1006: Can AI Make Good Design?  🔗 **Source:** [1006: Can AI Make Good Design?](https://share.snipd.com/episode/2531af65-49a7-4b79-a111-27537b141414) ## Highlights - **AI Explores Patterns Not True Creativity** - AI excels at exploring variations and regurgitating existing patterns but doesn't invent truly new creative ideas. - Wes Bos found AI produced a “warm stew” of previous work and often seeded outputs with the same familiar elements like repeated names (e.g., Sarah Chen). Transcript: Wes Bos Welcome to Syntax. We lost the first like 20 seconds of this episode on Riverside, so I'm here recording it again. I went to find the shirt that I was wearing. I spilled shawarma poutine on it so it's dirty. So listen up. We're going to get the editor to try cut it in perfectly. And I'm just going to start saying what I was saying at the beginning of it. We have a show on AI design, and don't turn it off just because this isn't the AI design show where guys are like, oh, I one-shotted this. But it looks amazing to me. And I think designers are toast. No, that's not what this show is about. This show is about we know that AI can write pretty good code. And that's great because it's very deterministic. It's you can be tested, things like that. But design, like can a model create something that is good, that is functionally useful, that converts well for the business, that helps the actual end user? We're in this like weird space right now where it seems like everybody is just ripping each other off. Every app looks exactly the same and there's tons of new tools to help you. So in order to get to the bottom of this, we're going to attempt to answer some of the following questions. Can AI be creative? Can it make you creative? Can good design be extrapolated? What about user experience? Can AI make a good user experience? You know, does it know about all of the complexities of your UI of what people are trying to get done? ([Time 0:00:00](https://share.snipd.com/snip/c24f6d6f-1022-4bed-8ea6-8dc6330ee9b4)) - **The Sarah Chen Testimonial Mystery** - Wes Bos tracked the same fake testimonial names across hundreds of AI-generated sites and discovered Sarah Chen and Marcus Rodriguez repeated everywhere. - He validated this by prompting multiple models and Google searches, finding Sarah Chen often appeared first in generated testimonial lists. Transcript: Wes Bos I got on to this whole thing because i get these emails every single day from these like slop coded startups. And they say, like, here, check it out. Like, let me know if you if you want to use it. And I always click through to the website and see what it is. And it's the same thing of everything. Right. Someone just logged into VZero or Lovable or Claude or whatever and type like make me a business make no mistakes and it shout out some business out the other end and what's interesting Is that on all of these i see the same name over and over again i see sarah chen on there and i was like where is sarah chen coming from and i i was like i gotta get to the bottom of this so first Of all i thought like maybe this is just the same person doing this over and over again. And it turns out that wasn't the case. It was many websites. So I was like, where is Sarah Chen coming from? Like, maybe it's coming from like one of these like site builders, you know, maybe they have like seeded it with a template that has Sarah Chen as a testimonial and just uses that. And I looked into it and it didn't seem to be the case because all of the examples that I had found, I looked on GitHub and the use of Sarah Chen in the last year has just skyrocketed. And they were all from like different generators, you know, some of it was in like PHP code, some of it was in like Markdown documents and PowerPoint slides. And there didn't seem to be any commonality between them all. So finally, I went into chat GPT and I said, give me a list of like five random testimonials or not random five testimonials. And Sarah Chen, bam, first one. And I went into every other model and every single one had either I think four out of five had Sarah Chen as the first. And then somewhere in there it had a variation of Sarah Chen in there which was absolutely wild so every single model is using Sarah Chen as it and I thought like hmm if you think AI is is Making you creative it's not it's making you like what programmatically what what did I what was I saying it was called hold on yeah yeah it's making you probable or like yeah i gotta ask Scott Tolinski How how did you notice sarah chen because for me my brain just glosses over that stuff you know the times that i don't are when i see my good buddy wes up there i'm like hey that's wes he's Wes Bos Got a little west quote but i don't think i would have noticed any random this guy emailed me and says like hey we are a business and then i like go to his website and it says Sarah Chen. And then there's like obviously an AI photo. And I was like, that's not Sarah Chen. So I googled it. There was another one. It was like Marcus Rodriguez was the other one that kept showing up. So I googled Marcus Rodriguez and testimonial. And I found, I'm not kidding, I probably found two or three hundred like startups dot coms that had him as a thing and in in many cases his like attribution of the um well company he's associated With was also the same it was like info tech or something like that uh so i like just by googling and i found like man this mar Rodriguez. And then beside it was always Sarah Chen was used over and over and over again. So it's just deterministic. ([Time 0:03:20](https://share.snipd.com/snip/b86c66f7-bafd-4363-b4cc-486320d61d05)) - **Table Stakes Kill Conversion Edge Fast** - AI makes high-performing tropes easy to produce, so tactics like HDR "wow face" thumbnails become table stakes quickly. - Wes predicts when everyone uses the same thumbnail style it will stop converting and creators will seek new aesthetics. Transcript: Wes Bos But let's talk about YouTube thumbnails for a second. The latest Gemini can crank out a Mr. Beast style thumbnail, which is like super high HDR. It widens eyes it does all the like stupid youtube thumbnail stuff that like we do it as well right because why because it works right but now that it they're able to just simply you can Just type type type and it cranks out what you used to pay hundreds of dollars for to a designer for this type of thumbnail. Now everybody, every single thumbnail is like that on YouTube, right? It's the wow face, all of that. It's the blue versus red, the thing down the middle, big letters on top. That is now table stakes. And I think that now that is table stakes, it's going to stop working in the next two or three weeks. I guarantee you, because everyone has figured out, oh, I don't need to. Previously, it was somebody like Mr. Beast. They had somebody on staff figuring out these and then other people started to copy those. And now it's simply just like a prompt away. And every single thumbnail looks like that. Now it's going to stop working because it's table stakes. Everybody can do it. Right. And now we're going to have to guarantee we're going to see like a shifting of YouTube thumbnails to something else. ([Time 0:08:52](https://share.snipd.com/snip/818c3816-c9c9-499d-99ce-198a50298a7c)) - **AI Can Extrapolate From Rules But It Converges** - Given a curated set of tokens, rules and examples, AI can extrapolate a coherent design but results converge to a recognisable template. - Wes used Claude plus Google Stitch to summarize features then auto-generate decent layouts that quickly started to look the same. Transcript: Wes Bos Meaning that if you were to provide it with a list of examples, a list of rules, colors, spacing, type, all of that stuff. If you sort of make a good base of design, can it then be extrapolated into something larger? And I think partially, yes. That's what we're seeing now with tools like Google Stitch, where you go into it and you tell it what you want. And in my case, I found what was helpful is I asked Claude to summarize all of the features and all of the layouts and whatever that needs to happen without describing what it looks like. And then I took that, copy pasted that into Google Stitch and it cranked out like a decent looking design. But after using it three or four times, you can clearly tell that they have seeded it with some good stuff, right? They have a whole bunch of templates. They have a whole bunch of good fonts. They have a whole bunch of rules about what colors work well with each other. They probably have a massive prompt of like, these are the things that look good. And then out the other end, you, the user, can say, all right, well, I want it to do this, right? I need form inputs or sliders or all that stuff. Then I want to use these colors. And then it can take those things and sort of extrapolate that into something that is looking half decent. But even after it's been out for like a month or two, I can now eyeball one of these Google Stitch designs. ([Time 0:10:38](https://share.snipd.com/snip/3e407622-63ea-4e19-9d15-1b1bbedcd63c)) - **Design.md Standardises But Risks Drift** - Google's design.md is a steering/validation spec for design tokens that aims to standardise rules across tooling, but it risks duplication and drift from code. - Scott worries it duplicates root CSS declarations and adds another repo file that could go out of sync. Transcript: Scott Tolinski So what is it? Like a couple weeks ago, Google came out with their design.md spec, which the idea for the design.md spec is that you are putting, it's kind of like your agents.md or any of these steering Documents that you have inside of your code base, but only for design patterns. And when I first saw design.md, I was looking at it and it's basically what it's a markdown file with front matter about your design tokens, your properties. Wes Bos It's a parser as well and like a linter. So it is a lot more structure than just a markdown file. But you're right. It is a steering document. Right. Scott Tolinski And so there are those aspects of it, too, right, where there is validation involved in this document but it is really in my my initial reaction was like this this looks like just a root Css declaration yeah there are more things in here but it feels like prime territory for drift for kind of it's like duplication of ideas that are expressed in code already but the idea Is is that your ai uses this as essentially a design document this is the rules of this library if you're adding a design to this it's following these dang rules whether that is the colors Fonts all that stuff the naming of things uh what belongs to what what the spacing are rounded corner spacing again yeah i still don't necessarily get it get the need for it but i think Wes Bos The big play here is that like it's a standard that can be translated between things so like if you have a figma that can output this and then like you have like a like a tailwind it can be Converted from one to another um but first of all i wonder like do we do we need this rigidness and second of all i say if there is rigidity, I feel like I would rather do this in like a CSS file With comments, maybe. Scott Tolinski I would love, I know this is not going to happen. I would love a standard way of just defining CSS variables because that's really kind of what this is. Because like spacing, rounded, all that stuff, just come up with a standardized name for all this stuff in your root CSS declaration, point there add some comments about the overview Or whatever bingo bingo now you're not having a ai document and a code document that could get out of sync the drift kills me i think there's there's a different world for this as well where Wes Bos The whole design tokens folks you know where you your sales force and you have 40 different implementations of your design. But I'm curious to see if there's any uptake. I certainly don't want another file in my root of my repo. Oh, yeah. Right. Another root of my repo. ([Time 0:13:54](https://share.snipd.com/snip/98c57604-ccbc-4d69-841c-3d8b886fd850)) - **AI Struggles With Deep UX Problems** - AI is poor at holistic UX because UX requires domain understanding, flow decisions and handling edge cases, not just arranging UI elements. - Wes argues booking-specific flows (e.g., U-Haul trailers) need human designers to reason about alternatives and user goals. Transcript: Wes Bos So much of design is not just making it look good. It's about making an interface that is consistent, something that has clear purpose, low friction to the user, right? And quite honestly, I don't think AI is very good at this. I think it's great at throwing buttons on there. I think it's good at having cards and putting cards inside of cards. But when it comes down to this is our specific app and we are doing this one thing, right? Are booking trailers you know we are u-haul that maybe that's a good example i freaking hate u-haul u-haul's app is absolutely awful the ux is i won't makes me long kill myself and it's Just the worst i don't think you could say that you could say i think, please. The UX makes me want to die and it just, it drives me nuts. Right. And what needs to happen there is that a designer that understands UX needs to come in and understand what is actually getting done here and figure out both the interface as well as the Flow for, for actually booking a trailer on U-Haul. Right. And I don't know that if you were to give AI these problems, that it would make a good thing because it just really doesn't understand. Maybe you could, I don't know, take a whole bunch of user feedback and pipe that in and summarize it and stuff. But I think at a certain point, you still need somebody who understands the stuff and has a good UX because it's just you're going to get these awful interfaces at the end of the day yeah Scott Tolinski I find most of the ux work that ai does even with strict rules and you can download a ux skill from skills whatever that malware and throw it on your system and prompt away make this ux more User-friendly and like it just flat out is bad yeah they're like i have such strict ux rules in mind and it's still just like falters with that i always i find i find myself saying move move Wes Bos Out of the way i let me do it like way more than ever i even think just like so much of that is just subjective you know like oh i'm taste i hate that like taste thing so much. But yeah, it makes sense. It's just like somebody needs to taste use this thing and go, oh, you know what? When I selected my location and then the trailer wasn't available, there should be a very good UX for being able to realize, oh, there is a similar trailer at a different place nearby, Or this is a trailer that I could get that is similar at this location. And yeah, there's just so much to it. ([Time 0:17:00](https://share.snipd.com/snip/956bd885-bbea-415e-8c27-938f4e928d11)) - **Some Design Is Algorithmic, Some Is Taste** - Some aspects of design are programmatic (algorithms like text wrap) while core aesthetic choices like color and trend-driven taste resist purely programmatic selection. - Wes cites Warp Terminal theme attempts as an example that looks worse than hand-crafted themes. Transcript: Wes Bos This is kind of an interesting one. And we've asked this many times to many of our guests over the years of, can you programmatically pick good colors? And I think the answer to that is no. So I think Warp Terminal is a good example of this. They have this concept of themes and you give them, I don't know, three or four colors of what you want it to do and then it tries to to skin the entire terminal based on those programmatically Lighter darker contrast you know all that stuff and it just it doesn't it's not a good looking terminal i don't think so it just doesn't look very good whereas if you look at something Like like a vs code terminal where you can control absolutely every little thing, it looks a lot better, right? I don't think you can programmatically pick good design. And then also you have this aspect of the colors are trendy. What comes and goes is very trendy. And now as everybody's ripping each other off, those trends move so much faster than they previously were. So I don't think it can be programmatic in that regard, but I do think it can be programmatic in other ways, like text wrap balance, right? Text wrap balance in the browser makes, if you have two lines of like an H2, and it will figure out how to balance those things. That is an algorithm, right? The browsers implement those and they figure out how wide the letters are and whatnot. So I think like that can be programmatically done. ([Time 0:21:05](https://share.snipd.com/snip/d22933dc-8bf7-452a-badd-e8c8cd3220fd)) - **Data Lets Algorithms Optimize For Outcomes** - With enough data and A/B testing, algorithms can optimise design for outcomes (conversions, attention), blurring into addictive product mechanics. - Wes notes platforms already tune feeds and UI patterns to maximise engagement and conversions. Transcript: Wes Bos And that's kind of an interesting question to me, because I think with enough data, yes, the reason why like TikTok, Twitter, all these things are so addicting is because they have these Algorithms that are able to figure out like what you actually like and then like with enough a b testing um you can figure out what all the dark patterns are or what are are all the best patterns You know like we had travis on who works at google and he he does exactly that is like he does design for what happens when there's a tornado and someone google's tornado what should pop Up on Google? And like, that's kind of time sensitive, right? So I'm sure they're very data heavy on that type of thing. And I don't necessarily know if that pushes itself into AI, but certainly you can start pushing that into algorithms. So I think yes. But then I think about like marketing. I saw this tweet the other day where someone's like, I replaced a social media person that was $500 an hour with this one skill. And it's like, no, you didn't. No, you didn't. You're not. Scott Tolinski No, you didn't. Wes Bos Again, we're talking about table stakes. As soon as people figure out what works in marketing, that changes, right? And it's just this constantly evolving thing. And now we're starting to see, I don't know if you saw this or not, but Grok's Imagine, where they can make video of somebody sitting in front of a camera talking. And it is almost indistinguishable. So good. And I asked myself, why do we need this? But I also was thinking about that. I'm like, man, we're going to have to get rid of our nice cameras and lights. Because as soon as everyone starts using these AI-generated influencers, I'm just going to scroll past. As soon as my brain detects nice lighting nice camera good microphone it's looking too good yeah good swipe ([Time 0:24:09](https://share.snipd.com/snip/0b18c612-90fd-4447-8875-25d70938795b)) - **AI Influencers Could Become Background Noise** - Wes observed Grok's Imagine can generate realistic talking-head videos and worries oversaturation will make polished AI influencers fade. - He and Scott noted audiences may prefer imperfect, human aesthetics as pushback against glossy AI content. Transcript: Wes Bos And it is almost indistinguishable. So good. And I asked myself, why do we need this? But I also was thinking about that. I'm like, man, we're going to have to get rid of our nice cameras and lights. Because as soon as everyone starts using these AI-generated influencers, I'm just going to scroll past. As soon as my brain detects nice lighting nice camera good microphone it's looking too good yeah good swipe past you know yeah so we gotta get here please play that i think that's a good Scott Tolinski Point i mean as we uh when i interviewed sarah bird at microsoft she said that the things that she's most interested in now are like the things that are like the most human and imperfect, Like Japanese pottery and stuff like that, because she's been working in machine learning for so long. And I do think that that is going to be an aesthetic pushback to this glossy, fake looking everything. Yeah. So I'm sorry, folks, that I look too good that you might think i'm ai he's not say that i'm not sorry comment down below who is more ai looking scott or i or me i i can't even somebody did think I look like ai in one of my videos and they were being serious why is this video so weird are you ai but it was an audio desync issue that caused oh yeah i think yeah well even like my little Uncanny valley every ([Time 0:25:39](https://share.snipd.com/snip/0b875e73-0ad0-45fe-b108-5e530f03efa7)) - **Automate Repetitive Design Work With AI** - Use AI to offload tedious, time-consuming tasks like background removal, slicing icons, and renaming assets. - Wes described asking Claude to slice 12 icons, auto-trim backgrounds, and export named files—saving ~20 minutes. Transcript: Wes Bos Think there is a range right there's like a tool aspect to it which is like i am simply just using these as tools to help me get my work done more productively help me be able to research it's Just whatever your end game is these tools can certainly help you like on one end of the spectrum there is background removal in photoshop and i don't think there's anybody out there That like as much as many designers hate AI, I don't think there's anyone and be like, you know what, we should be using the like pen tool or the little transform selection tool and like Cut out around like I'd spent my time doing that and cutting backgrounds out. I'll take the AI background removal any day. You know, that's so important for me. Scott Tolinski Yeah. I mean, those types of algorithms, people act like that's just completely unrelated, but it's not. Wes Bos Yeah. Well, I think like that one seems relatively harmless, whereas some of the like on the other side, like there's the complete brand design, you know, like I'm starting to spot like I always Think about plumbers. You see a truck driving around your neighborhood which is a plumber and you can i get you can almost eyeball which model they use to create their plumbing logo you know like you can you Can start seeing the ones that are like two years old and you can just tell what it is and now you're starting to see them and they have that like sort of like um that orange yellow haze to It um that all these like chat gpt images that get kicked out are and i think there there certainly is a space for it i think the best designers are going to figure out how to use this to create Good experience to create beautiful things but not just actually do the dumb brain stuff for you and like we've seen this over the years, right? We had clip art and then you had these like corporate blobby people on all these websites. And then we had these like pre-purchase packs. Something you may not see, but you'll probably notice it now. If you ever see like an inanimate object, you know, like a heart or a burger, and then they put legs on it. And it's like, it's like a burger that walks like the mascot for my new burger shop is a burger that walks or like the mascot for my new coffee shop is a coffee cup with legs and a little eyeballs. And it's just walking, you know, like I've, I have stickers that do exactly that. And that has become such a trope that like, you could just go online and purchase like 100 eyeballs and legs and then you can just add them to whatever it is that you want. You know, I'm a lawyer. I'm going to take the gavel and put put legs on it. But then everybody started doing that. And now I can't unsee it every time I see that we have a Canva, you know, Canva brought decent design to absolutely everyone before we had canva it was just word art clip art bad stuff that People could sort of just put together and then canva came out with like half decent looking stuff you can still spot it but is that bad you can spot it a mile away yeah you could spot it a Scott Tolinski Mile away i mean i think with any of this stuff that like you're in that same line that you're drawing your AI logo and a lot of that stuff is word art. Version of that and and if it's not then it's probably derivative or it's just a copy paste of some other basic logo man that's that's a genre of twitter that drives me nuts is people being Like ai is changing everything i made this incredible logo and it's like we can tell there but this logo sucks yeah that's the most frustrating thing and and also like do regular people Wes Bos Care they probably don't i'm gonna be honest yeah they probably one thing that drives me batty is that i get all these tiktoks for suno metalcore you know like like ai music and all the Comments are just like i don't care that it's ai it's it's a banger and like the country music has had this for a long time like they know people know programmatically what makes a good Catchy riff and, like, what will people sing out and what will be a good bar song. And people don't necessarily care because it's just, I don't know. I find myself. Slop for the masses. Yeah, it's slop for the masses. So that's another aspect of it as well. Like, I hate to be the guy who's like no uh you won't do that because like we had those people with with dreamweaver right don't use dreamweaver it kicks out crappy code you know i did use Dreamweaver yes i know yeah that's a kind of uh interesting part but here we are before we get into our final thoughts here let's talk real quick about just tools that people are using. Figma, MCP, Paper, Pencil. Claude is surprisingly a very good design tool. The other day, I had a grid of, I think, 12 icons, right? And it was on a white background, and I needed to, programmatically, I needed to remove the background. I needed to slice them up into their own. I I needed to trim them, and normally I would go into Photoshop and I would slice and dice and remove the background and then put everything in a layer and then export them, and that would Probably take me, I don't know, 20, 22 minutes or so. But in my case, I just told Claude, hey, slice and dice, the background trim transparent pixels interesting and rename each of the icons to what they actually are and it just looked at The thing and gave a perfect name for it and just beautiful icons out the other end all nicely in a folder all exactly the same size and that was just such a good use case of using these tools Rather than having to like spend time clicking buttons. Because like being a designer is not about knowing which buttons to click. You know, it's not about the medium, not about the tools. It's about the end game of what you're trying to actually get done. Scott Tolinski I think that's such a good, because it's what you just said there is you outsourced the busy work to AI, the stuff that is obnoxious, the stuff that takes forever. You're clicking around. ([Time 0:27:58](https://share.snipd.com/snip/decfee4b-19d2-4197-83ee-69e7aae15681)) - **Use AI To Speed Work Not Replace Taste** - Designers should treat AI as a tool to speed research and busy work, not a replacement for creative judgement. - Wes recommends using AI for background removal and mundane edits while preserving human taste for final decisions. Transcript: Wes Bos I'll take the AI background removal any day. You know, that's so important for me. Scott Tolinski Yeah. I mean, those types of algorithms, people act like that's just completely unrelated, but it's not. Wes Bos Yeah. Well, I think like that one seems relatively harmless, whereas some of the like on the other side, like there's the complete brand design, you know, like I'm starting to spot like I always Think about plumbers. You see a truck driving around your neighborhood which is a plumber and you can i get you can almost eyeball which model they use to create their plumbing logo you know like you can you Can start seeing the ones that are like two years old and you can just tell what it is and now you're starting to see them and they have that like sort of like um that orange yellow haze to It um that all these like chat gpt images that get kicked out are and i think there there certainly is a space for it i think the best designers are going to figure out how to use this to create Good experience to create beautiful things but not just actually do the dumb brain stuff for you and like we've seen this over the years, right? We had clip art and then you had these like corporate blobby people on all these websites. And then we had these like pre-purchase packs. Something you may not see, but you'll probably notice it now. If you ever see like an inanimate object, you know, like a heart or a burger, and then they put legs on it. And it's like, it's like a burger that walks like the mascot for my new burger shop is a burger that walks or like the mascot for my new coffee shop is a coffee cup with legs and a little eyeballs. And it's just walking, you know, like I've, I have stickers that do exactly that. And that has become such a trope that like, you could just go online and purchase like 100 eyeballs and legs and then you can just add them to whatever it is that you want. You know, I'm a lawyer. I'm going to take the gavel and put put legs on it. But then everybody started doing that. ([Time 0:28:39](https://share.snipd.com/snip/e386ed42-2ea7-4885-be42-0f5ebfc87f47))