Meaghan Choi - Designing Claude Code (and what's coming next)

00:00:00Ever wonder what it's like as the firstdesigner working on a generationalproduct like Claude Code?>> I DM'd my manager. I was like, I thinkthis is going to be big. [music] I wantto design Moon with Light for this.>> What does it take to design these typesof AI experiences?>> There's a huge school of thought thatdesign is all around owning the qualityand polish of everything. I think it'sreally important to have that, but itdoesn't have to be all of your time.Knowing when to [music] use these toolsfor execution and offload and knowingwhen you need to go deep and do thethinky work is a skill [music] set and Ithink that's one that I see people doingwrong.>> How are anthropics new products changingthe way that we think [music] aboutbuilding software?>> In the same way that it takes a lot toconvince a non-engineer to download aCLI tool and need the terminal to beworking and yet we've seen a lot ofpeople do it now. I think has a similarkind of leap. The way to look at it isit's an entire paradigm shift [music] inhow you work with claude. Welcome toDive Club. My name is Rid and this iswhere designers never stop learning.[music] This week's episode is withMegan Choy, who's the design lead forClaude Code and is pretty quicklybecoming one of my favorite people inthe industry to talk to. And it alsokind of feels like the way that designoperates at Enthropic [music] isconsistently a few months ahead of therest of the industry.

00:01:15So, there's a lotto learn from in this conversations anda few things that surprised me as well.So, we're going to dig into her designprocess, the way that Megan usesartifacts, and all of the ways that themodern shape of designer is evolving.But before we get into that, let's startat the beginning of Megan's anthropicjourney.>> So, I joined late 2024. This is backwhen no one even knew outside of likeprobably the core tech industry whatanthropic was. Like, everyone was sofocused on chat GBT and OpenAI at thetime. And Enthropic was primarily an APIbusiness. It didn't really have aconsumer arm. was like a clawed app, butit just hadn't had its takeoff momentyet. And so when I joined, I rememberpeople ask me, "Where are you going?"I'm like, "Oh, I'm going to Anthropic."They're like, "What's what's that?>> What's that?>> What's that? [laughter] Never neverheard of her." I was really excitedbecause I spent a lot of my careerworking on developer platforms and inemerging tech specifically. And I justfeel like when you are building in likethe tech research area or like new tech,developers are the most innovative andthe most critical of everything thatyou'll ever build. So you get to seelike the real use cases live of whatyou're inventing. Maybe about 3 or 4months into my time at Anthropic,someone had like demoed this like weirdCLI pipeline of having Claude

00:02:30edit yourcode directly. Took like 45 minutes. Itwas so slow. It required thisextraordinarily complex setup of likesetting up remote workspace, downloadingall these scripts, like doing all thislike local dev that was extremely heavy.and then it could kind of sometimeswrite good code sort of. But I rememberseeing I was like, "Holy crap, that's socool. I gotta try it." So I did and Igot in touch with the people who weredeming it, which was Boris and Adam andCat and a few other folks internally.They were doing it as an experimentalpod. They didn't have a designer becausethey're like, "We don't need a design onthe CLI." I I DM'd my manager. I waslike, "I think this is going to be big.I want to design Moonlight for this." Mymanager's like, "Sure, go ahead.>> Cool.>> Go do it." And that's kind of how I gotstarted. and it never stopped.>> I love that the expectation would belike, "Well, we don't even need adesigner. It's a CLI." Which I thinkactually probably would have been myreaction even like a year and a halfago. It wouldn't have been obvious to mewhere a designer would slot in. So,where did you slot in? Like what weresome of the things that you even werelooking for in terms of like thechallenges or opportunities that wereunique to designing within the terminal.>> I think there's actually a lot of designof the terminal that's like theinvisible thinking that goes behindbuilding a product. That's one that Ithink I expect and hire for at Anthropicand

00:03:46I think has been the long unsunghero of product designers everywhere islike the mental model, the interactionmodel, like the base primitives thatcompose how you think about what you'reinteracting with. So there's a lot ofthat that goes into it. And then theultimate form factor of the CLI thatmade it usable was a chat. So you therewas so much to design in that back andforth in between the chats that made itfeel more warm, more delightful thatcommunicated the statefulness of it. Ihave actually designed for CLI before.This wasn't my first time doing it.That's why I knew that it would need thehelp and I was like wow this is going tobe fun. Like the challenges andconstraints of designing for CLI make ita very innovative space to playactually.>> Okay, let's talk about that then becauseI'm sure there were a ton of evolutionsthat were happening as you're gettingusage and playing with internally. Sohelp us understand what design lookslike at a CLI level. Like what were someof those iterations that you were doingand the role that design played?>> So I would say some of them are justlike how you communicate the toolcalling and statuses. Like it's theexact same kind of design thinking thatyou would have on a UI except theverticalizationof how messages stream. You can't scrollor jump back in the same way. Likeeverything scrolls together in a CLI. Soas things render in, you have to really

00:05:01think about where it's loading, how muchinformation that you're showing. Becauseyou're serving developers, they havevery specific expectations with the CLIactually that you want it to beextraordinarily information dense. Likeyou would show more in a CLI than youwould ever probably show in like a guey,like a graphical interface, becausethere's no such thing as progressivedisclosure. So it's just disclosed,especially in the early days when themodels weren't as good. We're like, "Oh,we got to show people what it's doingbecause we're not sure if it's going togo off [clears throat] the rails alittle bit." And then keyboard shortcutsare massive. I would say like there arepower tools like Figma and other placeswhere they do consider keyboard heavierbut like the primary way you interactwith a terminal is keyboard shortcuts. Idistinctly remember we had this hugedebate for a very very long time betweenmyself and some of the engineers aboutwhat the shortcut should be for modes.Like if you've ever done plan mode orauto mode, what we call it and then whatthe keyboard shortcut is and if we haveto change the UI so that you know whatmode you're in. There was a fewiterations there, but the biggest debatewas actually the keyboard shortcut to doit because it needed to be fast, but itcouldn't overlap with any existingshortcuts in the terminal, which aremany. [music]>> Real quick message, and then we can jumpback into it. So, I've been looking at alot of portfolios through the talentnetwork lately, and many of the best[music] ones all have one thing incommon. The

00:06:16designers are using Jitterto bring their work to life with motion,which is by far the easiest way to giveyour designs that premium feel. AndJitter makes it simple. There are nocomplex key frames that you get in AfterEffects or Figma or other features thatyou don't need. It's just a reallyintuitive product that I've been usingfor years to animate my work and Icannot recommend it enough. You can getstarted today. Just head todive.comclub/jitter.I've anticipated Desen's [music] recentrelease for a full year. It's calledsurfaces and it enables you to designand prototype directly on top of anexisting production interface. All thepages and flows that you need arepreloaded inside of DES [music] asstarting points so you can iterate onthem with your team, but also run usertesting or inspect [music] states orcheck component properties. No othertool renders your actual productvisually in the cloud like this, whichis a pretty big deal for Teams. And youcan get started today. Just head todive.club/desen.[music]That's d ssn. Now on to the episode.You're building in a space where there'ssimultaneously so much muscle memory,but then the core concept of what claudecode was

00:07:31wasfundamentally novel. Probably evenstretched some engineers in thebeginning. Like were there mindsetshifts that you were designing for inthose early days? Like how do you evenget people comfortable with what a CLIcould do? I think the biggest hurdle toovercome was making it feel like a chatyou could go back and forth with. CLIkind of have that where you like printand then it like you send a command, itgives an output, sends a command andgives an output. And so we leaned intothat a lot to like anchor the paradigm.But I think the fundamental shift wasactually it reading your entire filebase and then writing to it as well. Andthose are like the two big mechanicaldifferences that helped make claude codeeffective. Like I think the most magicalmoment that you always see people havewith cloud code is when it actuallymakes an edit for you with the contextof everything that it knows about whereyou're working already. And you have toremember this is back in the era whereeveryone was like copying and pastingparts of their files into like a chatand then copying and pasting the outputback into their files. And we all knewthe pain of that. So seeing it happen infront of you without having to do thatis like the moment that people realizelike holy crap this is somethingamazing. Like this is going to changehow we work. And so we were optimizing alot for that first feeling. Actually,>> the first feeling for me was

00:08:46I rememberI was watching Dan Hollik use cla codeand he was the one who introduced me toit. I don't even remember when this beenmaybe last fall or something like thatand you you type Claude and then you getthat full screen and you have the giantorange box. That broke my brain a littlebit cuz I I had never even my I had noCLI experience at all and seeing that orI'm like immediately whoa what'shappening like where am I? This is verydifferent. I think that was actuallyprobably the most fun thing I everdesigned in the Claude experience. TheClaude mascot, beloved, fun, incredible,generated by Claude actually, which Ithink is a really nice touch. I thinkit's really important that you feel alittle bit of warmth from these tools.You don't want to be super superdetached from them. And also, I thinklike Claude on its own has a very uniquepersonality that makes it delightful towork with in my opinion. And so, Iwanted to really represent thatvisually. I love AIE art. I think it'sso fun. There was this deep period ofNFO files or info files if you rememberback in the era of software developmentwhere like you would have to downloadall these you would have to script todownload a lot of apps and a lot of themwould include these like incrediblyartistically drawn asy arts like thingsthat you couldn't even imagine. And atthe time I was

00:10:01like yeah I want to bringthis into our app. Like I think thiswould be really really fun. And we did aton of exploration with the brand teamactually on like all the backgroundsthat you see now, like all theanimations that you see now to reallylike make a world that feels like you'rein the computer with Claude. And I thinkthat's the vibe that we were trying togo for.>> It reminded me of the old like aim awaymessages. Did you ever do those where Iwould do the same thing? I would takethe blocks and I would make littlescrolling patterns and like artdrawings. So it was like this was mylittle home was just my little dots onmy aim away message. Okay, I want tokeep drilling into your reality assomeone who is designing but isn'tslinging pixels around nearly as much asI have, at least in in previous roles inmy career. And another release thatstood out to me was this artifactsrelease where you kind of break it downand you realize, well, okay, the inputscan be anything and the outputs are likecompletely non-deterministic.What is the role of design in a featureprocess and and release like that? Sogive us a little bit of a behind thescenes for Artifacts and what that waslike.>> I think there was a really strong kindof emerging culture of people wantingmore than just a transcript as anoutput, like something sharable,something visual. We've like learned

00:11:16alot over the years that it's not justlike text that people want, but there'slike a ritual visual. You know, we'reall designers, so we understand that.And Claude is really good at coding,like excellent at coding. And so as thatis emerging, there was a pattern offolks and I used to do this in myworkflow as well of constantly askingClaude to generate HTML files for thembecause it's fast, it's snappy, it getssomething like put together. Claus'sreally good at making it and then it'sreally quick to iterate on top ofactually. And if done right, you canalso share it>> which was a big part of it. We wouldstart earlier starting to takescreenshots of these files afterwardsjust like send a HTML file to each otherso you could review it.>> I text HTML files way too much>> which is crazy. Like it's crazy thatwe're sending each other HTML files.It's actually insanity. It's like>> I didn't even know you could. I was likeI literally asked Cloud I was like Ihave this HTML file. I was like I justdidn't even think that iMessagesupported it. Just like what are youtalking about? Drag it into iMessage.And I didn't hit enter. I'm like no way.I do it all the time.>> Right. Right. But if you're sendingsomeone an HTML file, it's like didn'twe invent the internet? Like why are wewhy are we [laughter] why are we doingthis right now? And maybe like this kindof highlights the workflow. It's likewow a lot of people were doing it. I wasdoing it. Can we make this somethingthat's easier to use? You don't have toask cloud to do it. So cloud does itautomatically. Oh, what are we doingwith it afterwards? actually want tosend it to someone or share it withsomeone. I want it to be live anditerable. So as we were using it, as webuild it out, we start to uncover the

00:12:31features that we want out of we're usingit and we build it into a product. We'relike, I most commonly use these forprototypes and when I'm iterating ondesign. Our data scientists mostcommonly use it to make dashboards. Ourengineers use it to make text specs orto visualize PRs or do research ordesign docs for the code bases. And thenyou want to send it to someone. So let'shost these. Let's make them sharable sothat anyone can access them. Oh, Iactually want to be able to send anolder version but keep working on it.Oh, let's add version control for theseso that you can share and pin a specificversion or you can share the latestversion so they're live. So, a lot ofthe way that we kind of iterate on theseis we build for ourselves. We test forthe broader adoption and then we addfeatures that we find useful and it endsup being useful for a lot of people orwe get feedback from other folks on whatis useful to them and we build it in andwe end up with a product. And theinteresting part of that is some peoplemight not consider it traditionallydesign, but I do like what is the shapeof this. It wasn't called artifacts tobegin with. We weren't even sure what itwould be. Like we were just sending HTMLthings. We didn't really care about thename or the packaging at the beginning.We cared about the workflow. And thedeeper we went into it, the deeper we'relike, how do we fit this into someone'smental model? So it's not something thatthey have to learn. It's just somethingthey get or have or is it important forpeople to learn what it is so they knowhow to use it? And those

00:13:46workflows arereally where the design comes in. The UIpart of itself is extremely simplebecause in my goal, we're trying not toinvent new UI. We want it to feel sonatural that it just becomesintrinsically part of your workflow.There's a designer here on the cloudcode team whose manure is excellent.He's like, we're designing tools forpeople to get work done. The moment younotice the tool is the moment we failed.Like it should be totally out of yourway.>> And I think that's a great way to designproductivity tools.>> Let's help people imagine how they coulduse artifacts. So, can you talk a littlebit about you mentioned sharingprototypes, what's that workflow looklike for you? what are you trying toaccomplish? So, I'd say like a standardway I would use artifacts is I'll get afeature request from a team for likesome design support and as opposed to medoing any design work, I'll just getClaude to generate an artifact or I'llask them to generate an artifact of whatthey should be or sometimes I'll justtell them to use cloud design togenerate that as well and then to sendit to me and I'll review it and then Iwill get my cloud to iterate on thatartifact if it needs any feedback orI'll DM them the feedback directly andthen we kind of have a live version oflike where we started. and where we'retrying to go. And then Claude canactually ship that final version. Andbecause it's baked into your codebase,once it feels mid-fidelity, I actuallyask Claude to build it based off of ouractual components. And so it no longerbecomes just an HTML

00:15:02artifact, but asharable prototype that then we checkinto the repository as well. And so itlike grows as we get more sophisticatedin a feature that then eventuallybecomes a pull request. Is there like afidelity cap on what like you know is itan approximation of what you're tryingto build? Is that like the ceiling whereit's more useful early stage or like I'mtrying to figure out exactly how thatwould fit into I'm actually trying toship something. I think the benefit ofit being in an HTML artifact is thatit's so fast to iterate on and whenyou're in the early stages, you reallywant to get those iteration loops in.And then once you have the right shapeof it, I think it's easy to switch froman artifact because Claude is aware ofit into a PR. That's kind of thetransition I go for. So we're going fromlike early ideation and feedback to adecision. Make it a PR so that you canjust ship it when you feel like it'sready.>> The way that I would send Figma files,like it just didn't really work. like ifI'm trying to send it to like an execmember, I could never get them to gointo the canvas or anything like that.But now I'm like I could have createdit's not even a prototype. It's almostlike an interactive report that includesa prototype.>> Exactly.>> I'm like, man, could you even builddifferent ways for them to like expresswhat they like or different intentpieces and then capture all of that?

00:16:17like it's kind of exploding what I thinkabout this workflow in terms of thedesigner's job of visualizing ideas,[snorts] sourcing opinions and and uhideas from the rest of the team and thenlike tightening that loop as much aspossible. I can see how this would be abig big tool.>> It's almost like a mini micro site anddecision log in itself. Like often Iwill ask Claude to actually pull thedata or pull any research I have fromlike our bitquery, Slack, any docs thatwe have on this and like summarize it atthe top and then put a goal that we'retrying to accomplish. Turn out likethree to four versions of the design. Ifit's more complex, I'll ask it to breakout actually the different work streamsthat we need to work on and thenproposals for there. And then I will askitforit. The interesting thing is I oftenhave multiple artifacts in one of these.Like I'll have an overview one and thenI'll have side ones for all theiterations on the overview one I want todo and then I tell it to send it allback to the main artifact and that mainart.>> Are they separate links or is it like atable of contents almost?>> They're separate links but they can linkto each other if you ask cloud. And soI'll ask Cloud like, "Hey, like I wantto do a deep dive on like the entrypoint. Let's say let's like prototypefive different entry points. I likeoption

00:17:32A." Mark that I chose option Aand leave a link there to the oldartifact. So if anyone's curious that Idid the exploration, they can like seethe decision.>> Wa dang, that's [clears throat] crazy.This is like a whole set of designchallenges in itself in terms of justthinking about the artifact journey thatyou want to create for people. And Ithink the beautiful thing aboutartifacts and probably the challengingpart of it is that it's so open-endedlike because it's truly anything. It canbe anything that you want it to be.>> My brain's racing right now. I'm likeI'm like I'm like 80% having thisconversation and then 20% thinking aboutwhat I'm going to do right afterwardsbecause something I was doing literallylast night was I am building on top ofthe GitHub API and it's little I knowmaybe a good chunk of the metadata thatI would be able to have available fromthat but there's so much that I don'tknow too. So I just sent Claude I waslike go do a research report effectivelyand given this set of goals what is allof the re relevant metadata that Ishould consider using but I didn't wantthat as a block of text. So I had itmake it into HTML and then I had itbuild like each piece of metadata as amulti select so I could just click andthen get like um okay like you knowthese are the seven things that I'veselected and can feed that back ascontext into cloud. What I really wantedto do was to send that to two otherpeople

00:18:47and be like,>> go through and let me know which piecesof metadata speak to you, like whatwould be the most compelling thing thatyou would want to see on a user profile.And then to be able to like compare theresults and then have like a versionhistory within that>> and yeah, very limiting when you'redoing like the standalone HTML, but Idid that wrong. Like I should have[laughter] been using artifacts forthis.>> Yeah. And it makes them just likeinfinitely more capable and sharable. Ithink the sharing is really the big partand like we're going to layer more ontop of it. We're like so early in thisjourney and a lot more is coming. So youcan just imagine all the things that youreally feel like you want to do, we alsofeel like we want to do and it's kind ofthere. Like we're thinking about it.>> That's the smirk of somebody who alreadyhas the next version built, butdefinitely cannot talk about it yet.>> Yeah, it's exciting. I I really like it.You know, I never thought we'd go backto HTML as like the main form factor forcode and the main framework we use, butit's pretty powerful. It's pretty great.Speed means [music] a lot.>> There's one question that I can't stopasking myself. What if companies appliedto talk to you rather than the other wayaround? And that question is thefoundation for the all-new dive talentnetwork and [music] it's working. Likeright now, I'm helping many of the mostexciting startups that I know to hirethe designers and builders who listen tothis show. So, if you're curious whatmight be out there, and

00:20:02maybe you wantto get on my list, or maybe you're evenlooking for your next design hire, headto dive.club/talent club/talent to jointoday. Okay. So, we talked about likethe social layer of artifacts a littlebit, but I'm assuming there's more onthe bone in terms of how [music] youwork specifically in the ways thatyou're collaborating with the models.You've been there, I mean like a yearand a half. We were joking before thisabout how it feels like we're aging indog years. So, like a year and a halfworth of reps brainstorming with Claudeis almost an eternity at this point. Andso when you're working on maybe earlystage more ambiguous meaty problemspaces, what are some of the ways thatyou've evolved how you collaborate withthe models?>> Man, I am like constantly having bothexistential personal kind of big brainthinky discussions with my claud andthen also very tactical execution work.And the beauty of how we now work hereis that it's all together. So I don'treally have to code switch in betweenthose two which is I think our old wayof working. It's all kind of built intogether and I have like one clawedchannel that's like my regurgitatedbrain in Slack. I actually do it all inSlack. It's crazy. It might be like acrazy workflow but most of my workSlack. Yeah. Yeah.>> I didn't see that coming.[clears throat]>> Yeah. If you

00:21:17see the claw tag launch andI think this is like a big one that uhwe can chat about a little bit more. Alot of my both thinking and executionwork is all done there and I'mgenerating so much ideas and I'm goingback and forth with cloud at a highlevel and a lower level of execution.Okay, pause. Let's talk about cloud tagthen because I did not expect you to beinterfacing predominantly in Slack. Sogive us a little bit of context for whatthe heck that is. I remember seeing iton Twitter, but it didn't immediatelylike it was clear that it killed a bunchof startups, but it wasn't immediatelyclear what it unlocked. So why is that abig deal? and and why do you think thatdesigners going to be using that more?Yeah. So, I think Claude TEG is a reallyhard concept to wrap your head aroundtransparently. Like, we had a lot ofchallenges internally and we're stilltrying to work through exactly how toexplain it to people cuz it's one ofthose products somewhat similarly toClaude Code that I think you have to usein order to really feel, but it takes alittle bit of hurdle to get to a pointof making it useful. In the same waythat it takes a lot to convince anon-engineer to download a CLI tool andbeing the terminal to be working and yetwe've seen a lot of people do it now. Ithink claude tag has a similar kind ofleap. The way to look at it is it's anentire paradigm shift in how you workwith claude. So today you're workingwith claude as a tool. You're giving itdirection and then

00:22:32it's following thatdirection giving you a feedback. It'slike in and out in and out in and out.And it's all single player. It's alljust you and it's all very sessionbased. like you spin up one claw, youspin up another cloud, they're kind ofseparate from each other. Yes, there islike empty files, but it's a little bitlike it still feels like everything'squite separated. And as you mentioned,sharing is a problem. What claude tagintroduces is you're not working with abunch of different clouds. You'reworking with claude, a single claactually. And not just you working witha single claude, but everyone in yourorganization is working with thatclaude. Man, how do I describe this? Ithink it's hard because like the humanmind is limited to like what we cancomprehend, but this is like a claw thatworks differently than a human. Like yougot kind of got to let yourself takethat leap and realize like oh claw canhave thousands of simultaneousconversations while you as a human canprobably have four or five.>> And so it's having those thousands ofsimultaneous conversations and it's alloperating on a shared or knowledge and ashared memory layer. It also has accessto its own set of tools as opposed tobeing limited to your own user MCPs.your organization can give cloud its owncredentials so that it can have its ownGitHub access, its own like Googleaccess, its own like asauna access andso it can take actions as it needs to beconfigured as a member of yourorganization. And so what that lookslike

00:23:48is as I'm working with claude onone feature, it will pull in all theknowledge of all the engineers workingon it in a different channel or in adifferent documentation or in adifferent codebase. all the PMs that goto market, like all the other piecesthat are organizationally happening, itknows about, but it also knows that I'mworking on the design side of it. So,when we're riffing together, it's givingme feedback and direction on what designcould look like and how it relates toall the other pieces that are going onin the organization as well without needme needing to be in like 30 differentmeetings for it to happen. I just likehold all this in my head like Claude'sholding it in my head for me. And it'sso proactive as well in a way that ourexisting Clouds are. it like runsinfinitely for everyone. And so asupdates are happening in other channels,sometimes I'm talking it's like, "Oh,hey, so and so just made a decision hereabout something you were asking about.This is how you can update your thinkingor this is something that you shouldtalk to them about [clears throat]because it's slightly different from howyou see it.">> And so it just like makes it so muchfaster to have these like bigdiscussions. And then on the ship sideof things, because it's so proactive,because it's so integrated, a lot of myPRs are written actually through Slackright now. Is that a crazy thing to say?>> That's crazy. Yeah. I [laughter] didn'tcrazy thing to say.>> I can actually share I put together likesome screenshots of what it looks likelike some examples that I copied andpasted with a little bit of redacting

00:25:03into what some of my conversations withtag look like.>> We had a lot of discussion on thisbecause>> it is a new mental model. We don't wantyou to think that this cloud which isuser segmented is the same as like>> this organizational cloud because it'sso open like there's a little bit there.Like we want people to shift how they'rethinking about how they work with thiscloud. The crazy part is we use cloudtag to build cloud tag so it's verymeta. Like I was using claude tag toupdate cloud tag and it was it's awarethat it's cla tag. I've had some crazyconversations with cloud where I'm likedo you know your cloud like what do youthink this should be because you are clatag [laughter]>> and then it'll go through the crazyit'll go through its transcripts of likeanthropic or ants that's what we callourselves using cloud tag and be like ohlike this is what I think it should bebecause this is all the feedback I'mgetting in the feedback channel aboutmyself that it's managing.>> Dang, that's trippy. It's so crazy. Butthis is an example of me. I have a MeganClaude channel. I keep it open becauseone really beautiful thing about this isactually you learn a lot from seeing howother people are working with ourclouds. Sometimes I'll say like, "HeyClaude, did you see what Boris was doingin his channel? I want that. Give me theexact same thing for everything." So,it's like very social in that way. Iknow there's a lot of challenges of likesharing skills and workflows. This isone where you can literally just say youcan see someone else's workflow and belike, "I want it." And because Claudedid that workflow there,>> it already knows it. It's it'll

00:26:18justbring it to you. You just have to ask itto do it.>> Yeah. You're not even like pasting ascreenshot or anything. You canliterally just reference any otherconversation.>> Yeah, I'll copy and paste the like DMlink or the like open link. It'll belike, I want this. Can you do this forme, please?>> Dang, that's cool.>> So, this is a message I sent to Claude.This is an example of like some of thenitty-gritty UI details that I wantfixed that I now just have Claude do.So, I was like, hey, there was a dropshadow there. There wasn't meant to beone. Remove it. Right. Super easy fix.Claude on it finding because it hasaccess to our repositories. It's findingthe page and the styling. It removed itfor me, pushed a new branch, opened aPR, post a screenshot to the PR. That'ssomething that I have a preference to.So I ask Claude, whenever you're makingUI changes, put a screenshot in the PRso I can see it. One thing I also did,this is something special about ClaudeTag is it also learns about you overtime. So what you're seeing in here isactually an iteration of me working withClaude and the preferences I have, whichis always draft a PR, but make it adraft PR, not an open PR, so it doesn'ttrigger CI. And then always put ascreenshot of the before and after inthe PR description so that I can seewhat it looks like before I open it. SoClaude generated this PR for me. Iclicked the link. I was like, "Yeah, itlooks good. Now open the PR and mergeit." And then the crazy part is it madesure CI was passing. It posted in thestamp channel to get a review from myengineers. And then it told me when itwas merged. So

00:27:33after me sending the DM,reviewing the output, I didn't doanything else. It was done. It was alldone for me.>> What the heck?>> It's [laughter] crazy.>> What percentage of code that you wereshipping is just happening like thisright now? And is there like a trend?This is obviously like a very low-levelvisual change. That makes sense to me.Like give me a lay of the land a littlebit in terms of just how big a part ofyour workflow this is.>> I would say more than half of my code isdone like this now.>> What the heck?>> And it's it's not like this is I thinkthe simplest example of it. SometimesI'm doing like full overhauls. Like Ihave one and this is me directing it. Ihave some that are just automaticallyrunning right now. So, for example, Ihave a job that does cleanup that likeevery Monday it'll just show me cleanupand then it'll spin up a bunch of PRsfor me like that and I'll just reviewthe before and afters and then I'll justtry and merge them. Like this example, Ithink it's even not as forward-lookingbecause I had to ask Claude to do thedesign change, but very often Claude isproposing based off of work that hasmerged in the past for me or work thatis forward-looking or discussions thatare happening in Slack between people atAnthropic about what the UX should be.I'll have Cloud signal those to me andthen I'll look into them and then Cloudwill typically generate a PR for likeproposals for what it could be. And it'snot just PRs. I can show some otherworkflows. So

00:28:48this is another one whereuh it needed a little bit bigger of anoverhaul and so I was like, "Oh, likeyou need let's clean up this page."There's a bunch of different stateshere. We had an advanced section. Iwasn't sure what it should be. I had aFigma link of what we were brainstormingdifferent architects of it. So I justlinked it to Figma.>> Quick, give us a little bit of contexton that Figma link. far are you actuallygoing in Figma right now?>> Um, this one was mostly around likeordering stuff around. I went throughlike four or five different orders oflike should these configs be in thisorder and it's faster for me personallyto do it in Figma. I think the designersinternally actually we debate about thissometimes because some people still useFigma and we all use it at differentstages, some people don't use it at all.For me, I use it when I know I'm fasterlike when the iteration I know I want todo, I'll be faster doing it with likeauto layout and stuff in Figma. That's aspeed thing for>> when now are you faster in Figma? Likego a little bit deeper there.>> Uh well, for this one specifically,there's like we had like 30 differentsettings and I wanted to know if itwould look better collapsed and whatorder they should all be and whathierarchy they should be. And with autolayout, it's really easy to move thingsaround actually myself. And so I put upthree versions of this Figma linkactually. And I had one that highlightedas like recommended. Part of the reasonI also was in Figma for this one isbecause we didn't have anything likeadvanced and so I wanted to have a

00:30:03reference for Claude to know what theadvanced pattern would look like becauseit's a new one that we're introducing.>> Sure.>> Very simple though. So it added acollapsible advance section. It's makingall these changes on how everythingworks. It's putting up a draft PR. Ilike really trust cloud at this point. Iactually kind of wait until it's done.It put up a before and after for me. So,I clicked into it and then the reallyinteresting part that I love is thatthere was some gaps in my design. So, itmade a call. Like, it made a design callon its own. It's like, "Hey, I made adecision because your designs didn'tinclude it. And I think it's better."Actually, it says like, "If you want meto match the mockup, I'll do it, but Ithink my call was better." And it wasright. It was like so right. I was like,"Oh, wow.">> That's happened to me a few times in thelast week where it says, "One call Imade. I saw this discrepancy. It feltlike you were pushing me this way, butif we went this way, I think it'd bebetter for XYZ reasons." and I read itand I'm just like damn [clears throat]>> right>> it's it's right [laughter]>> so imagine claude doing that not justwith the knowledge of your designpatterns but with all the decisions thatare making on the architecture of howthings work and all the product and goto market decisions that happening forexample I was doing this at one pointand the team had decided to rename afeature and it just like oh by the wayeveryone decided to rename this so justas a heads up I'm going to update thecopy here as well I was like wow thankyou>> oh man that's so cool that's so cool

00:31:19>> and then I looked at it I was like heylike should we pull out this specificific setting because I think it might beimportant to pull out. So these are likethe kind of discussions I have withclouds like do you think I should dothis? I actually don't know. I'm askingto figure it out and it has an answerfor me why because it's one of the mosttouched controls. So it's pulling fromour data and it's saying hey actually Ithink it's important that you keep it inhere because it's a config that peopleuse based off of usage patterns and ithas that because it has its own accessto all of our like data logging. Puttogether a prototype for me so I couldclick into it. looked into the Porterprototype, opened the PR, made sure it'sCI pass, all good to go, merged.>> Okay, I have questions. [laughter]I think the first one is actually I knowthat there is somebody somewhere that islooking at this and they're like, she'sshe's just outsourcing her thinking tothe models. We don't even think anymore.We just let the models do whatever wewant. That's not design. What would yousay to that person? There is so muchdesign work to happen that needs tohappen right now that what you're seeingis everything that I believe I canoffload because the models are capableenough right now so that I can spend mytime on like the really really hardgnarly problems that need deep designthinking. I think we're all in an eraright

00:32:34now, especially if you're workingat one of the labs or one of thecompanies that have really adopted thisfastpaced shipping where you're justit's hard to keep up with everythingthat's going out right now. And so thesefew examples you see are the examples ofhow I keep up actually and how people atAnthropic are keeping up because we'refinding ways to automate the parts thatcan be automated. So we can spend a lotof time on like the really hard designthinking problems that we're workingwith. Like for example, a really hardthinking problem that's part of it islike how do we explain to people thatthis is an organization's claude? It'sone cloud that your entire organizationworks with. It's not just your claude.Will people want separate access in aspecific channel and how do we let themconfigure that so that you still havethe power of claude knowing everythingabout the rest of the organization butit's not breaching any securityrestrictions that you might explicitlyhave in your organization. How do we letyou know the difference in betweengiving claude access as an entity inyour organization versus just using yourown user off MCPs which MCPs are alreadykind of hard to understand like how dowe make that distinction between you.Those are the really important piecesthat we spend a lot of time thinking onthat we won't offload because it's aboutthe user model relationship and behaviorand like these design polish pieces orsome of these updated ones are likereally easy

00:33:49ones to offload.>> But you're still doing a lot of thatdesign work and even your own internalbrainstorming by having back and forthwith the models, right?>> Mhm. Absolutely.>> Let's talk about that piece then. Likeyou talked about like the gnarlyproblems. You also said like theexistential conversations. So like whatyou just showed in my opinion would belike very executiondrivenwork. Let's just get rid of all the P3things at scale. And I'm sure there's alot of people listening who that's kindof their usage of AI and they haven'treally taken steps toward more of thehow do I get Claude to accelerate mythinking on like the big picture stuff.So, like I know it's a little bitwishy-washy what these back and forthslook like, but for people who are notmoving in that direction at all, likehelp us understandwhat your practice looks like when youare collaborating with Claude on some ofthese really, you know, the the theambiguous gnarly problems.>> I think there's two pieces to this. Thefirst one I would say is that tacticallythere's a skill set that designers arelearning to develop. Some already haveand some don't. That is kind of aninverse of how you see the design rolehave been historically where there's ahuge school of thought that

00:35:04design isall around owning the quality and polishof everything and everything has to bequality and polished and I think it'sreally important to have that but itdoesn't have to be all of your time. Andso knowing when to use these tools forexecution and offload and knowing whenyou need to go deep and do the thinkywork is a skill set. And I think that'sone that I see people doing wrong. Likeyou might be asking Claude to doexecution when you you should actuallybe asking Claude to do the earlythinking. The execution is wrong. Likethe shape of the execution is incorrect.And Claude is not good today at tellingyou if your idea is wrong, it'll justexecute on the thing that you ask it todo. And so there's a responsibility onyou as a designer, which I think is areally important skill set to have toknow is this the right shape of productor should I be discussing with Claudethe shape of it and not the execution ofit like the mental model and theprimitive rather than how it looks tosomeone and that's a really big part ofdesign that I think people oftenoverlook. First, you need to be able torecognize that as an individual. Onceyou recognize that, I think when you areworking with Claude on the more initialside of things, I tend to have myexplorations be a lot more amorphous.Like there's not a specific output I'mtrying to get

00:36:19with Claude. I'm like onlyhaving a discussion. I say like, "Hey,I'm thinking about an idea and I'm notsure where it should go. Let's keep itopen-ended." Like I'm working withClaude as if I would be working withlike a product partner almost that Iwant to go wide with me in like ourexploration process, but I'm also askingit to pull in all the knowledge that ithas already from research, from data sothat we can make informed decisionsforward. And then a process like thatmight look like, hey, like I'm trying tomake, I don't know, dummy like a weeklymeal planning app with me and myhusband. What does a meal planning appneed? Like what are all the parts of themeal planning workflow? You need toorder, you need to like have theingredients, you need to like know whatfood you want to cook at home, you needto know when you want to go out. Likeit'll propose the entire workflow forme, not from a design perspective, butfrom almost like a functional productperspective. And then some people arelike, hey, like what is a meal planningapp actually? And like I'll have theseexistential like is it an app? Is itjust actually like a workflow? Like do Ieven need a UI for this or can it justbe nothing? and I just ask you to dothis and then sometimes Col be like Ithink you like apps so you should haveone but your husband doesn't like appsyou can just text him what happens atthe

00:37:34very end it's like those kind ofdiscussions that you have that you wouldtypically prompt yourself as a designerthat you can prompt with Claude now andhave like a that brainstorming partnerwith you so that's where a lot of itgoes and then that thinking isn't asvisual as you think it is though likesometimes I'm making mind maps withClaude or like tables with Claude but Ithink a lot of it is just likeconceptualizing the pieces There'sanother thing I want to talk about thatis you've kind of talked around it acouple times where you mentioned how alot of people like there's this schoolof thought where design is like thingshave to pass through us to go out thedoor and we're the ones that uphold thethe bar for craft and quality and I gethow working at like the tempo thatyou're working at maybe that is actuallyimpossible but it also does feel like itis directionally where a lot of thisstuff is headed as lines get a littlebit more blurry in terms of who ownswhat and cloud helps everybody doeverything to an extent. So, one of thethings that you mentioned last time wetalked was how you thought designersneeded to be more comfortable letting goof design. And I think we've seen like acouple examples here of what that kindof looks like, but why is that evennecessary? And are there times evenwhere that has felt uncomfortable foryou?>> Oh, absolutely.

00:38:49I think in the ease ofof anthropic, we're very experimentaland we're so early in this explorationof like the shape of the product. Like Isay this all the time, we're 1% in thisjourney of like what the shape of AIshould be. And sometimes knowing when itneeds that craft and polish or knowingwhen just the shape of it is actuallywhat you're trying to test is reallyimportant. And so you might launchsomething or test something that looksterrible, but what you're actuallytesting is the underlying mental modelof the shape of product. And I don'tthink it's worth polishing that thingactually because it might not even makesense. You might have to rewrite itbecause it's actually not the rightshape of what it needs to be. And thoseare the times where I think actuallyspending a lot of iteration on thatpolish isn't a good use of time. And Iexpect the designers on my team to knowwhen something needs to be polished andwhen it doesn't. that experimentation ofmental models doesn't require that highof craft and it's better done actuallyin speed in the open in iteration asopposed to spending a lot of timepolishing something that's actually thewrong full wrong shape>> of what the product should be. Thesecond part about handing it off is Itend to find the more we can shareresponsibility of what we think qualitylooks like in our product and not be thegatekeepers but educate and kind ofdelegate around us so that both ourengineers and product partners feel thatquality

00:40:04is something that they own anddrive too. You want to like lower thefloor and raise the ceiling for everyoneto be design shepherds a little bit orquality shepherds. And so lowering thefloor looks like making it so thatanyone can feel empowered to say I thinkthis is the right quality and you teachthem why it is or not or claude actuallyteaches them if you have goodautomations in place and good designsystem files in place to teach you whatgood looks like and then raising theceiling together meaning I don't thinkand it never has been historically onlydesign's responsibility to mean qualitythe most successful organizations feellike it's owned by everybody and so weneed to make sure that we're not pushingthe bar on our own but we're pushing theare with everyone around us and you dothat by giving people that shared senseof ownership over what quality looklike.>> What are some practical next steps thatdesigners listening could put in place?Maybe it's like a workflow or a tacticto raise the floor and empower otherpeople on their org to contribute at ahigher level when it comes to design andmore polished output.>> One that I do which has somewhat workedis I'll stack nicer PRs on top of myengineers PRs in the workflow that youjust saw. I actually don't do themmanually. I do them pretty automated.Like once Claude learned that that was ahabit I did, it just started doing it onits own, which is crazy cuz it's veryproactive. Uh that's one way because itdoesn't slow down the

00:41:19progress and itdoesn't erase the work that thatengineer does. It kind of gives them analternative or an iteration on what theydid. And typically when they see that,sometimes I actually stack or iterateoff of my PR so that it still has theright idea of what they wanted to do,but it like is in the right shape ofproduct as well. And then I have beenexperimenting with some workflows whereI like have Claude understand the thingsthat I think about. Who are we servingthis for and what are we trying tocommunicate them? You know like thebasic question what are we communicatingto them? Like does this align with ourdesign systems? Has this gone throughthe past with like our contentautomation? Like how does this fit intothe broader ecosystems of our product?Should this be a product or should it bea feature? Does this need to be named orcan it be invisible? Like there there'sall these things that I'm constantlyasking my engineers that uh Claude haslearned over time that I like to ask.And then I'm slowly putting togetherlike an automated reviewer that kind ofdoes this for me and we'll put up PRs orthe interesting thing about beingintegrated in Slack is it's not alwaysabout putting up PRs anymore. Sometimesit'll just send a message to someone andbe like, "Hey, Megan thought this aboutthis. Like what do you think? Here'slike a prototype that you could playaround with to like see what she wasthinking about." And so massaging thatlanguage is very similar to like justworking with people. And that's reallyhelped. So I would say invest in likehow you define your quality and how youuplevel it and then share that as

00:42:34aworkflow that anyone on your team canuse and don't do it to stop them.Although it is important to know when tostop and hold the bar, but try and do itin a way that like is additive and isteaching and guiding rather thanstopping. I think>> so much of what we've talked about todayexists outside of pixels. And I thinkit's kind of what's made theconversation really enjoyable to me. But[laughter] I got to imagine still thevast majority of people that arelistening are designing interfaces,right? Like historical B2B products. Andeven for myself in the last year and ahalf, I felt more and more of the valueprops around the products that I'mmaking shift towards like what's thebest model to get context through likean MCP and back to cloud and close theseloops and like speed up iteration andthat kind of a thing. It feels a littlebit uncomfortable to me to be totallyhonest like like actually quiteuncomfortable. I'm like, is this we justaccelerating toward this interfacelessworld? And I'm wondering how much youthink about that and if you have anyideas around where does the interfacehave lasting value and moredefensibility when it does kind of feellike you can almost do anything directlyinside of a chat. Now,>> I've been really toying around with thisidea recently and it's something thatI'm slowly developing a strongerconviction that

00:43:50we're still going toneed a lot of fixed interfaces. Sothere's two concepts now that UI canfall into. There's fixed and thenthere's like adaptive or likenon-deterministic is kind of and I thinkthere's just so much that needs to bereliably stable that you don't want torelearn and you don't want to be dynamiclike a login screen, billing settings,like there's a lot that actually feelsimportant to keep stable and I thinkthat UI will be really important.Knowing what should be fixed and whatshould be adaptive or dynamic is areally important decision for you as adesigner to make as you're designingthese features because I think oftenpeople will assume that everythingshould be customized because it can. I'mvery much of the opinion that a lot ofpeople don't want everything to becustomized. The right things should becustomized and adaptive. And then on theadaptive piece, I think it's like what'sthat container layer from what isadaptive? How do you express in a visualsystem that something's adaptive versussomething's fixed? How do you let userscustomize it in a way that theyunderstand that they're customizing thetool that they're using or they'recustomizing the output that they'regenerating? These are all like reallycritical workflows to get right. And soI think what I would guide peopletowards is like yes there's going to bea lot that's non-deterministic

00:45:05but thereis still a lot that should be fixed andit's your job to figure that out. Andthen in the things that arenon-deterministic how do you make thateasy and clear to people.>> Yeah makes sense. Honestly all of thestuff with artifacts is even challengingwhere I've been drawing the linesthinking about it. Like I've hadmultiple instances now where I'mdesigning you for a simplistic examplemaybe like a search results but thesearch results is natural languagebased. It can basically return anythingand the shape of that is totally tied tothe query where it's like okay I don'twant to have an entirely dynamicinterface. I want to probably have likea set of components and maybe map thatto a system prompt where like we'regetting repeatable shapes where there'slike a little bit of familiarity butcloud can kind of do anything you wantwith it. And and I guess I thought thatwe were speedr runninging toward a worldwhere that was how my future Claudeinteractions would look where you'rehaving like, you know, a bunch ofdifferent component combos that aremaybe giving me different ways ofexpressing intent outside of just typingit. And now I'm like, man, maybeArtifacts has swallowed that entirelyfrom me where it's like I'm actuallyjust not even interfacing with Claudeinside of the chat. It made me almosteven more skeptical about the level ofinterfaces that are going to exist inthe future. I don't know. It's it's theinteresting thing that I'm I'm thinkinga lot about. It's like where where arewe going here?>> I kind of separate

00:46:20my school of thoughtin between I'm generating a thing that Ineed to look at and I'm just doing a jobthat needs to get done. Like noteverything has a visual output to it.And so I think for the things that arejust doing jobs that get done they justneed to be infinitely configurable andwork for you but the end of the daythere's a starting point and that stillneeds to be designed so you can knowwhat it can do. Like there's a lotaround like teaching people. And then onthe dynamic side of things, I do thinkyeah, like having these be dynamic iswhat makes them so powerful. They shouldbe and we should lean into it and becomfortable with it. And where youcustomize might be on like its visuallanguage or the tool you use to edit itafterwards.>> Are there other unmapped designchallenges that you think that we as anindustry are going to be staring in theface in the coming months and likedifferent types of problems that AI iskind of bringing to the forefront as theway that we interface with these modelschanges. I think a really big one thatwe're we're going to have to reckon withis when you can build everything shouldeverything be built. Just because it canbuild and just because it can be afunction doesn't mean it needs to be. Itadds a lot of product bloat. It makes itvery complex. Minimalist design had itsplace and I don't necessarily thinkthat's the solution. But I think interms of like full functionality, thingscan be a little bit simpler than theyare right now. And not every idea isgood or necessary. Sometimes adding itis worse.

00:47:35but because it feels so easyto build now. Everyone's buildingeverything. So that's a big one that'sreckoning through. Like having thediscernment to decide if it's worthbuilding or not. Not even worth buildingbecause you can just build it, but ifit's worth adding to your product, Iguess, and launching it.>> Okay. I kind of want to tie a bow on allthis. Like we've covered a ton. It'sjust new, right? Like there's so muchnew in this conversation that itstretches me in all of the best ways.And I I think it does produce a lot ofthese existential questions of like whatthe heck is my role moving forward,right? And so I'm curious if you haveany thoughts around like what's the newshape of designer look like that's goingto thrive in the future. We talked aboutthis skill around being able to decipherbetween whether I should be in executionmode versus shaping mode. Are thereother things that you think are going toseparate the best of the best designersin the years to come? I'd say the firstskill set is just in the next few yearswhere we're still in the emergent phaseof this technology. Being veryopen-minded and curious, I think isreally useful. Like hold your opinions,develop them, and then be willing to letgo and update how you think aboutthings. Both what your role is, but alsoyour product is extremely helpfulbecause everything's very dynamic. So,one of them is just like

00:48:50having fluiditybuilt into you and being able to leaninto that ambiguity and be curious andenjoy it and like find a place to startand move forward in spite of the factthat it's chaotic. That's a really bigone that I think will serve peopleextremely well. I think it's a hard onebecause design used to be so structuredand processed that it feels like youreally want to cling to that. But I'murging people to do the opposite whichis like lean into the chaos, develop newways of working and like continuouslyupdate. And I think that will reallyserve you well. So that's a big skillset that I think we need to practice.Kind of like let go of the rigidprocesses and roles as you know it andjust lean into building good productsand you'll find your way. The second oneI think is just maybe discernment isjust another word for taste and I'm justgoing to like kick myself in the[laughter] future. But I thinkdiscernment on what should be built, howto best build it in a way that suitsyour product and what you're trying toexpress and what role you should play inbuilding it and where that craft shouldfall is like a big one. Like that's thecore builder process right now. It'slike if you identify a thing that shouldbe built or an idea, how do you shape itin a way that it will land into yourimagination from your imagination into aproduct and then knowing when to call itquits and be like that was actually bad.Let's not do it.

00:50:06Looking at someoneelse's and being like that's not greatand these are the reasons why oridentifying something that's really goldand be like yeah that's great. Let'skeep riffing on it. We're like two bigshape iterations away or like one big exexecution riff away from getting there.like that discernment on like wheresomething is and how it fits in andwhether or not to go after it is reallyimportant. And then the third one Ithink is like feeling full ownershipover finishing the ship because this erarequire like the the second skillrequires knowing the difference inbetween the shaping and the executionand like when to lean in and what'sgood. Sometimes that execution andpolish will come after, sometimes it'llcome before, sometimes in between. Andowning that loop, teaching your peersaround you how to do it. Elevating yourentire organization, not just design,but everyone around you is a reallyimportant role for design. And you needto feel your responsibility for that inthe product. And you need to act on itas well. And so those three kind ofskill sets, ways of life, outlooks arelike specifically actually what I hirefor on the team and specifically whereI've seen people really succeed.>> It's interesting. I I feel like in a lotof the conversations that I'm havingeach one there's like a pull like wecan't just stay in like the traditionaldesign box that we've drawn forourselves over the years, right? Soyou're kind of pulled maybe

00:51:21more intofront end and more technical or maybemore toward product and on the surfacelike you look at anthropic and it's likewell you're working on the models,you're designing a freaking CLI, right?like you would expect the poll to betoward like quote unquote engineering,you know,>> but everything that you've said thisentire conversation, like you were aproduct strategy expert, you know, likethat's kind of what you're bringing tothe table right now. There's a slight attimes hands-offness in terms of what isgoing out the door, the the craft, theslinging pixels around, like it's verymuch so, like you said, what do webuild, what do we not build, and why?and that kind of being almost like theanchor of the core set of skills that ispropelling you in your career. I don'tknow if that resonates or not, butthat's definitely like kind of what I'vebeen hearing over the last hour.>> It absolutely does. And I think maybethis is something that a few folks inthe industry have started talking abouta little bit is like maybe as we allbecome builders, we actually just havearchetypes that we lean into and likeunique spikes of skill sets that wehave, but our unique role doesn'tchange. Cuz I think the skill sets thatyou just mentioned, they apply to peoplewho like lean really heavily into frontend and have a high degree of polish.It's just like what they're talkingabout is more at a feature level or morelike very vertically deep as opposed tohorizontally broad. So I still thinkthose characteristics are important.

00:52:36Itjust depends where you are in the lifecycle of a product and what you choseto anchor on. But I think I would expectall designers to be able to flex intothe higher and lower level. I thinkthat's where the career is going. Beforeyou could specialize only in one in oneof these. Now I think the idea is thatyou're developing them equally. Likewe've expanded the skill set that wewant you to be able to do. Well, interms of curiosity being the definingcharacteristic that comes up on thisshow, you've definitely dangled a fewvery interesting threads to pull on forme and I'm sure a bunch of peoplelistening. So, I very much so appreciateyou coming on, Megan. You were quicklybecoming one of my favorite people inthe industry to learn from. And I'malready going to say we're going to haveto run this back on I don't know whatthe timeline is because things changetoo quickly. But, uh, I reallyappreciate you coming on today andsharing what you're thinking about andwhat you're building.>> I'm glad. I'm glad. Thanks. And I wouldadd like if there's one thing thatreally excites me about the designindustry right now and what motivates meto come out and speak with folks and tohave these conversations with you islike to really encourage folks to knowwe're early. We are all able to shapewhat we want right now. Like that's theexciting part of it. So participating inthis discore, trying these things out,developing your own opinion [music] isthe fun part right now. We should all bedoing it together.