. maybe for people who aren't paying as close attention, this is certainly a flashier version of that. And so it is more of a demo about where we are headed.

Maybe demo is not the right word. It is more of an avatar of where we're headed. Is that the right word? I don't know.

You tell me. All right. Well, we're about to go. We're like, who knows?

Nik (14:07)
I don't know. I don't actually know what word you want to use there.

Dan (14:12)
So what do we think is vibe designing here to stay as a term?

Nik (14:17)
It's funny, I like the term vibe coding, vibe designing, I guess. just feels,

it just, it doesn't roll off the tongue. think that's my only child vibe design. Maybe it just doesn't roll off the tongue. was vibe coding. There's something about the words. ⁓ but yeah, I think that this is something, I mean, I imagine people are going to do this more. I think people are, I think I've said it before on this, like this is the new WordPress template is, you know, some generated.

Dan (14:25)
haha

Mm-hmm.

Nik (14:43)
file based on a design MD set of skills.

Dan (14:47)
I just think that all designing is just vibing so.

Nik (14:52)
Yeah

Dan (14:52)
It's just an unnecessary word. Vibe designing is just design, man. all right. Shall we move on to John Maeda's state of design in tech 2026?

Nik (15:08)
Yeah, is a... Maeda's been doing these design and tech reports for many years now. And so this is the newest edition, specifically keeping up with the times and thinking about AX or agent experience, among a couple other things. Dan, yeah, tell me about it.

Dan (15:25)
I'll hit on some of the very big high points here that he goes into. The first big thesis is that user experience design UX is moving more towards AX. This idea of we are moving from execution to more evaluation.

And in doing so, he lays out six different points that he thinks are really important for designers to start to embrace. listeners of this podcast will definitely know some of these that we've talked about. Some of them are new.

my general hot take is, the report as always, it's well-framed, it's readable. This report I think is pitched at a lot of individual designers trying to stay relevant.

And the article, a lot of it boils down to just keep learning everyone, which is inarguable, but also a little bit empty.

Did you have any overarching hot take?

Nik (16:27)
Meta is making a lot of good points about where we're at today. I think a couple of the things that I really resonated with, or I thought were pretty astute.

let's actually talk a little bit about each of the points. So one of the big points that made a mix is that in this age of agent experience, that you shift from the Gulf of Execution, how do I do this? The user figuring out how do I do a task with this software with this system to I tell an agent what to do.

and then it does it and then gives me some feedback. And I have to check, did it work well? Did it do it correctly? Did it do a good job? And I do think that that is an important shift. If you're going to be building agentic systems where you're now going to let agents do most of the action and you expect your users to simply ask for something, you do need to be thinking as a designer, how do I give feedback? How do I give interfaces to allow people to check, monitor?

make sure that the agents are doing what they want. Because you're now taking away the direct manipulation that the user would do and know I did this correctly because I figured out how to do it I can see the results immediately. One point that then made a makes is why coding programming, this has worked really well. And it's because you can see the output immediately, like,

You get code and the code works or not. Now, there's a lot of questions about, is that code robust? Is that code scalable? Is that code secure? And that's actually a bit harder to do. I mean, you can read the code, but it's somewhat tedious to do. But still, the ability to vibe code stuff and say, I asked the thing to make a thing that did this and it does it. I can just see that it does it, works really well. So there is a cool question here of, where else in the kinds of things that we design

might have a similar sort of pattern where you can build agents where it's very clear to tell it what they've done well or what they haven't done well. And so then this gets into the sort of six points around the learning. What should designers learn? So the first one was evaluation is all you need. Basically talking about how when you work with AI agents and AI systems,

A lot of things boil down to evaluating, are they doing what they're supposed to or not? And then telling them, yes, you did it, or no, you did it, and then updating. And so you act as more of a critic, and you actually have to define the evaluation. So one could argue if you're designing an AI to create design systems for you, you might create an evaluation set of this is what good design looks like in our company, and this is what bad design looks like. So move towards creating the good stuff.

Dan (19:01)
design is not coding. There's not often a does it work or does it not task. There's not any one right answer. There are better answers than others in design, but it's not like does it work or not. It's like doesn't work well or not and

Reducing design judgment to just writing evals. It takes this kind of rich skill and makes it sound like like you're doing QA that kind of makes me feel not not great. The other thing here about evaluation being all you need is that implies a maturity with AI agents.

that just frankly doesn't exist yet for most real world products. just hasn't happened yet. We're not there yet. That sounds like it is an automation, push the button, watch it do its thing without supervision. then at the end, oh, you can just evaluate it.

I just feel like this is jumping the gun a little bit we're not at the point yet where evaluation is all you need. You still need to do a lot of other things before you can get to an evaluation. Just one off the top of my head is like, what does good look like in this context? Being able to understand what is the strategic value of what we're doing?

What is the value of it to our users? just understanding the context around it is a task that is completely left out of this.

Nik (20:25)
I'm smiling here because in a way you've just sort of described this process of like, we got to get to all that we got to do all this rich design work and all this work ahead such that by the end we can start to evaluate and maybe the output is to get these evaluations, but it isn't that we just start with, well, let's just come up with the evaluations. You actually have to do all this deep thinking.

and deep work to get to that point. And by the time you are there, yeah, I mean, you might be right. Maybe we are moving into just more of QA, not just, but we're more moving into QA because we've done all that work. And so.

Yeah, I don't know. guess when you're... the way you're saying it, my mind is going like, well, okay, designers, we should still be okay. We still have a lot of work to do to even get to the point where we can start effectively using evaluations to then steer and have generative systems that are working for us that are then automating certain aspects of what we're doing and or certain aspects of what the product and the services that we design are.

Dan (21:25)
The second concept that Maeda talks about here is computational thinking. And this is definitely something that he's been beating the drum on for 30 years now, which is that every designer needs to become a programmer. And that was kind of the stance for 20 years. And now he's saying that designers should instead become computational thinkers.

that the designers need to understand systems and loops and structures and feedback. And while you don't need now to be a coder anymore, you do need to know how to speak machine. So, and this one was a little bit puzzling for me because I thought the whole point of doing this vibe coding stuff was that you didn't have to speak machine, was that you could say things in natural language and it will understand

What you want it to do that you don't have to know things like loops and structures and feedback because you can actually You can actually just say those things and it'll figure out the loops and stuff that it needs you can talk to it about the goals

Nik (22:35)
I might have been interpreting this under the lens of agent experience and designers who are creating agents. Because if that's the case, I think you do need to understand loops, structure, feedback. I think all this stuff of designing the experience that people have with agentic systems, I think that actually is pretty good. And I agree now, you probably don't need to be a coder anymore, but you do have to be thinking,

Dan (22:42)
Mm-hmm

Nik (23:00)
in these ways. And actually this is something that I like a lot and partly it's because a lot of the ideas that seem to be making better agents are actually the kinds of things that we as designers like, like reflective loops on action. I mean, you know, the way that a lot of the coding agent loops work is that they build something, look at it, say is this good or not? Am I done or not?

I could be better, how can I be better, what should I do, can I update my plans? And actually I think that the way in which designers think can translate really well, or at least some of these schematic or systematic ways we think we think, these models of our thinking, can actually map over to how we design agents. And it seems to make them work better, which is also the kind of unexciting thing. I think that if you're trying to be a designer,

and you're not building agents per se, I think you're right, Dan. think that we're, we, maybe that's like, folks who are designing the AI design tools and systems, like they probably should think this way, but eventually you can get it to the point where, no, you can, you can think like a human.

and you can speak like a human and then the machine will translate and turn that in. But I don't know, I mean, there's maybe an argument here of if you understand these things better, do you work with these agentic tools better? Do you work with agentic systems better? Possibly. But again, that's been the argument from like, do designers need to code? like, I don't know, some designers need to code and they're going to come up with...

new work that maybe pushes bounds, but I think there's plenty of designers who don't need to code and they're gonna come up with new work and push boundaries. So I don't know. I think we need more designers.

Dan (24:35)
these are fair points. And I will say the one thing that I really do agree with here is just the system thinking. If you can understand and think about systems, I think that's going to get you very far, particularly as you start evaluating the outputs of AI and AI agents, which are not very good at generating systems, which

is something that people who were doing deeper dives than we did on Google Stitch was that like, well, the systems start to break down after a couple screens. so being able to know systems and do systems thinking enough to tell when, hey, something's off here because it doesn't match the system is a valuable skill.

Which brings us to his third point, which is around taste matters.

the idea is that AI capabilities do not replace human judgment. Designers still need to know what feels right, what's timely, and what's worth making. It's the human advantage. And we've talked about taste a lot on this podcast. is it really a moat Is it really

going to be the thing that's going to save us. I feel like this is a lot of reassurance rather than analysis. Taste is a human advantage is pretty comforting to designers, but

Taste matters, feels a little like copium.

Nik (25:57)
Yeah, we've talked on this show and we've read writers who have basically critiqued the idea of taste and like, is that really what you need? You know, on the one hand, I think that I like the fact that we're talking about it. I like the fact that maybe this moment is...

pushing designers to reflect on what their own taste is and maybe starting to reflect on how that sets them apart and how that differentiates them. And I don't know if that's such a bad thing.

Dan (26:25)
No, I agree with that. And I think definitely you and I as educators, I know that I am trying to, in a certain sense, impart my taste on my students, being able to see things through my eyes and being able to see what I would feel is off. And they can choose, like, hey, don't actually think that's off or not.

I mean, they'd be wrong about some of these things. They're like, I don't care if these things aren't aligned. You know, those kinds of things. But I mean, they may all love purple and that is their choice. Again, they're wrong. But it is, you know, I agree that I have definitely been reflective about what is taste to me in this discourse. So maybe maybe that is a nice output from this.

The fourth one that he talks about is master the new instrument. He's saying AI is like an instrument and you can direct it, you can be collaborating with it, you can become dependent on it. But the main thing is to master it and have it amplify your judgment. What did you think about this one?

Nik (27:38)
Yeah, I I teach a class on how students can leverage AI in their design process. And the goal here is to get students to understand this as a new tool or a new instrument. think an instrument's a reasonable metaphor. You got to learn how to play it. You've got to learn what's possible, what's not possible and in exploring it and exploring its constraints.

and capabilities, you then can start to learn how to use it in a really effective way. So, you know, I thought that that's like an it's a reasonable thing. I think that designers should be trying this stuff out. And I think that they should be figuring out where it works for them and where it doesn't work for them. I actually tell my students, you know, I don't know what is going to be best for your process. And by the end of the class, I think you should figure out which were own.

AI augmented processes. so, and in the same way that I think, not everyone plays an instrument the same way, right? The guitar gets played in thousands of different ways. Some people do learn how to play very similar to other people, but there's always new ways to, play it and to new style and ways of engaging. yeah, the metaphor kind of works here.

Dan (28:50)
I mostly like this one too. I do think that the difference between learning how to play a guitar and learning how to play AI is that a guitar is extremely stable and has been so for at a minimum electric guitar decades the regular guitar hundreds and hundreds of years.

can master it because it's stable, but AI tools, they're not. What it means to be next to or collaborating with AI right now might be different in a year from now. It might be different, well, it'll definitely be different two years from now. there is always that lack of stability

And different models, and I've been reading a lot about this idea of the jaggedness of models where it's hard to tell if one model updates what all of a sudden became less reliable, what all of a sudden became more likely to hallucinate. You don't know because all that stuff is kind of under the hood. And that is, I think, a big difference here when mastering.

Mastering AI, I just think is a very different beast than mastering the guitar

Nik (30:04)
Yeah, another point to this, and I do think this is good, is kind of these three models of either collaborating with the machine, being dependent on the machine, or...

being above the machine and steering the machine. And I think that those are generally reasonable ways to think about it. I think ideally we want to push away from people becoming overly dependent on this. think models where you're collaborating and or steering it makes sense. I would probably say, yeah, that dependence is not really what we want.

And then from there, thinking about kind of how do you sometimes collaborate and how do you sometimes steer? I think they're interesting and it's something that designers probably ought to be doing and maybe reflecting on their own process when using these tools. one, do I feel like I'm becoming dependent or do I feel like I'm collaborating or steering? And then how do you kind of operate well within say either steering or collaborating?

Dan (30:55)
I think dependence is an interesting word. I'm thinking about your guitar metaphor and being like, well, I am completely dependent on the guitar to make sound. I can't make these sounds on my own. Same with AI, right? I can't make this kind of code on my own. I can't connect with this API on my own. can't.

generate these user interfaces on my own. Like I think there's always going to be some element of dependency in that you can't do the thing just by yourself. You can ask for help to do it, but there's going to be things that the AI is going to be better than you at. And I think navigating that is an interesting design space to be in. What happens when the AI

is better than you at certain things. And maybe you don't even know that.

Nik (31:45)
Yeah, and actually if you think about collaboration as one of the models or teaming and sort of this concept of human AI teaming, you know, at this point, I work with Cursor and I'm working with AI agents in Cursor to write all of my code. And if you took it away from me, I would be pretty bummed. It would be like you took my whole team away. It'd be like I lost my whole team. And I, you know, so yeah, you're right. There's a dependence there. Even if I'm not...

Do I feel like in my overall design work, do I feel I'm dependent on the AI to do any design work? No. Can I go back to sketching my ideas? Honestly, can I go back? Can I go back to coding myself? Yes. But boy, would, I would be real bummed if you took these things away from me. And in the same way,

if you took away, like if I was working with a team, even if I was say the, ⁓ a manager or if I was a CEO and I had a great team I was steering, if you took away that team, I'd probably be pretty bummed. Now, if you're like, if you're replacing the team, maybe that's different, but if you actually like said, no, no, you have no team anymore. You must work on your own individual contributor again. Yeah. You, your ability to potentially create work at the level.

that you were may be significantly diminished. And the other thing too is that some people may actually become virtuosos of sorts of this collaborative team oriented thing. Like we work, you might work best when you work on a team and direct collaborate and that actually working on your own, you're nowhere near the capability you have. And thus, yeah, of course I like what you're saying there. There is a dependence, but maybe that's a healthy dependence.

Dan (33:11)
Isn't that what an abuser would say, Nik

Nik (33:13)
I have no idea.

Dan (33:16)
All right, let's talk about the next one, which is tasks do not equal jobs. Jobs are changing at the level of tasks, not disappearing all at once is the idea here. That tasks are being automated, but jobs are actually a collection of tasks. And I agree with this.

Google Stitch, someone said when shown the interface design was like, well, that's 1 % of my job. That is not my entire job as a designer. It's just a sliver of it. And I think

That is 100 % true. I think so much of people's perception around jobs equaling tasks and then being like, well, this thing could do everything that a designer can. And no, that's not true. The actual production part of design is a piece of the job, maybe 25%, maybe 40%.

But it is definitely not the whole job. What was your take on this?

Nik (34:25)
Yeah, I really agree with this point. mean, yeah, jobs are a collection of tasks. And honestly, I think within design and many forms of knowledge work are even more than that. I mean, we are interested in as part of the job and why we hire people is how you think. If you equate everything to just be the tasks you do, that's sort of this factory model.

of work. It's sort of this mechanized model of work. Like everything you do can be broken down to a task and then that task is something where you're an input output machine and that's, you know, I'm not so sure that that's really what most jobs actually are. And so yeah, I agree with this point a lot. I think that if you do work,

And this is a point that made a makes production heavy tasks are more exposed. I agree with that. think if, if you have certain tasks, which are very input output, they can be wrote, they are somewhat mechanized. Maybe they're not fully automated yet, but they're mechanized. And even your own thinking is mechanized when you do it, which there are aspects. mean, there's aspects of my job, which are mechanized in that way. Yeah. I think that stuff can get automated and it, you know, likely would be because there's probably some efficiency and some benefit to that.

Dan (35:35)
And I wish he had stopped there, but no, there is a sixth one, the tree shape designer. And this is an update to the T shape designer. And the idea is that you are now a tree and you go deep into design, but you also grow branches into AI and systems and evaluations.

and computational thinking. is just a restatement of most of the points in this piece dressed up in a new metaphor.

and identity archetype.

That was my feeling on tree shape design. Or should I say that was my chopping down the tree shape design.

Nik (36:18)
So I'll take a little bit of the more positive spin on this one. ⁓ know, actually, so T-shaped, tree-shaped, yeah, the metaphor is kind of cute here. It doesn't, on first reading, it didn't feel super different to me than like we've heard about pie-shaped designers, as in like the Greek symbol pie or M-shaped designers, or you hear about all these different kind of shapes where they're kind of growing and expanding, they're broad, they're deep.

Dan (36:18)
Do it, do it.

Nik (36:43)
Yeah, fine. The tree metaphor can work. Actually though, the thing that I somewhat liked about this is that trees often, not always, but are often part of forests. And actually every tree can be somewhat different.

You know if I'm gonna if I'm play with the metaphor here, I'd like a forest of designers with a variety of different shapes and skills and Likely you need to figure out as an organization. What's the forest that works for you? I realize I'm leaning in to it really heavily, but you know, I don't know it kind of made me

Dan (37:13)
You're growing

your roots into it. You're stretching out your branches to embrace this metaphor. You're a tree hugger. That's what you are, Nik Okay. And that was John Maydaz, state of the designer in tech, 2026. Now let's move on to our last story.

Nik (37:18)
Yeah.

There you go.

Dan (37:34)
which is all about design skills and coding design skills. And we're looking at two different sets of those from Marie Claire Dean or MC Dean, who is a design lead at shopping and AI at Google in London. And

Nik what are these design skills and how do we put them into practice?

Nik (38:02)
Yeah, so the design skills, these are very similar to the stuff we've talked about on the show before, like impeccable, and even what we just described earlier with Google Stitch's design.md file, which is it's sort of a set of guidelines, rules, statements that are useful to guide AI agents in how they create something.

So for example, you could have a design skill for color and in that you can kind of define the kinds of color, the ways to think about color. You can have design skills for animation. And for example, you know, put all of, Dan Saffer's rules of micro interactions into that file. And now all of a sudden, your agent understands that or at least can build with that.

The thing that's cool about these, and the thing that I'm excited about is, one, they're plain text representations of design ideas, concepts. Two, a lot of people are sharing them. Like they're putting them out there, open source, online. MC Dean has done this with 103 skills. Some of the things that are

Dan (39:02)
This

is God's work here. This is amazing.

Nik (39:06)
Yeah, and it's cool because in a way, like it's, it's this interesting way of capturing design knowledge that, I don't know if I've seen before, you know, people talk about this. We might talk about how we design or what we're doing here, but actually this is a very interesting moment in time where we are trying to say, I have to tell these machines how to do something and I want them to do them in a designerly way. So I'm just going to write it all down and give it to them.

But then I'm gonna share it with everyone. I think that is really cool. I think it's an interesting way to basically see into the mind of other designers, like how are you thinking?

Dan (39:39)
she talks in here that, about some of this being a knowledge preservation problem. And it's a great way to encode institutional knowledge. So it persists beyond any individual person's tenure. And I think what you're saying is that it is a great way collectively

as a professional group of professionals, this is a great way to start to encode our professional knowledge in design and HCI so that it persists even when a designer or an HCI person isn't there, that these things are still accessible. Literally accessible.

Nik (40:18)
Yeah.

Yeah. I, and I do think that that's really interesting. ⁓ the other thing that's, neat about it is that people are creating them for all kinds of stuff. You know, I mentioned some things impeccable was one where it was very much about, I would argue design aesthetics, visual, you know, a little bit of interactivity. you know, Dean has actually created a whole set of skills that are more about

strategy or design handoff. Like what would I need to do to have the agent hand this off to a developer? And actually it's probably so that the agent can within its own loop sort of hand off better. And that's really interesting. So there's all these other skills that designers use, are more kind of our thought skills that are kind of being written down and encoded. Another one and

actually there's two articles that we're referencing here is a whole set of skills around inclusive design and accessibility. And this is also really cool because this is the kind of thing where for a long time, you people have had challenges with this. hard. It can be hard to make things accessible. There's a lot of things you need to consider.

And so now the fact that you can share information, you can have these skills that you can actually just give to your AI agents. And then now they somewhat understand this. mean, that's, again, there's going to be limitations here. I'm not going to claim that any of these are like, well, okay, cool. We've solved accessibility because we've got, 40 skills here. One though, I think it's cool because we can test and see if it works too, because Dean has made these.

open, you can take them, you can build on them, you can improve them, try to update them. in a way the skills.

push us to this somewhat science fictiony future where it was like i can just download new skills right you know neo in the matrix yeah i know exactly but it's sort of like well my agent now knows how to design better my agent now knows how to strategize better now and i think that this is

Dan (42:07)
I know Kung Fu, yeah.

Nik (42:18)
It's kind of interesting. It's interesting to see, it's interesting to explore. And I think the other thing that is something that maybe our listeners, you're utilizing this, is get in there and actually like observe these, inspect the skills, update them. The cool thing is that it's totally doable because it's just a text file. And so you can kind of update them, you can add to them, you can make them your own.

And I think that that's one of the things that I really like about this stuff, as opposed to it being built into the systems that we download and say, well, this is what I get and I don't get to change it, or I have to prompt change it. I've got to try to steer it through this way where this feels a little more explicit and it feels like something I have more control over. And I really like that.

Dan (42:58)
Is there a central repository for all these things?

Nik (43:01)
Yeah, there are like skill aggregators that have all kinds of skills for AI agents where they can download. Actually, I think open claw systems are utilizing these where then the open claws can actually go download skills when they realize they need something. I will say that is really insecure because people can just write all kinds of stuff in skills and then put them up and have all kinds of access to your things. So please be careful.

if you're deciding to use OpenClaw and give it the ability to download its own skills on its own. Yeah, I don't know though. mean, if there are other, like I'm not sure if there's something out there which is like a design skills.

I mean, Claude is putting together a marketplace. And so if you're utilizing Claude code, things are being sort of aggregated there. So that's one area.

Dan (43:53)
Maybe we just need to vibe code it ourselves.

Nik (43:56)
Maybe. Or-

Dan (43:56)
Sounds like a

good summer project for a student.

Nik (44:00)
Yeah, or one of our listeners needs to let us know ⁓ hey, here's the design skills, like an aggregate of all of them. I will say it wouldn't be that unreasonable because the cool thing is, is that everyone that I've seen for the most part shares their stuff on GitHub as open source. And so you could actually even just do a meta list, like awesome design skills, and then put in the GitHub repos. You can have a Git repo that's a big repo for all of the...

that links to all the other repos. I actually wonder if that exists now because that is a common thing, the awesome something on GitHub.

Dan (44:34)
So let's say you are like me, a noob, vibe coder, five designer, and you want to start playing with these things. Do I just download them? then what do I even do with them once I have them? Do I have to do some kind of installation

or how does that even work?

Nik (44:56)
Yeah, so what you do is you would download the skills, which are markdown files, so they're text files, but then you do need to tell if you're using Claude, for example, you have to say, hey, here's the skill files. If you use other editors, so if you use something like Cursor, they have their own way of including skills, and you basically include the skills in the project folder that you're working in. And then the agent,

basically has access to those files or you can install things that sort of a system level within the Claude invite for i know Claude you can do it in sort of at the

Claude environment level and then anytime you ask you can Claude can choose to use those skills or You can actually use commands to use a skill That's the other thing that's kind of cool about the way that skills are architected is that you can slash command a skill so when you're prompting say please use this skill and just include it within What you're doing next?

Dan (45:48)
Right, I remember our famous slash delight this from Impeccable.

Nik (45:53)
Right, and so now you can have slash design ops handoff. can have, slash.

Dan (45:58)
accessibility decisions, et cetera, et cetera. Do I have to go into the terminal to install these things?

Nik (46:00)
Yeah.

No, you don't have to. I believe you can install them in like the Clawed desktop app,

Dan (46:14)
stuff like this all sounds amazing, but i'm always like, my god. Okay. How do I get this cool little thing? into

what I'm doing in an easy way. And so much of these environments are still set up for people with a technical background. And I think one of the things that will open it up for a wider group of people will be having some of these things easy to just install almost like extensions on a browser where you're just like, well,

Just add this thing into my Claude or into my Gemini and go in a way that I don't have to mess with all this other stuff. personal rant or slash personal rant.

Nik (46:58)
if you are a Claude desktop user, if you go to the plus button, there is a skills thing there, and there's a menu that's manage skills, and

You can add skills. can actually upload the skills and it just says, you know, upload the folder of Markdown files. Here you go. You can write your own skills and they even have a little tool. If you want to co-create a skill with Claude and then save that. you can actually sit there and have Claude create a skill with you. And then it'll save it. And then it becomes available as a slash skill command within the Claude desktop app. So you don't have to go into the terminal, but the cool thing is that if you do that, I think within Claude, ⁓ if you do use Claude,

Claude code would then also have access to it as well.

Dan (47:40)
that's cool. Amazing slash amazing. So here's here's one final question about this. Does this slow down? The delivery of my results. So for example, if it has to go through all these different. skills and apply them.

doesn't that slow everything down

Nik (47:58)
Yeah.

basically, it's all more tokens, right? Every time you take in a skill, that's more input tokens. Every time you do something with it and go through, the reasoning loop, like it's going to be using more. And so, yes, it probably will slow things down, but arguably the benefit of that is that it hopefully does something that's closer to what, you want because it's utilizing that skill to do it. rather than creating, the generic

Purple UI, if you have your design aesthetics skill that says, we don't use purple, we only use this color palette, we use these fonts, we do these things, it should basically, without you having to tell it explicitly, please don't use purple, please don't use this, please use this font, it just looks in the skill, goes, yep, this is the design skill, this is how we do stuff, and it creates it there. So yeah, it can slow you down, it probably will cost you more tokens.

but in theory it should do things in a better way.

Dan (48:51)
I think this brings us around full circle is that these skills are codifying design quality and design taste.

And that's an interesting place to be in because then it's like, well, I'm using a skill that is, let's say I make that is my design skill, my sensibilities. In theory, you can be as good of a designer as I am because you are then seeing through my eyes. And that's an interesting place to be. And I wonder if I can sell.

myself as a design skill that you can then buy. Why is there not a skills marketplace yet? See, you open source people. This is not capitalism. MC Dean giving away all of her good stuff. How dare she? Socialism, that's socialism. No wonder she lives in London.

Nik (49:29)
Yeah.

So.

Well, there are-

I mean, I think that...

I this is a potential direction where things are going. mean, they already are talking about them as marketplaces and it's a marketplace where you're not maybe paying, but in theory, one could say in the future, you could pay for that. And I imagine we'll see that soon. I actually do imagine that people will be able to sell that in the same way that you sold a WordPress template.

Dan (49:48)
Ha

Nik (50:04)
You could now sell a design skill and just say, hey, I have already written, if you like this style, you like this way of doing stuff, you can pay a couple bucks for my design skill, add that to Claude. There you go. Now I will say.

To your point, it's a really interesting way to try to codify these things, but it's still all language. And there's still so much of what we do as designers that I'm not sure is well represented in language, because these are just markdown files. Even when I want to talk about style, sometimes I may be better at describing my style in a way with a mood board of images rather than with language-driven

set of instructions. So I still think there's probably limitations here.

Dan (50:46)
Yeah, for sure. I mean, there is the unconscious competence problem, right? Where I know a whole bunch of things, but I don't remember that I know them. And I certainly don't know how to describe them because I just do them without even thinking about them. And now I have to sit down and make a skill and think about like what I would do in all these theoretical situations. That's a pretty hard translation problem, but

A very interesting translation problem,

Be cool to have an AI that helps people translate their unconscious competence into skills.

Nik (51:19)
Yeah, and that may be where some of the skill creator interactions that AI apps are developing, like Claude's skill creator, is trying to go.

Dan (51:27)
Right, but I bet they are still trying to do it in language. You may not be good at language. You may not be a good writer. You may not be a person who thinks like that.

and being able to have something that teases that stuff out of you through examples and activities could be really interesting.

dollar idea here.

Nik (51:46)
or a great research project.

Dan (51:48)
or a

great research project or a zero dollar idea an idea that costs us money to make one of the other well i think that's our show for today thanks for listening be sure to rate and subscribe and tell your friends about us because we're starting to cook here on our 11th episode and we'll be back again next week

with some hot takes, some hot new skills, and we'll see you then.