Kris Puckett - Becoming an AI-native designer

00:00:00We hear a lot of design leaders come onthis show and talk about the importanceof curiosity and wanting to see peoplepushing the limits of what this newtechnology can do. But what does it looklike to really be an AI native designer?That blinking carrot is this invitationfor you to just tell it like I don'tknow. I don't have any idea how youwork. Here's how I work. Like what canwe do together? And then the next thingyou know you've got like this full iOSapp that has like a bunch of LLMsintegrated into it and it's doing likemetal shaders. How do you createexperiences that are personal to you andreflect what you can bring to the tableas a designer?>> We've had quality, we've had craft, butI think what people are going to bedesperate for is more that human side ofthings, right? Like soul is going to beone of those things that is like thatit'll be the word of the year like thisyear and next year is like soul. Welcometo Dive Club. My name is Rid and this iswhere designers never stop learning.This week's episode is with my friendChris Pucket who's led design atDropbox, Mercury, and now at Stripe. Andhis story is a perfect example of whatit looks like to really lean into thismoment in time with AI. To be curious,to build your ideas, and also to findways to level up

00:01:15other people on yourteam. So Chris is going to share alittle bit about his practice at Stripe,but also goes really deep into how he'susing this new technology to improve hispersonal life as well, starting with hisnew mobile app Epilogue. The goal hadbeen to design my own and build my ownsoftware. And like I had this dream forlike 20 years to be honest. I got myfirst iBook G4 in 2005 and I thought itwas so cool that people made their ownsoftware. But you have to know how tocode and that was one thing like back incollege I was a freshman and I was likeI'm going to go learn how to code Macapps. OS 10 was the thing that we calledit then. You had to learn a bunch ofdifferent programming languages. Likeit's an object-oriented programminglanguage, which I did not understandwhat that meant when I was a freshman incollege. So, I went to Barnes & Noblelike any goober does and you're likegoing to go to the books book section.They have all these like PC books and onthe very bottom right corner with like10 books on developing Mac software. AndI didn't understand a single thing thatthey were talking about. And so I waslike, "Oh, I'll take a course. I'll takelike a CS 101 course." And it was allveryengineer, math-heavy,things that I don't naturally gravitatetowards. And I kind of convinced myselflike, "I'm not good at doing this. LikeI'm I

00:02:30This is not for me. I'll justfocus on other things."But the dream had kind of always beenthere of like, "Oh, how cool would itbe?" And then the iPhone came out andand the App Store came out and I waslike, "Iwant to be someone who makes thesethings." And so I constantly would trylike whether it was Treehouse or thecourses to learn Swift or otherprogramming languages to kind of helpget there and it just never reallyclicked. Uh and I'm not entirely surewhy, but what ended up happening thissummer, I was using one of my favoritetools called Perplexity. I've been usingit for a few years and I was using it inways that they had never really intendedit to be used, which was like I wouldcreate spaces for each book that I wasreading. I love physical books, alwaysbeen a big reader, and I love being ableto ask questions about the books. And Ibig pen and paper person and physicalthings, but then you miss out on thepower of like the digital. What ends uphappening is I fill out my notebooks andthen I forgetwhat questions I had or I wouldn'tactually follow up and look them up. So,I was actually on an airplane flying andI was reading a book and I had so manyquestions about it and I'm usingPerplexity spaces to create individuallittle libraries for the questions I'dhave. But you can't add quotes and youcan't add notes and it doesn't linktogether. And I thought it'd be so coolif like the books that I was reading,they'd

00:03:45all link in some way and I couldkind of keep track of it. Goodreadsnever really like hit it for me. Thedesign is unfortunately like not my jamand then it's not really built for whatI was trying to do either which was likea private version of me just askingquestions. And so I wanted a a way to dothat and it just kind of clicked likeClaude Code had just come out. Whywouldn't I try to do this? So I actuallystarted building an a book reading appin Cursor. I was running into so manybrick walls. Like it really wasn't atthe time built for developing iOS apps.And so Claude Code had come out and Isaw on Twitter a bunch of people likeraving about how much better it was forthis specific type of like creativeflow.So I gave it a shot. And what I ended upfinding was I could build my ideasthrough that so much more effectivelythan I ever could before and I was abletoget started on this journey. Real quickmessage and then we can jump back intoit. All right, I just had my aha momentwith paper. I'm now thinking of italmost like a shared whiteboard that Ican use alongside Claude Code. So if I'mbuilding an idea, I can tell Claude toexplore six different variations andit'll immediately start designing onPaper's canvas where I can then hop inand refine my favorite concept. And thenwhenever

00:05:00I'm ready, I can seamlesslypass my design back to Claude Code orCodex to bring it to life. Paper is thefirst fully connected canvas and it istotally changing the way that I thinkabout my design process. So if youhaven't tried it already, head todive.club/paperto get started. So Jitter just releasedimage to video and it's a pretty bigdeal. All you have to do is upload anasset and Jitter will instantly generatea short video clip from it with the helpof AI. It's an easy way to add motionand depth to static backgrounds, maybecreate little clips for brand visuals,or just in general make any scene feelmore alive with subtle movement. It'ssuper fun to play with and availabletoday. Just head to dive.club/jitterto try it out. Now on to the episode.talk to us a little bit about how yougot momentum. Like going from I want tobuild this thing and there's so manydifferent paths that you can take interms ofwhere do you start? How much do you tryto like plan it all out up front or justtake little baby steps? Like what weresome of the learnings that youencountered on this journey of goingfrom an idea that previously you wereincapable of bringing to life to, youknow, we're looking at this thing on thescreen here and it's beautiful.Ah, yeah, thanks. It actually led

00:06:15to apretty fundamental shift in my creativeprocess. And I realized how much I waslike self-limiting my own creative workby the phrase like I don't know. And Iwas constantly like waiting. I wasn'tsure what I was waiting for exactly, butthere was this like need for permission.And what I found just working with CloudCode was mostly that I don't need tohave fleshed out plans. I don't need toactually have like the perfect spec docsor PRD docs to get Cloud Code to startworking. I think probably the more hottake I would have here is that I endedup finding that Cloud Code was a reallyeffective like co- conspirator in thisprocess. Like co- creator with me. Thatthere are times that in Epilogue, theapp you're seeing here,where it would spit out a piece of UIthat I probably would not have designedon a canvas that way myself, but Iactually really liked the initialthought. And so then I could just startlike reshaping what it was doing untilit was something thatI loved and I was proud of and I thoughtlike this was more representative of thevision. So, what I ended up doing wasrealizing I could just ask questions.So, in terms of like getting started orgetting the momentum, I think it'spretty easy to get momentum going whenyou start seeing progress happen prettyquickly. But, for example, this is aI'll come back to this welcome sheet ina little bit.But, on the library grid

00:07:30that we seehere, this was really, really like funto get set up because I had never doneanything like this before. I neverworked with APIs before. I barely knewwhat an API was, to be honest.And so, in order to get an iOS app, orany app, to pull book covers from booksthat you would like to search for orread in your library, you have toconnect it to some source of informationthat is up-to-date. And in our case,it's Google Books API. And so, forexample, if you were to search for TheOdyssey, you would get a ton ofdifferent results, but I wanted thisspecific cover. And so, I'd be able tofilter through the results to find whatI'm looking for.Connecting the service, like gettingmomentum, the minute that I was able totell Cloud Code, like, "I don't know howto do this. You literally just need youto connect this service somehow." Itwould walk through the steps. Okay, youneed to go to Google Books, you need togo to the API, you need to create a key,copy the key, do the thing, let's makeit secure. It would walk me through allthose pieces.That is a huge, like, fuel in the tankfor momentum to be like, "Oh, okay. Oh,I'm doing stuff. I'm doing stuff." Andthen you see it show up on the screen.They're like, "This is wild. Like, thisis it's working. I'm literally tellingit what to do. And if I get stuck, I canliterally just say, 'I don't know. Idon't know what happened. It's broke.Can you fix it?' Like, I don't I don'tknow why this is not working." I take ascreenshot of whatever I'm looking

00:08:45atand I just say, "Now what?"My screenshot folder is is>> Yeah, it's bad. unhinged, like, un-unhinged, for sure. If anybody was evergo through it, they'd be like, "You haveso many small little pieces of UI thatdon't make sense." So, yeah, it was easyto get the momentum filled, and then theI don't know became this really wild,like, invitation for a deeperconversation of like, "If you look atthis is Xcode.If you've never looked at Xcode before,uh or even if you're used to tools likeFramer or Figma, canvas-based tools,this does not make it easy to figure outhow would youbuild your app? Like, how would you getit into simulator? Uh there's a playbutton up here. That's great, but youhave to actually know, like, where areyou targeting it to? And then, don'teven get me started on like how youactually upload an app to the App Store.On the back end, it's wild, and if Ididn't have Cloud Code walking methrough the steps the first time to doit, I would have just never been able todo this. I appreciate the honesty. And II think that's why I like this story somuch is AI in so many ways just allowsus to reach further and and take on moreambitious ideas. And like this is prettyambitious. And it's not even just aboutgetting like Google Books working. Imean even some of the design detailsthat I noticed like you have theseuh I think you're using like metalshaders even. Like talk to me a littlebit about that because

00:10:01that's somethingthat for me AI has made very interestingall of a sudden where it feels much moreattainable to have these interestingmotion and animation effects. Like forsomebody who's like I want to dosomething like that, literally no ideawhat to start. How did that part of thisprocess work? What I really wanted andlike the entire intent of this app is tohave this ambient reading companion. Andambient can be like cuz like awishy-washy word, but it's like it'sthere, but it's not there. And I kind ofwanted a lot of the UI to feel that way.And so the book covers really felt likesomething that would be a natural way tomake it feel like kind of alive and withyou.Uh And it wasn't until recently that Iwas able to get the metal shaders towork. I don't know anything about metalshaders. So what I did was I used thisparametric design technique to be ableto pull in one, I had to teach CloudCode about metal shaders. So what I did,there's a great YouTube video out aroundmetal shaders, like intro to metalshader.And I've got this video research agentthat I created that summarizes all thecopy, pulls it out, and helps me createa skill. And then there's the the Bookof Shaders, which is all the copy uhcontent in there. And so I would I readthrough all that, didn't understand mostof it. But I would copy-paste the likeselections to it to the research agentto develop the skill.

00:11:16So I taught CloudCode like here's everything aboutshaders that is available on theinternet. Like let's compile it all intoone skill. And then let's take this ideathat I have of being able to extract thebook cover colors using Color Cubeto create a mesh gradient that we applya metal shader on top of. That'sliterally like how I started thisconversation this past week with Claude.It ended up building this. And then Iwanted to be able to adjust all of theparameters. So, I can adjustlike color temperature, so make itdarker, make it more prominent. I canadjust like the origin, so like where isthe shader kind of coming in from. Andall of these pieces just help mefine-tune you know, what exactly I wantto see in the shader itself. Then I cancopy the parameter valuesand go back into Claude code and say,"Okay, I've found like the tweak that Iwant now. Go ahead and update it." Andit will update it, then I can previewit, and then we can make that like thehard-coded value.And I've also got like different presetsto test out of like what would actuallylook really good. So, we can kind of gothrough and see now I don't really likeany of these with the color palette, butthat is the general idea. And so, myprocess was to like I don't know how todo something. I don't know becomes kindof a superpower cuz I'm assuming I couldwork with Claude code to

00:12:31figure it out.LLMs are going to hallucinate likecrazy, going to make stuff up. So, I'mconstantly also rechecking itsinformation. Like I'm asking it all thetime to like grade your own plan. Didyou hallucinate? Did you pull fromaccurate sources? Just because I don'ttrust them. And anyway, I'm not kind ofyou know, trusting the AI fully is likegoing to need to check ourselves onthat, but the output I think is prettysolid. I'm really pumped about this.And another piece of this was that Iwanted to make a welcome card. Again,this is mostly for me at this point. So,when I read sometimes dad life, work,whatever,I'm behind on my reading. And so, Iwanted a way to like have a littlewelcome back card.The welcome back card was initially I'llpull upsure I've got it here somewhere. Thiswasthe initial version of the welcome backcard that I designed.So, this is like theiteration process that I would gothrough, which is tell Claude code,"Okay, I want to create a welcome backcard. I want it to have the book that Iwas most recently reading. I would loveit to have a quote that if I have anysaved quotesrelated to the book, those surface, andthen call to action, maybe some stats."So, this is what we came up withinitially, and it's using my ambientambertheme

00:13:47here. So, the idea of like Epilogis like you're by a fire, a cup of tea,a book. It's fairly like kind of warmwith the the orange amber glow to it. Itlooks good, but I kind of thought, "Wow,I just made this a cool metal metalshader system that looks really good.I'm really proud of. I would love to seethat actually on these welcome backcards themselves." So, we kind of keptrefining, and I had to change some ofthose parameters specifically for thecard, and this is what we ended up with,and I think it actually looks a lotbetter. Maybe even zooming out, sincestarting to work on this, you're now atStripe. So, having put in these reps,having, you know, leaned into the, Idon't know, like what did this processof building Epilog unlock for you as adesigner that you're now bringing intoyour practice in your day job? In theprocess of building this app and doingthis,I kind of found that like, "Oh, Iactually don't need permission." Like II can just go build my ideas. I didn'trealize how powerful that was until Istarted seeing like, "I have a app inthe App Store." Now, it's not like supersuccessful by any metrics, but it's beengreat for me. And so, like success, Iguess we could define it verydifferently, but it has taught me a lotthat I don't need permission to build myideas. I don't need to worry aboutforgiveness. Like that's a weird powerdynamic. And in terms of like the workand how it's showed up in my role,I think especially at Stripe, it'shelped

00:15:02me just kind of take a little bitmore like initiative and risks andchances. So, like I'm putting myself outthere. I'm creating these things, andwhether it's, I don't know, skills forClaude for the team, or any way I cankind of fill the gap. But mostlyinternally, it's just built a lot ofconfidence that like if I'm stuck onsomething, I don't know doesn't have tobe a door that's shut. It can be easilya handle that I just turn and like walkthrough to figure it out. And I knowthat I have like these other resourceslike Claude to kind of help me. Youmentioned the building skills for theteam, which I think is kind ofinteresting because so often in theshow, you know, what building with AIlooks like is is building, you know,maybe a one-off design tool or some kindof a creative pattern generator. Andthose are all amazing, but I I thoughtit was interesting that you pointedpermissionless building almost more onlike internal processes, ways ofworking, moving the needle in thoseways. The easy thing to use AI right nowfor is to generate ideas to buildsomething like a product, an app,website, whatever it is. And it'sI feel like it also just does reallywell on Twitter. Like there's a lot offun in it. It's fun internally. I thinkevery company that's doing a lot ofdesign stuff right now, they all havetheir internal tools they're buildingthat are like, "We built our ownprototyping tool that, you know, teamsare using." And like that's all reallycool. There are other

00:16:18massive unlocks to me for AI aroundworkflows, around like how whether it'scontent design, things that you wouldn'tsee, organization, like analysis. Sothese Claude code skills, you can I meanyou can create anything that'srepeatable pattern. You can create aClaude code skill for. And so I try tofind ways to fill in the gaps of like ifthere are people that are kind ofstruggling with time management, you cancreate a skill for that. Like there's somany different pieces of it that itdoesn't have to be super technical inorder to take advantage of Claude code.And I think that's the one thing that Iam trying to like help any team ordesigner that I've talked with or workon on this because I think we're stillso early. Sure a lot of your guests thatyou have on and I've watched a lot ofit. I mean, I feel like every episode. Ithink we are all very early in the AIphase. And we all feel like this is allvery normal. I think there's a verylarge amount of people out there thatare still like their usage of AI isWell, I talked to ChatGPT a couple oftimes or asked it some questions.And it it has completely like not evenscratched the surface of what'spossible. Whether it's building an iOSapp or building a way that you wouldorganize your day, like Cloud Code cando so much more for us. So, a greatexample, too, was like we created thisprocess for pulling out all of the tasksthat have to be done, whether it's theGoogle Docs or like all the differentpieces of

00:17:33the MCPs because our contextis so like large now. Like, if you're ina large organization, you have likeGoogle Docs, Linear tickets, if you useLinear or Jira or whatever managementsystem, all that stuff is so spread outand all the communication is so spreadout. But, you can connect it to CloudCode to have it pull and compile allthose things into a single like unifiedsource. Or even personally for me, likeI started using Linear to have CloudCode update its own statuses. So, I hadCloud Code A, create all the projects,B, take my like brain dump that I had ofhere's everything I want Epilog to do.Let's go and like build a bunch oftickets. It wrote all the tickets forme. And now I'm able to just have it gothrough the tickets every morning beforeI start my day. Like I come down, I canget it fired up on a couple of things.Then I can come back, see what's done,review it, knock it off the backlog.Like Cloud Code can do a ton of thosepieces that take a ton off of our mentalload so we can focus on the work that wereally want to be doing.Uh, and I want to help teams unlockthose pieces. All right, so let's play alittle role playing, then.Yeah, I'm coming to you. I'm a designer.I'm maybe leading design at a startup.There's like four of us. And, you know,maybe team of 40 total, you know,smaller company. Basically, my usage ofAI is I've just started making like apattern generator in Lovable,

00:18:48and I useChatGPT chats to ask questions. Andthat's about it. I don't have anythingrepeatable or system-level implemented.And honestly, I don't really even knowwhere to start. How would you help metake those first few steps? That's myAnd that's my favorite, too, cuz thelike I don't even know where to start isthe greatest starting point. And it isso simple. Like you're going to Peopleare going to hate this. You just have totell it I don't know where to start andlike ithelp you. So Claude code for example hasan ask tool question ask question tooland it is built into the system, but itwill have like if you ask it I want youto ask me a bunch of questions. I'mgoing to tell you who I am, tell youwhat I do. I want you to ask me a bunchof questions, help me figure out whereyou can help me. It will do that and itwill spit out a handful of maybe acouple generic ideas, but it willprobably spit out two or three that willswitch something on your brain of you'relike I I had no idea I could do that. Ididn't know that was a possibility.Let's explore that. And then you startdigging into it and all of a suddenyou're like 2 hours into a 5-secondquestion that you typed in and you'rebuilding out these like new systems. Andso the primitive the foundational pieceisI don't know is actually a great placeto start. I'm a design lead at astartup, got four people that you know Isupport and here's what I do, here'swhat my

00:20:03general day. Help me figure outwhere you can help me.It will change how you interact withthis tool faster than it using lovablewill. Like not a knock on lovable or anyof those other tools like they're reallyfun for creating visual artifacts andcreating the actual output, but in termsof understanding and unlocking new waysof working for you, it's starting withthat foundational piece. Uh and then ifwe wanted to get more tactical of likeother things it can do. I think a lot oftimes we don't think we've gotrepeatable patterns. We do, we're justnot great at identifying them. And sothat's another piece that I think Claudecode can help with where you're likehere's my day, here's my calendar,here's likewhere I spend most of my time. What am Imissing?The what am I missing piece is like agreat question to ask because at theircore LLMs are just pattern matching andso they are fantastic at very quicklyidentifying patterns that even thoughthe human brain is an unbelievable likethe greatest pattern matcher in history,we still have these gaps. And so you canuse AI to become this mirror that's likereflecting back at you some things thatyou might be missing about yourself andyour work and where you're at.One thing that LLMs I don't think doreally well at is like creativeexploration, but you can teach it to dodivergent thinking very quickly.

00:21:18And that is a huge unlock for teams aswell. So like, you know, the standardfor designers right is crazy eights.Like it's such a funmaybe not anymore. Maybe that's like aage I'm showing to my age, but you It'sstill cool, honestly. We knock it. Ithink it's still cool. I think it'scool. Like it plays. I mean I don't knowabout you. Like the idea is you shoulddraw like on a whiteboard or a piece ofpaper. Like if you don't know what crazyeights are, you draw like four boxes.You take one idea and then you try tolike just quickly likecreate variations through that.LLMs would actually be really good atthat. You just have to kind ofcommunicate like that's what you want todo. Um so there's things you can do tolike teach that. But I thinkfoundationally people don't love thisadvice even though it's the best advicefor them, which is if you don't knowwhere to start, you just start typing.Like that blinking carrot is thisinvitation for you to just tell it like,"I don't know. I don't have any idea howyou work. Here's how I work. Like whatcan we do together?" I guarantee youwill get some interesting ideas. And ifyou were to follow up and say, "That'scool. It's super surface-y. Let's godeeper." Then it really starts diggingin more and more. The more context yougive it about your work and what you'retrying to accomplish, the better theoutput will be. And really you just needa little bit of gas in the tank to getstarted. And then the next thing youknow, you've got like this full iOS appthat has like a bunch of LLMs integratedinto it and it's doing like metalshaders. Like that's just it's a wildentry

00:22:33point.One thing that Dive Club has madeabundantly clear to me over the lastyear is that the practice of design ischanging. And the old process of gettingfeedback just doesn't quite cut it intoday's world. That's why I'm excited toannounce that Inflight is officially inopen beta. It's the feedback tool thatI've always wanted and it's built for aworld that moves at the speed of AI. SoI can share my prototypes, give contextin video walkthroughs, and Inflightmakes it easy to get the exact feedbackthat I need to move forward, whetherit's voting on directions or maybe evengetting the green light to ship a newidea. And all of this is available in asingle link that I can drop into Slackor maybe even share with power users totest out a new prototype. I use InFlightevery day and it's totally transformedthe way that I share work. So, I'mexcited for you to try the product andif you ever want to jam about it, justemail me at rid@inflight.co.You're kind of lighting up whenever Imake you put on your consultant hat. So,I'm just going to keep leaning in hereand I'm going to give you another not ahypothetical, this is very much so mysituation and a workflow that I've beenexperimenting with more recently, whichis having Claude, maybe I have, youknow, an area over here and I'm building

00:23:49something, but maybe I want to explorevisually some permutations of acomponent or something on a canvas andactually have it kind of work on thingsin the background. And I'm I'm findingmyself in this situation where I havethis prompt pretty often where it'slike, "Hey, I have this thing. Can youjust Basic I want to say crazy eights.Like, can you just help me see thisdifferent ways? Like, I just want to seethis differently." And And it even tiesback to something that you werementioning earlier where you were like,"Maybe you don't, you know, love thatfirst thing that Claude createddesign-wise, but it helped you, youknow, see something slightly differentlyand took you in a direction you wouldn'twant to go." I am finding that as avalue proposition with AI likerepeatedly.Now, my question is, how do I invest inthat type of prompt that I keep goingback to the well on, which is like,"Hey, help me see this differently." Howdo I push past the thing that I have andand broaden my horizons for what thisspectrum of possibilities could be? Isthat an opportunity to build some kindof a skill or or just anything that Ishould be doing to A raise the ceilingfor that moment and then also B allow meto take that step more efficiently?Yeah, 100% that's a great great callout. Skills are at their core justinstructions. So, if you've seen onlinepeople talk about skill files,skills.md,

00:25:05and they're files thatAnthropic created and then now I thinkthey're a standard across the otherfrontier models like Gemini and ChatGPTand the others.And it's just a markdown file. It's justtext and all it is is it's a set ofrepeatable specific instructions thataugments what Claude code can do. Youcan call the skill up. So, for example,if I were to go into mine right now,I've got a handful of skills in here.So, I've got likeuh this liquid glass skill that I made.The liquid glass skill I had to createbecause it was a repeatable problemespecially early on in this. So, I madethe very stupid decision of making anapp that was specifically for me and Iwas really excited about iOS 26.I was like on the fence about liquidglass, to be honest, when it like wasannounced. I was mostly excited aboutsome of the foundation model things thatApple was doing. So, the on-device LLMscalled like the Apple Foundationsframework. And in order to test thatout, what the idea was, what if I coulduse that to help some of the things Iwant to do for Epilog. And so, when Istarted working on this and iOS 26 cameout, Claude code did not have access.Like all the information around liquidglass and how to do these patternsproperly was not available yet. Apple'sdeveloper

00:26:20library, Apple's likeknowledge base, is all in JavaScript.LLMs don't read JavaScript very well.So, if a page is entirely in JavaScript,LLMs have a really hard time. Like Ican't just paste the link to Apple'sdeveloper docs and it reads it. Itreally struggles. And so, what I had todo, like I kept pasting. I even think inmy clipboard there's like some liquidglass principles that I was constantlyrepasting over and over and this is oneof them.Around different shapes to use, like howto use it. And so, I was always doingthese things. When skills came out, itallowed me to offload all of theinformation that we had spent a ton offrustrating conversations together,Claude Code and I, to get the nav bar todo the liquid glass thing and to havecustom icons and all these pieces, I cannow teach it how to do that. And that'swhat a skill is. And so the skill isloaded when I mention liquid glass cuzyou can tell Claude Code in in thefoundation file, the Claude MD file, tobe like natural language associated tothings. So I don't have to always invokethe skills, I don't have to do theforward slash command to get my skillsup. I can just tell it I'm working onliquid glass and it will pull the skillup to be able to understand theinformation. So for you or for anyoneelse that's doing these repeatedpatterns, a skill is a great way to dothat. The best part

00:27:35about skills is thatyou don't actually have to know how tocreate a skill because there's a skillcreator built in. You can tell ClaudeCode, I want to make a skilland it will ask you, what do you want todo for the skill? And you will be ableto tell it so for you, your repeatedpatterns that you're trying to do andthen it will come up with a plan andthen it will output the skill and thenyou can actually have it review its ownwork. Like what would take this skill tobe like an A+ based on the intent thatyou heard me say and like give ityourself a grade and then it'll usuallycome back and it'll say like maybe a B.Almost I feel like Claude Code's likepretty humble about it. So I'm like,okay, cool. Like I don't want B work, Iwant A work. Like Claude, we have no Bplayers on my team and then it willrefactor it a little bit, give somethoughts of like what it could do that'dbe more advanced.And then you have a skill and it hotloads, which means you don't have tolike log out of Claude to use it. Theminute that the skill is installed inthe folder, it's good to go. And soanyone can do that. Like if you haverepeated patterns that you're doingoften, you can do that. If there arelike code or prompts that you want todo, so less instructions and more justlike every day I'm around this timedoing this specific prompt, you cancreate a loopthat I think is just/loop.

00:28:51And these run prompts or slash commandson a recurrent interval, so you can tellit like how often you want to dospecific things and that will run aspecific prompt for you. This isdifferent than a skill where a skill islike a set of instructions and a skillcan include prompts in order to toaccomplish the thing, but this is reallymeant for just like you have a prompt.Like my prompt for every morning isI want you to review my health metrics,look at my HRV data and tell me like howunhealthy I am and it will come back orlike tell me my energy levels and that'sa prompt that I could use on a loop if Iwant. And I've got other systems that Iuse for that, but you've opened up thenew rabbit hole now, which is more ofthe personal side of things, which Ithink is really interesting andsomething that I'm enjoying in this inthis conversation is that it's you know,we're we're we're broadening outside ofmaybe what I would typically give ananswer of in terms of what it looks liketo be an you know, like an AI nativedesigner, you know, it's like, well, no,there'sso many different opportunities that youcan leverage this stuff if you just knowwhat tools are available and and andknow how to ask the right questions. Andso you've mentioned loops. I have noloops in my life. I have really no ideawhat this could even unlock. So keepgoing a little bit more into like howare you identifying opportunities to useAI in your personal life? I just go onthese walks

00:30:06and I'm like thinking aboutmy life and I'm thinking about I'm adad. I've got two kids. Like I want tobe a good husband. I want to be a goodfather. Like I want to be in the bestpossible shape I can be. I want to likecreate everything in my head. All theselike swirling pieces andthe thing that AI has really helpedunlock is that instead of telling mywife the same repeatable things over andover and over,like, "Oh my gosh, I just wish I couldlike make an iOS app." I can walkthrough, go on a walk, brain dump, andthen ask Claude code like, "Okay, here'severything I just was I've been thinkingof in this last hour. Where can we getstarted on?" And to your point or yourquestion around like, "What does mean tobe AI native in this? I feel like it'smostly just we could take the clicheroute of like this is what AI native isright now. Like Ireally wanted a really nice blur and Iwanted a nice card open for truncationand I spent ridiculous amount of timefiguring out how I wanted this to lookeven like the more less animation tofeel ultra polished. So I mean I spent alot of time with cloud code polishingthis. That is definitely one aspect ofbeing like an AI native designer. Noteveryone in their life is going to beinteracting with this. Like chatsurfaces are only going to be more andmore prominent. Despite I think whatothers might say in the space, I reallybelieve we

00:31:22are talking right now andthat is the only way that humans outsideof like the written word mostlycommunicate. Some nonverbalcommunication, written word, but for themost part like you get together withfriends and family, you have a dinner,everyone's talking. And so I I justfundamentally think like that is theeasiest ingress for people to work withLLMs. So that's not going anywhere. SoAI native has to be beyond just like Imade a really cool shader or I made thislike dither effect that every otherperson is making. I was doing that todayand then I was like oh my gosh this islike why am I doing this? Like there's ahundred of these that are way betterthan what I'm making right now. So I'lllet someone else do that and just borrowand like give them credit. To answeryour question like I mostly thinkthrough the friction points personally.I am not unique as a person. Everyproblem that I've had has beenexperienced by thousands and thousandsand millions and billions of people inhistory. Like they're just not special.So if I'm having this problem, I've gotto figure it out and then probably otherpeople are as well. Or maybe it's notlike a huge TAM of like people that wanta reading app. There's probably otherpeople who would want something likethis and so I could share that and likethat would feel cool to put this out inthe world and see what happens. It comesdown to problem solving at the end ofthe day, you know, which iswhy I I guess again I'm like so gratefulto be a designer in today's era becauseit's just all of a sudden

00:32:37it feels likethere's whole new categories of problemsthat are worth solving. Where it's like,yeah, technically we could have solvedanything at some point, you know, but itjust wasn't worth it, you know, like itwas just too much of an investment orlike it didn't make sense. Even down tolike your permissionless building thing.I thinkone way that my brain takes that is likemaybe I am somebody that felt morepermission to build, but I also the flipside of that is like I was alwaysworried about like wasting time or notyou not doing high ROI activities. Atthe end of the day somebody's employingme.Where now all of a sudden it's like,actually I can go really, really far in3 hours to figure out if this is evenworth doing, you know? And I don't knowthat I there's not even a question inthere yet, but it's like that'sthat mental model shift.>> I I love that. I deeply connect withthat. The fact that I can experiment andexplore.So I've got Open Claw and I havelike my Mac mini and I did that and Ifeel like a little bit of a tool sayingthat, but I think I love it. Like it'sit's it's exactly like what I've alwayswanted Siri to be and more. It lives inmy message, althoughMatt Smith and I were texting today andand he got me all hooked up on Telegram,so now I'm doing that and iMessage, butit's

00:33:52not the point. The point is one ofthe challenges though is that like Iwant to get into like elite shape. I'vehad some health problems over the lastyear and like things are getting better,but I'm using my Apple Watch for all themetrics and I want to see how thesethings are resolving and how the piecesare in play and all this stuff. Biggestproblem though is that you can'tdirectly connect like my Apple Healthdata to the Mac mini because I havedifferent iCloud accounts on them cuzI'm not a lunatic and I have likeseparation of faces.So my Open Claw setup has its own AppleID and there's not really a great way toshare health care data. You can howevermake a local network and share thembetween them, but you have to have aniOS app that's running that sends thesync in the background. So,in 2 hours yesterday,with Cloud Code, I built this entire appthat does it. Uh and I'll show you thesimulator for this. I hope this builds.Anyway, the idea is like you'll see avery janky app on the screen. And whenyou see it, like it's not anything to tobe proud of, but it works and it worksfor me specifically. And that is likeall that matters. And so thepermissionless building, like Iliterally can just say start sharing andit will on my phone,

00:35:07it won't do thishere. Well, it'll ask me, yeah. So,everything that's in Apple Health thatis tracked now gets sent three times aday to my Mac mini uh on a local areanetwork and it works. And that's great.So, in terms of like the permissionlessbuilding, like I'm able to in 2 hoursbuild this Okay, I have this idea. Itell Cloud Code about it. Like, this iswhat I need to have happen.Make some linear tickets. It makeslinear tickets for me on this piece andthen I can go through, work through thebacklog. I'll triage everything, markeverything complete, and I'm on my way.This is kind of like your interface forOpen Claw? Is this like Talk me throughthe setup cuz and I guess I'll sharelike where I'm coming from is I'm theperson who has like 50 different OpenClaw bookmarks and I haven't taken asingle step.And at this point I don't even know likeis it hype? Is this something that Ishould be leaning into? How do you thinkabout it and maybe you could talk to themajority of us listening who I'massuming are also on the fence. Yeah, uhI don't think it's hype. So, Idefinitely don't go towards hyperbole.But I'm not like ultra skeptical. Itried it becauseit actually solved a huge problem I washaving. So, I would was already usingCloud Code

00:36:22locally. Every day I'd have aterminal window open. I would screenshotmy daily health briefings, all right,like health metrics, I would send it toit, and then it would analyze it, andthen it would look at my calendar, andit would look at my to-do list, and itwould say like, here's where we're attoday. But, it's all local. And if Iwanted to ask it a question aboutsomething, if I wanted it to help keeptrack of something on the go, I couldn'tdo it. And like, I could use the Claudemobile app to start a new Claude codesession remotely and try to do that, butthey didn't really mix. Not until like 2weeks ago where there was really like agreat way for us to like bridge betweenbeing on the go and there. So, when OpenClaude came out, it was like thisincredible moment of like, this iseverything I've ever wanted in anambient assistant. And like, I reallywant that. I want to be able to fire offquestions about research. I want to beable to just take things off of mybrain. I just have pen and paper, stuffeverywhere, ideas. Maybe you're likethat, too. I think the creative piece ofus is like, so many things I want to do.I can just tell my Shadowfax, that'swhat I call it, the the instance,>> it. to go build my like, here's what Iwant to do, here's what I want totackle. So, is it hype? No, it's justyou have to have maybe more measuredunderstanding like what it will do, andlike what it can do for you. But, rightnow, like I talk to it all the time tothe point where my wife will jokinglylike, hey, you should tell Shadowfax toto add this to your to-do list,

00:37:38and it'slike, oh, that's a good call, I'll dothat. But, the point is like, it'shelping me do the things I want to do,like live the life. Linear is aninterface mostly for me to tellShadowfax like, I want to build myvarious like projects, things I'mworking on. Uh, I did like a deep auditof all the things that were bothering mewith Epilogue. So, I would do a voicetranscription to Shadowfax, and then itwould create all the tickets for me. Andthen I could have my local Claudeinstancego through these tickets and startprogrammatically like knocking them offone by one. And so, we were able to, youknow, a week ago like power out all ofthese long-standing issues that I'vehad. Um, but this is the way that I cankeep track of all the work in flight,where it's at, and then also have justlike the help I need, and Shadowfax isthe one that does that like throughiMessage and so it's helping me kind oftackle all those pieces. Like you saidearlier, it does feel like we're just inthe first inning with all this kind ofstuff. And yet that little piece therewhere you could literally just be on awalk and you have ideas and executionplans. Maybe you were having a back andforth with Claude in you know, voicemode or whatever or just dumping andthen it literally is turning it intotasks and then working on your behalflike on a separate device. It's like Idon't know there's all the clichés aboutthe the 10x

00:38:53engineer and everything, butlike I get it, you know? Like the thedelta between all of you designersbecause most people listening probablyare designers, but the delta between adesigner in an org who is kind of justnot scaling yourself. I'm not even surewhat the how to even refer to it versusthe person that, you know, literallycould be doing and exploring andaccomplishing10, 100 times more. It feels like thatgap is widening. I think especially onlike the hiring manager side, too, liketalking with either like early stagefolks or people that are like lookingfor new roles and they're not engaged inthe space. Like it is a little bit of aflag. Plato has a great quote where hesays that bewarethe barrenness of a busy life. And solike I worry a little bit that the ideaof just doing more and more and more andmore and more like yes, it enables that.I do not like having agents runautonomously. I do not like having ifyou're in Claude code, the dangerouslyskip permissions world where you canlike there's a little prompt that youput in that Claude can just go and do abunch of crazy stuff and you're skippingpermissions as in like it might deleteall your files. It might. It's it can gooff the rails.But the idea behind that is people dothat so that Claude code can just likerun

00:40:08for hours and hours and hours andhours. That is for sure one mental modelabout working with LLMs is that I canjust tell you to go do all these thingsand you do it and then I just takewhatever output you have and I'm likehappy with it and I move on. But I thinkwe're seeing this even in like Twitterarticles, for example, where everyarticle starts to sound the same becausethey're all using AI to write thearticles, even if it may be likegenerically their idea,but there's this like push, I think,just to do more and everyone's kind oflike there's like anxiety around that.And I want off that cycle. I do not wantto be riding that bike race with anyoneelse because this is not how like I viewthese things. Like they are a forcemultiplier, but I want them to befocused. I want it to be something thatI feel is still like authentically me.So, that's the one thing I was reallyproud of with with Epilog is that eventhough Claude Code designed all like Ididn't use a canvas tool to design thelayouts or pixels, I just told it here'sthe hierarchy, here's the layout, let megive feedback. I would give so muchfeedback until it felt like, "Yes, thisis authentically something like Idesigned. Like I like this. I feelcomfortable putting my name on it."I had to be very involved in the loop.And so, I want designers to feel like tomotivated that these tools canaccelerate the things they're doing tohelp accelerate

00:41:24explorations, to taketasks off their plate that they're notgood at doing. Like, you could literallyconnect Claude Code to Google Docs tolike create structures of a document foryou that you fill in yourself. Like,I'm terrible at that. I'm allergic tospreadsheets. Claude Code can write andcreate spreadsheets for me now that Ican then and go fill out. Like, here'sthe generic idea of what I want thespreadsheet to accomplish. Me sittingdown for an hour trying to like figureout how to do it is not effective.Claude Code doing it is effective aslong as my content. So, you know, thepoint I'm trying to make before I likego on this crazy diatribe is that it'seasy for us right now, I think, to havelike, "Oh my gosh, we can just do allthe things." And it's like, "Yeah, butis it worth doing? And do you feel proudof the output itself?" Which is why likehaving linear or these tools kind ofkeep it in check where I can review, Ican go back and I can review each of theticketsuh were being built of like, "Okay, thisis what I wanted to accomplish, so wedid." And then it's marked as done and Ican go in and I can review the specificticket. And show me what this means inthe app. Did it accomplish what I'mwanting? Is it slower than other people?Yeah, but I feel more proud of the theoutput. Were you leaning into thisexperience or set of skills while in theinterview process at Stripe? Did youintentionally position yourself assomebody who iscomfortable

00:42:39and like really pushing whatAI can do and and bring to the table inyour role? I mean, we brought it up,like had conversations for sure. I don'tthink I shied away from the fact thatthere are people in this space who areso much better and smarter at this stuffthan I am. And like I'm well aware oflike, I feel like I'm ahead, but Iprobably couldn't articulate to you myown skill set very well. So, I tried tobe very cautious about like, I love thisstuff. The last 10 months using thesetools, especially cloud code, haveunlocked a whole new creative way ofworking for me that actually worksbetter for me than a canvas does. Andso, I definitely would articulate that.I would highlight like my own thoughtsand methodology and process. And I havea ton of opinions on like how AI nativeexperiences should be built and likewhere we should be indexing and movingtowards. And currently right now, MCPsare they're not going anywhere. Anyonewho says an MCP is dead is like, that'shype. Like that's crazy. But, they'regoing to change quite a bit and we'regoing to see a lot more companiesleaning more and towards better semanticAPIs and CLIs than they will an MCP,which is a lot of acronyms thrown out ifyou don't know what the space is.Essentially, like MCP is a way of justto talk among different layers, but ittakes a lot of information

00:43:54like back andforth to accomplish a goal. So, like thecontext window, the window that the AIcan hold and think gets filled up veryquickly with an MCP. And there's moreefficient ways of communicating attimes. And so, I think the models aregoing to start to get more efficient oflike which tool that they use and whento do things. So that affects howcompanies are going to design theirproducts. And like do your products havean agentic focus to them? Or iseverything readable by an agent becausethere's going to be a lot more agentsdoing the work for people. So you haveto like think through that as adesigner. So I definitely would positionmyself like in the you knowconversations whether it was Stripe or afew others I was talking with aroundwhere I was at with my skill set orwhere I thought the industry was going.Whether or not I'm right or I don'tthink it really matters whether you'reright honestly because the thing thatthe flip side like the hiring managerskeep coming on and it's like they say,"Well, curiosity is the most importantthing." I mean I I've heard like 20people say that over the last 6 months.I I I don't even ask the questionsometimes now because it's like, "Well,I already know the answer is curiosity."And it's just like, "What does thatactually look like? How do youdemonstrate curiosity?" And to me a lotof what you're showing, I mean this iskind of that, right? Like if I'm hiringsomebody and you're walking me throughthings that maybe 6 months ago Iwouldn't even have considered in the

00:45:09boxofdesign, you know? But hardly evenadjacent to design. But now all of asudden it's like everybody's becoming ageneralist, everybody's wearing all thehats, and people wantemployees who can come in and just makean impact and see problems and solvethem and reach for any combination oftools to solve problems. Like we're notjust solving problems on a canvasanymore. We're maybe not even solvingproblems in code anymore. We're solvingproblems at like systems levels andpositioning in the marketplace. Youknow, it's like this is curiosity andthat's kind of why I was wanting to askbecause I think it's getting a littlebit more difficult to put this into abox of like, "Here's how to prepare foryour portfolio," you know? But if you asa designer can show something like this,like it's a pretty big deal in today'smarket, you know? Like there's a windowin time toget a serious leg up by being able tothink and act like this. Yeah, I 100%. Ithink especially if you're a designerlooking for a role right now, which isnot what you brought up like kind oftangentially a little bit like if you'reinterviewing stuff, I think theconfusion that some people have, they'relike, "Well, I haven't I haven't builtan app or anything with with AI yet."I'm like, "You don't have tonecessarily. You can if you want to, butonly like if you're proud of what youmake in there, but you have todemonstrate

00:46:24that you understand thesesystems and you're like you're actuallybuilding something with them. Whether itis like another shader plugin system,just do something with it to see andexplore and be able to to talk to it andlike talk through the process. I wouldsay the thing that is like subtle amonghiring managers right now, you look atso many portfolios. You maybe have 6 to10 seconds to look at a portfolio truly.Like you you know, I know people who arehearing that don't love that cuz youspend all this time into it, but if yoursite is on Squarespace and it's a basicSquarespace template, you're probablygoing to get passed on even if thecontent's really good. Because likethere there is just not a reality inthis world right now where you could nothave Claude code, ChatGPT, Gemini workwith you quickly to have a more customversion of your own template. There's alot of reasons why someone would just doa a basic Webflow template or whateverit is and that's like not a knock onthese companies at all. It's more of astatement of like we're moving so fast.I hate saying this like the bar is sohigh. Like everyone wants the same typeof designer, the same shape of designerthat if you're not using these tools todifferentiate yourself a little bit likeyou are going to get left behind andthat sucks. Like it's a weird thing toto view and see, but if you're resistantto change, it is really tough. Socuriosity for sure I could see it beinglike a little

00:47:40bit of a cliche andthey're probably trying to say the nicething, which is like if you're notshowing that you're invested in wherethe space is going, it's probably a no.There's probably going to be companiesthat would be a good fit for that, butall of the the larger tech companiesright now, they really want to seepeople that areengaged in the space. And there's justlike again going back to thepermissionless building, you don't needpermission to do it. You literally Idon't know is a superpower. Like youdon't know how you don't have to knowhow to create your own website. You canliterally ask it like I don't even knowabout hosting. Who should I use forhosting? And it will spit out a bunch ofdifferent options that are great, youknow, oh like I can get you set up onVercel right now and you can have a freewebsite and it would take us like maybean hour and you have like a staticlanding page that looks really good. Andyou're like, okay, let's do that. Let'stry that. And just see what happens.Like that would be far better than justusing the stock templates. Everythingthat you're showing today is such aexpression of who you are, you know?Like all the things that you'rebuilding, it's for you. It's systemsthat are like deeply revolved aroundlike your willingness and excitedness tolike improve and and I grow as like aperson in your career, all that kind ofstuff. So it's like I don't know. Youryour website, the side projects, it'sthe billboard of who you are, you know?Like it's not about a case

00:48:55study or yourwork anymore. It's like who are you, youknow? What do you want to put forward?And you can get a prettyyeah, we do judge books by their cover,you know? That's the facts. I'll sharethis cuz you mentioned the website andthis is like not even true, but I Iliterally went through this and I foundthis WebGL shader that I really like,this ocean shader. So you can like movearound.>> Oh, cool. And it'sresponds to the time of day. So the sunis in the correct position for Denverbased on the time of day based on likemy IP address.And then I just have like, you know,little writing, something about me. I'malso not like looking for a role cuz Ireally am like stupidly grateful for thecompany I work for. So this is just youcan ask it questions about me if youreally wanted. And then I did likereally silly things like I had a chanceto like, oh, what would it look like toadd rain to it? And like what if it wasat night? Uh and so but I now want thewindows to like respond to the the rain,you know? So I don't know how to dothat. Claude does it. What I have now,purpose of sharing this is not like a tomy own horn, it's like I could not havedone this on the previous platformwithout a ton a ton a ton of customcomponents that I did not want to designmyself. And I don't think I would havehad the output as good, but I had thevision for it. Also,

00:50:10I'm saving like$500 a year on the site and the thing,which is great for me. I'm pretty stokedabout that piece. But most importantly,I have something that feels morepersonal to me. Because I do think likeI've been saying this for for a littlewhile, like soul is going to be one ofthose things that is like it'll be theword of the year like this year and nextyear is like soul. We've had quality,we've had craft, but I think what peopleare going to be desperate for is more ofthat human side of things, right?They're going to be longing for thiseither they an era they've neverexperienced cuz they're younger of likethat MySpace, you know, our MySpacegeneration where you're like yourMySpace page was deeply personal to you.Like in 2004 and three, five, like myMySpace page was complete custom ChrisPocket perfection at that time.And I think that we're going to want tosee that come back. And I think peopleare going to want more of those likeyour portfolio looks and feels like you.And that would be the hope, right? Withwhat I'm trying to design. So, I there'snot a reason not to have that. So,differentiator is also like are you init or does it look like everything else?Are you using Instrument Serif in allyour stuff because everyone else isdoing that? Like there's so muchpersonality that each of us has that Iwould love to see that reflected intheir design. I'm I'm on like such astupid soap box. I'll get off. No, it'sperfect. I I honestly like

00:51:25I can't thinkof a better place to end it, honestly.Like it's super inspirational and you'relike the perfect example of what itlooks like to just run at all of thesenew possibilities and tap into like thecreativity that's obviously been insideof you and all of a sudden she's beenunleashed and you're just firing inevery direction. It's beautiful to watchand Chris, I really really appreciateyou even just demystifying a lot ofthings for me personally. I'm interestedin a lot of the things that you'retalking about. I know there's stuffthere. I know there's more I can grow. Iknow I can go further into skills andall these different personal systems.And I just having somebody kind of breakit down has really been super helpfulfor me even. So, appreciate you takingthe time today. And Red, thanks so much.It's been an honor to be on the thepodcast. This is like my second podcastof all time. So, beyond grateful tofinally be here with you. Appreciate thetime.Before I let you go, I want to take just1 minute to run you through my favoriteproducts because I'm constantly askedwhat's in my stack. Framer is how Ibuild websites. Gen way is how I doresearch. Granola is how I take notesduring crit. Jitter is how I animate mydesigns. Lovable is how I build my ideasin code. Mobbin is how I find designinspiration. Paper is how I design likea creative. And Raycast

00:52:40is my shortcutevery step of the way. Now, I'vehand-selected these companies so that Ican do these episodes full-time. So, byfar, the number one way to support theshow is to check them out. You can findthe full list at dive.club/partners.