How Anthropic, Every, & Ramp design with AI

00:00:00Welcome to Dive Club. My name is Rid andthis is where designers never stoplearning. This week's episode is aspecial one because it's actually a liverecording of a panel discussion thathappened last week in New York City. Weget to hear from three incredibleguests. The first is Megan Choy, who'sthe design lead behind Claude [music]Code and Co-work. The second is DanShipper, who's the CEO of EveryY and oneof my favorite thinkers when it comes toAI. And the third is Bradley Zipperwho's a design engineer at RAMP. And[music] the entire point of thisdiscussion is to do a deep dive into notonly the future of design but what[music] AI workflows look like at someof these prolific companies. So withoutfurther ado, [music] let's dive in.

00:00:51[applause]Welcome back everybody.I've always wanted to start talkingafter a live applause. That felt prettygood. I'm not going to lie. That feltpretty I think I could do that. Uh,three amazing guests who have all givenwonderful workflows and we're going togo a little bit deeper here and talkabout the different things that peopleare observing in the industry, how theway that we work is changing some of thebest practices that are still kind ofactively being discovered inside of ourorgs. And I'm going to try something alittle bit weird. Haven't really preppedthem at all at all. And I'm going to askus to put on these hypotheticalconsultant hats. not so hard for you Danbut you know maybe maybe over here alittle bit new and uh we're going tojust help people think about differentavenues for transformation right like ifwe want to grow as builders as orgs whoare leveraging AI to you know thefullest capabilities and reallyrethinking from first principles how weoperate what is that going to entailright and so talking Megan with youearlier uh something that I learned isthat Megan does a lot of onsites withteams teaching these design orgs how tomake the most of cloud and building andusing AI. And so I'd like to start with

00:02:06you and we can put on again this not sohypothetical consultant hat and thinkabout you know you talked about likethis 2011 level course almost thatyou're putting people through. So whenyou think about the transformation thatyou're hoping to have design or see,what are some of the main mile markerson that journey that you would look for?>> Yeah, I think I met a few folks todayactually who attended like some previoustalks back when cloud code was lesspopular and I think the theme of 2025and like for orgs first getting intothis is like let your designers getaccess to your production codebase.That's like the starting point of thisconversation. It is been a very gatekeepand for a lot of good reasons forsecurity, for privacy, for like yourability to understand that you're makingthese changes. It's been segregated fora while, but it's so so so important nowthat these tools are available and likethe more information access they have,the better you'll be able to build toget access to that production code to gothrough the process process of shippingto production because the closer you areto what your end users see, the closeryou have to like influence the product.The second, and this one is deeplyuncomfortable for a lot of designers, isin the same way that we're asking ourengineers to let us in and help us code,you need to

00:03:21be more comfortable lettinggo of design. And that means that a lotof features can go out without you. Alot of the time, it's actually verypossible for people to get a V1, a V2,even a V3 out there and have it bepretty good and have it be prettyaligned with your design system. And,you know, you build the automations inplace so that there right checks andbalances. But I think it like goes bothways. And so those are like the two bigmilestones that I often see. And I thinkit's crazy if you think about it thatlike we all want access to thatproduction code base, but when it comesto having our engineers design for us,the initial reaction is like, "Oh mygod, I don't want to do that." But it'sso important, I think, to like shiftyour mindset a little bit and likerecognize that we're all kind of havelike this new ability to build. Allright, let me push on that for a secondbecause I've talked to quite a fewespecially larger teams but not evenmassive teams where they're veryintimidated by getting designers intoprod and there's this story that you cantell around well if we just make aseparate repo and a playground thendesigners are coding and that's you knowthe 8020 value. What would you say tothat leader?>> I just like then you're maintaining tworepositories. If you ask any engineer ifthey ever want to fork their codebaseand maintain two versions of the samething, they will always tell you that'sa terrible idea because you have tomaintain two things and they're alwaysgoing to be out of

00:04:36dates. Also, part ofbeing in the production codebase ishaving access to all the tools that yourorganization is building into thatcodebase because you bet your engineersare also on this journey learning withus and they're scaling themselves. Italso gives you access to the actual dataendpoints so that when you do loggingand you're doing data queries, you canactually see what's happening. There'sso much built into a real codebase thatyou just can't replicate in a sandboxthat it's just you just really needaccess to it to really experience thepower of what it's like to build aproduct that your users end up using.[music]I've anticipated DES's recent releasefor a full year. It's called Surfacesand it enables you to design andprototype directly on top of an existingproduction interface. All of the pagesand flows that you need are preloadedinside of Destin as starting points. Soyou can easily make changes, exploreideas, and then when you're ready, youcan share your prototype in a singleclick. Destin is the only design toolthat allows you to design on existingpages and flows pulled directly fromprod, which is a pretty big deal forteams. And you can get started today.Just head to dive.club/desen.That's D SS SN. [music]Never in a million years

00:05:51did I think mydesign workflow would change sodrastically in the last few months. Anda big part of it is Paper's [music] newsnapshot tool. It's a Chrome extensionthat lets you copy any [music] componentor element from your live website andthen paste it directly into Paper aseditable [music] layers. So all the timeI'll grab something from prod, haveClaude immediately spin it up sixdifferent ways on Paper's Canvas, andthen when I'm ready to send a conceptback to Claude, it's seamless becausePaper's Canvas uses [music]real HTML and CSS. That workflow feels alot like the future of design to me. Andyou can start doing this today. [music]Just head to dive.club/paperto try it out. Now, on to the episode.All right, Bradley, I'm going to go toyou for a second here as the residentdesign engineer. this person who's beenkind of bridging and seeing both sidesof the table for a while now. In thisworld where many more designers areshipping, touching production, doingthings in GitHub that they may or maynot fully understand, how is thisshaping what you think really soliddesign engineering collaboration lookslike? I don't know if it's so much thatwe all need to becomedesign engineers or designers

00:07:06orwhatever box or label you want to put onthings. Um, I think it's more aboutwhat does it mean to put care andintention into something? Like Megan'ssaying, we can probably get to a placewhere seven out of 10, we're at a sevenout of 10, like out of the box, a thingyou want to do, an idea that you have,we can get a seven out of 10. Anyone cando it. And that's actually great. Um,but that just means you have more roomto care and put love and intention andthought into the things that you build.And um, I think that that that's what itin my mind is about. Sure. It's like youshould understand this uh this animationyou're putting out and uh what it meansto be performant on the web and uh whatit means to not make me rage quit yourapp because you have like a a menu andthen another menu. But when I go alittle bit outside the boundary of thatsecond menu, it like closes the menu andI'm like, [sighs]you should care about those things. Butthat is all just care. Just you care.It's just you caring. And now you havemaybe a hopefully a little bit more timeto care.>> Okay. I want to touch on the little menuthing because I think it's exists

00:08:21inthis box of responsibilities that all ofa sudden have kind of fallen potentiallyon designer plates. Like even speakingfrom my own experience, that's somethingthat previously I would make a linearticket. I would add polish and then Iwould make a comment and be like pleasethis is really important, right? Andthen mostly get ignored for the nextlike quarter. And now I don't even makethe linear ticket. I just open up a newwork tree and start working on it andit's empowering but also the slope isslippery because I can find myself justpolishing and polishing and polishingand polishing and so given the newcapabilities I'm curious how do youthink about allocating your time as adesigner who all of a sudden may or maynot feel like you can shoot lightningout of your fingertips with thedeploying code and pushing to PRs andstuff like that.>> Sure. In the beginning, it takes a longtime to to do these things and to putthat care and craft in because youhaven't done it before. You've neverdone it before and maybe you juststumbled upon like uh John Famos'stweets and you're like, "Wow, [laughter]shout out." Uh, I should I should dothat. And then you take that tweet andyou give it to Claude.>> Come on, man. This is hitting too closeto home. You give it to Claude and thenyou maybe you just hope it works

00:09:36ormaybe you ask why does it work and heynext time can you make a skill out ofthis? Can you make a skill? And and youdon't even write the skill. It justmakes a skill. Um so I don't know. Idon't think it's that uh uh uh maybe inthe beginning it takes a long time. Westart to spend a lot of time polishingthings but um I I think that that whatare you doing? If uh you can get toseven out of 10 like this, you probablygot a lot of time back. You just maybedon't realize it.>> All right, there's a bunch of things I'mgoing to put a pin in. One is the sevenout of 10 one. Dan, I'm coming to youfor more organizational transformation,but really quickly, I'd love to hearfrom you. This idea again of likeeverybody can code as much as they seefit. You have an engineering backgroundand yet you're leading design of a youknow coding product. like how do youthink about the allocation of your workand maybe have you seen a shift over thelast even couple models?>> Yeah, we're a very lean team on thecloud code side I would say and so Ithink I just like I really trust thedesigners on my team to prioritizethemselves and transparently sometimesthat looks like not prioritizing polishand I think that's okay. There's a fewtruths that I hold in my head as we'reworking that I think really help guide

00:10:52how I see prioritization should ship.Part of it is because we're working at alab and so our job is to be likeresearching at the frontier. And so onething I like to encourage the team tothink about and this is like an ethos inthe cloud code and team in like theproduct team at Enanthropic is like isit worth polishing something that's notgoing to be here 6 months from now? likewe're so early in this journey right nowthat we actually don't know what thefinal shape of these products are goingto be and so is your time better spenton that future looking thing where itrequires deep thought and kind of like alot of thinking and like cloud can't dothat yet and requires like a really bigcanvas to explore or is it better spenton like those 30 tickets. I actually gotsome great feedback at one point from myengineering team that I was not spendingmy time wisely pushing these PRs.They're like, "Yeah, you're helping thepolish work." But right now, Claudeisn't good at design. So, I am doing alot of manual look over it. And itwasn't just like a great use of my time.And so, I think it's hard to internalizethat as a designer cuz it feels like nowyou can do it all. And it feels likeit's all on your shoulders and yourresponsibility to maintain that polish.But we like in the same way that nowthat we can do it also, your engineersshould be able to do more too. Like it'sa shared responsibility to care aboutthat polish and about that care. And so,it's not all on you. You can still bugyour engineer

00:12:07to do polish on featuresin the same way that your engineer cannow bug you to implement something onthe front end. Like it's it's all of usand we're all in it together.>> Yeah, I like that a lot. I've felt thatmyself where it's just so easy to knockout a little task, feel good about it,and then immediately just jump to thenext polish ticket and you can kind ofall of a sudden like six hours are goneand things look great, but then you'rekind of like, did I work on the highestleverage thing? And I think it hits onthis idea of, you know, you kind ofreally have to be self-aware around notonly how you're working as anindividual, but how the org isoperating. And so maybe Dan, if I cantoss it to you. When an org comes to youand says, Dan, we know AI is a big deal.We don't feel like we are evolving asquickly as maybe we should. Can you justhelp us get to that next milestone?Where are some of the main opportunitiesthat you look for or signals that youwould use to get a sense of like wheresomeone's at? Like if you're going toget this or to the next level, what doesthat typically look like when you'rewearing that consulting hat?>> Uh the main thing that I always look atis what is the CEO doing>> and maybe more broadly what is theexecutive team doing? Because I thinkfor the first, you know,

00:13:22three years ofAI, there was a lot of lip service paidto, yeah, we have we got to have AI orwhatever, but it was always like andwe'll do an AI working group. And umgenerally that has not worked. Um andthe organizations that seem to do thebest are the ones where the leadershipteam is like in the tool all day. Andthat's the only way that you'll you'llhave an intuition for how to manage ateam who's in the tool all day. and it'snot outsourcable. So that's we spent alot of time with leadership teamsactually just being like okay literallyopen cloud code and make something andthey they love it but a lot of them arekind of afraid or don't really want todo it. But I think that's that's one ofthe big leaning indicators to me.>> You're a pretty good prototype for whatit looks like to be the leader who liveswith the tools and has hands in the claypretty much all day is what I'massuming. So maybe we could zoom in onyour workflow for a second. And if youkind of put you know you today up andthen over here is you maybe I don't know3 6 months ago something like that. I'mcurious in terms of how you areinterfacing with the models on some ofthese more you know ambiguous creativeopportunities. Where do you see thebiggest deltas in terms of how you areworking what language you're using toolsskills

00:14:38anything like that? I mean, Ithink there there was this big momentthat happened in I want to say Novemberor December of 2025 when Opus 45 and GPT53 came out um or yeah 53. It was likethis moment that I think a lot of us forus at every like we had been using cloudcode for about we've been using it forabout a year and and had not been havenot been looking at code for about ayear. So at that point it was stillpretty risky and people were like areyou [ __ ] kidding me? You're you'renot looking at the code. Um [snorts]but I I think we could see that that'swhere it was going and we were willingto deal with some of the like weirdnessof that.And for me uh because you know I'm I'mwriting and I'm in meetings and I'm likedoing all this stuff like I don't I liketo vibe code but I'm not going to do anyserious engineering. Um, but at whenthat stuff came out, when O Opus Opus 45and uh 53 came out, I was like shippingPRs to our products and I was like,"Holy [ __ ] [ __ ] this is crazy." Um,so I think there that that has been abig shift because the the

00:15:53uh the modelsare just much smarter. They're much moreindependent. They don't do as many dumbthings. And I think we've also figuredout like cloud code you guys likecreated [snorts] this I think a lot ofpeople had models of what is an agentthat does work for you look likeand prior to cloud code there was likethe Devons of the world or or or evencodeex the original codeex was uh it waslike you have a sandbox agent in thecloud and what what was really specialabout cloud code when I first tried itis like no no it's just on your computerit has access to all the things that youhave access to and that just opens upthis whole new territory and I think uhas that has started to catch on and asthe models have gotten better it it hasbecome this new work operating system ina way that I think is really reallyspecial and uh so so the big powerupsare yeah I can you know that that appproof like I just made that in betweenall the other stuff that I do um I haveshipped PRs to like pretty much everyone of our products but don't reallyknow the code base. Um, sometimes meshipping PRs is actually really bad. So,um, [laughter]it's it can be it can be annoying. Yourmileage may

00:17:08vary. Um,like I said earlier, I I strugglemightily to respond to emails on timeand I'm now just perfect. And it's it'scra that's really crazy. Like all thethings that I would normallyprocrastinate on, I just it just getsdone. And you know I have a lot of Ihave a lot of thoughts on you know justgenerally what happens to experts andexpert workflows and and where where tospend time but I I think themy whole day is very different becauseI'm just in codeex or cloud code all dayand I use I do all my work in there. Iwant to talk about this phrase AIfluency, which feels like it's probablythe next buzzword buzz phrase. And let'sjust like increase the fidelity of whatthat could look like, what good lookslike inside of an org. And I'm going tostart with you two because youshamelessly plugged your designer roles.And let's imagine that somebody issomehow inspired by your pitch and theywant to join, right? And maybe we fastforward six months and my question islikewhat does good look like, right? Likewhat is a behavior that they would bedoing on a regular basis that would getyou to point at them and say, "Yeah,that

00:18:23that is what AI fluency looks like.That is what we were looking for when wewere putting this design role outthere." If in six monthsI can look at you and say that you havefound a wayto cut out all of the noise and trulythink about

00:18:44what what matters most, right? It'slearning. How fast can you learn? Whatis the iteration of you learning andthen acting upon that? Um and then whatwhat do you do with that learning andhow do you proliferate it throughoutlike an organization that is now at liketerminal velocity where everyone isdoing things and putting things out? Umthere's this this this idea that um youknow we can't alignment doesn't exist. Iagree with this by the way. Uh but thethe cost of alignment is is quite highand the cost of redundancy is lowbecause we can kind of clean stuff up.uh but I think if in six months you arethe one who is um able to understandacross your your whole product um verydeeply right um [snorts] howyou can simplify all the way down fromyou know we have 15 things that solve 15problems to we have four things thatsolve 30 problems uh I think to me thatmeans that you're leveraging the toolsbecause I think the reason that we don'tget to that today or we do now, butwe've never gotten to that is because wespend a lot of time [gasps]doing administrative work or um figuringout how to set up calls with researchersand

00:19:59uh since we did shamelessly plug uhRAMP, my time I've I've been at RAMP for4 months now and um my first week I metwith the research team and I've neverscheduled a research call in my wholetime at RAMP. Every week, four of thempop on my calendar and they tell meexactly what I need to talk about andthey know what I'm already working onand I just go and I just show up and Ijust get to learn from customers and Iget to hear about what they do and whatthey care about and what resonates intheir world. Um, and then I get to applythat across the entire organizationbecause um, I don't have to read all thedocs. I can just synthesize them and andI can understand and and look verydeeply at umI don't know what comes next. Where arewe going? Um [snorts] and I think that'sAI fluency to me. Just just carve outmore. Be be selfish. Be selfish.>> Okay. So we have systems thinking theimportance of internal workflows.Something I've been feeling a lotrecently too. Dan, I'm curious to hearfrom you. you know, you let's say youfind the senior product designer in thisroom 6 months from now, what would getyou to the point where you're like 10out of 10 higher?>> Overall, when I think about people whoare who are experts at at what they doand in in AI, I I sort of split up andand we can say maybe designersspecifically, I split it up, but

00:21:14I thinkthis applies just as well to engineersor, you know, product managers orwhatever. I split it up into twobuckets. One is>> [snorts]>> um because these tools make it makecompetence cheap for everyone. There'sthis whole there's this huge glut ofpeople doing design work in every thatare not designers. So a good designer isgoing to figure out how do we you how dowe make systems to harness that designwork and use it rather than like be likeno that sucks or we we can't use it.Right? But that's actually hard. there'sa lot of systems to build on that side.Um, and then the the other thing theother bucket is how do I use these toolsto make something that no one has evermade before.And uh the first one is you know there'sall this there's all this routine workthat I can I can sort of automate and Ican harness the routine work that allthe other people in the organization aredoing and then there's all this otherstuff I can do that was just would justnever be possible otherwise. So, um, andand I think the kinds of people thatlike to do that are very curious.They're very playful. The, uh, the kindsof people I love to work with aremulti-dimensional. Everyone inside ofevery is like, you know, Kieran, who whoruns Kora, our email

00:22:30agent, was like,uh, he's, you know, super technical, buthe was a composer and a baker beforethis, like professionally. Um,and I think that kind of person, allthese tools are just this amazingplayground. And you can what I love iswhen the design team's like, "Yeah, I'lljust go I'll just go make a little appfor that." And like, you know, and oftenit's it's a little internal tool orsomething like that. But sometimes it'slike, you know, the um that that graphicI showed you with the uh the ZenosParadox graphic, like I just pulledDaniel, one of our designers in like theday before because it was something got[ __ ] up and I I just needed someone todo it and like we just got it done andit's really good. Um, so there's reallythere's like there's something aboutworking super fast on new things andbeing like really excited about that andusing the tools to harness what is nowpossible and like push it to a levelthat would wouldn't be possibleotherwise. One thing that Dive Club hasmade abundantly clear to me over thelast year is that the practice of designis changing and the old process ofgetting feedback just doesn't quite[music] cut it in today's world. Andthat's why I'm excited to announce thatInflight is

00:23:45officially in open beta.It's the feedback tool that I've alwayswanted and it's built for a world thatmoves at the speed of AI. So I can sharemy prototypes, give context and videowalkthroughs, and Inflight makes it easyto get the exact feedback that I need tomove forward, whether it's voting ondirections or maybe even getting thegreen light to ship a new idea. [music]And all of this is available in a singlelink that I can drop into Slack or maybeeven share with power users to test[music] out a new prototype. I useInflight every day and it's totallytransformed the way that I share work.So, I'm excited for [music] you to trythe product and if you ever want to jamabout it, just email me atridinflight.co. [music]I want to return to something thatBradley said, but maybe I'm going tohave you talk to it because I knowyou've been thinking a lot about how tolearn [music] as an individual with themodels because a lot of what you'retalking about is like, you know, beingcurious, trying things, making sure thatyou're continuously improving as anindividual. And so I'm curious when youthink about your journey as somebodywho's constantly tinkering with thesemodels, how have you been able to growthe muscle of making sure that you arelearning as you use these tools?>> We're also hiring in case that wasn'tclear. [laughter]

00:25:00Um, but I thinklike we are so early. I cannot emphasizethis enough to everyone here. We havereally only solved two use cases and theuse cases of everything that can besolved right now. Search and coding.There's so much else out there that toassume or to think that we're likeanywhere close to the end right now islike we're just not. We're just so faraway from that. And so like I operateevery day imagining like thinking in theworld like this could all be obsolete.like every pixel I push, everything I'mtrying today could be obsolete next weekdepending on the next model that comesout, depending on like the nextinnovation that comes out. And I thinkthat belief that like there's nothingreally that I that you need to like holdtoo dear and you need to be willing toupdate your belief system constantlylets you be very exploratory in thedifferent directions that you're going,but also like leads to you wanting tolearn more different directions. Like Ithink that's truly how we need to beliving today because we're just so earlyon. And so once you like embody the ideathat we're like like we're so not nearthe end point right now we're like 1%and we all have the power to shape thenext 99. It

00:26:16becomes like an opportunity.It becomes exciting actually. and itbecomes fun and like it be and likeeveryone's next 99 is going to look sodifferent from each other because thepower of these tools lets you be socustomized that I think it just givesyou the opportunity to learn and focuson the areas that you want to learn andfocus on and so it just it just becomesfun and interesting like learning forthe sake of joy I guess is part of it.Can you talk a little bit more aboutwhat it's like in your position whereyou you really are having to operate inthe future, right? Like you're designingfor maybe the next model release or thenext six month set of use cases. Howdoes that change your daily weeklypractice of design?>> Probably in two fundamental ways. Thefirst is that I think because we are ofcourse still shipping products today andwe want our users to have a good time.We need to be able to hold two truths inour heads simultaneously. The first isthat we want the products that we shiptoday to be excellent and as good aspossible. We want to serve people. Andthen we also have to believe that theproducts we have today are wrong and wecan rethink them from the ground up. Andso you flip back and forth in betweenthese two all the time. And so when I'mlooking at the future, I constantlyimagine what is annoying to me today.What do I not want to do anymore of? I'mbuilding

00:27:31for myself as much as I'mbuilding for you all. And what does myteam complain about? what like when Italk to people what are things that theydon't want to be doing more of and alsowhat are the things that they do want tobe doing more of and not just the peoplewho are talking and showing it cuz Ithink it's very easy to talk and showbut when you observe people's livedexperiences of how they're using thesetools like what do they look likethey're enjoying doing and what are theparts that like aren't fun like if youwere to tell me last year that peoplewould like having four terminal windowsopen I would have been like that's crazylike who who wants to live that way andthen we saw our team do that and so Ithink a lot of it is like through yourobservation and your experience and thenbuilding like based off of like thepeers around you. So it's actually a lotof fundamentals like you you hear thislike wow that's design well yeah cuzthese core design skills are stillextraordinarily important in our day andage and like you can keep building onthat foundation you just can test newthings faster now. Okay, so there's thisthing that I'm experiencing where I'minterfacing with these models. I'mfiguring out the thing that I don't wantto happen twice, right? And I haveClaude or Codex, I make a skill, right?And then you just kind of dump it intothis black box and you hope that ithelps everybody. And the entire processof collaborating on top of the learningsthat we're all having as individuals hasbeen

00:28:46largely opaque for me. And I knowthat you are operating in orgs whereyou're taking that set of problemspretty seriously. And so maybe reallyquickly, can we go down the line? I'mkind of curious what are you seeing ordoing or experimenting or having successwith in terms of making sure that thelearnings that you're having asindividuals are propagating to the restof the org and benefiting the largerteam.>> I do think it's very isolating workingwith these models. Um, let's just alladmit it. like if you spend 8 hours ofyour day talking to claude, you haven'ttalked to another person and it can kindof feel a little strange. Uh we're also,as I mentioned, very early in thisjourney. So, a practice that we havethat we're doing on the cloud code teamis to pair for a while. What a concept,pairing with another person.>> Is that even legal?>> Yeah. Like, how often do you wish youcould work with another designer on aproject and you can't? Well, what we dois uh we'll schedule an hour. We'llshadow each other. You'll just literallywork on a thing that you're alreadyworking on. you'll see what anotherperson does because it's actually reallyhard to explain your workflow tosomeone. When I was trying to do thisdemo, it was like really hard for me tocome up with what I'd talk about. Um,but if you just watch them in theiractual workflow, they'll do things thatlike you don't even realize that thatthey don't even realize are special thatthey're learning cuz they just did it somany times now. And then, uh, we dothose every month. We cycle through thedesign team. It's

00:30:01still a little bit ofa newer practice, but the engineeringteam has been doing it for a while cuzpair programming is pretty common. Andit just helps you learn from each other.like we are all experts to learn fromright now and it also builds like a thereally important camaraderie of workingon a team. So that's something thatwe've done and it's been quitesuccessful actually.>> There's no great solutions yet like wehave we have an internal skill libraryfor example but then it's always like isthis skill up to date and should Idownload it and or should I just make myown and skills turn out to be prettypersonal in a lot of ways and so you candownload it but then you have tocustomize it for yourself anyway.There's a there's there's something thatneeds to be solved there. I think thethe thing that feels the most solvedright now for this kind of thing isactually just Slack agents. We have alot of those. Um some of them are thingsthat we build internally and it's justlike you know a claude code on a on aMac mini uh like connected to Slack orwe also use this thing called Victor umwhich just raised a bunch of money. Wealso have our own um agent a agent thatwe're building. And the the reallyinteresting thing about doing that,especially if your organizationenforces, and we we don't enforce this,but we highly encourage people to use itin public channels,

00:31:17is you get to seeother people prompting. And that is asurprisingly intimate thing right now.Um you're kind of like this is thisshould be between me and my AI, youknow? And so I think that's a reallyinteresting way to to spread stuffbecause what you do is you you get umpeople will have ideas like you saidpeople have ideas that you never wouldhave thought of and they will do thingsthat you never would expect and thenthat just like seeds something in yourbrain like oh I'm going to do that. So,I think I think that's really cool. Andalso, if you have multiple agents in uhin your workspace, they actually sharethings really easily and really quickly.And that's also really cool. So, one ofthe things I have is um I built thislike little thing called mail room thatgives my codeex an email address andit's but it's it's it's on my email. So,you know how Gmail has like the plusformat. So it's like Dan plusc andthere's like a little string and Codexjust knows to check that email and thenthe agents that we have in Slack knowthat that's my CEX's email. So they justlike email stuff that I need to do tothat codeex and then it just does it andI get it. Um and that's like anotherinteresting agents are so fast atsharing things and I think that thatsolves a lot of this problem once youstart

00:32:32getting it to work.>> What about you Bradley? What about uhcollaboration, knowledge transfer,things like that? I mean, I think I Iwould harp on the the the Slack bots.Um, when I when I joined RAMP, likeseeing people,that was only four months ago, seeingpeople um use Inspect in publicchannels, in every public channel,something's broken, um, immediatelysomebody's on it, they just tag inspect.Uh, in fact, like, uh, there's even afunny um funny instances where likesomeone will say like, "Oh, this thingshould be different. something's wrongand then umthere's there's like a a pattern wheresomeone will come in and like drop alink to something in this in this caseit's like they just add inspect and theydon't say anything else and it justreads the context of the thread and yousee you see it work and um that'sinteresting when I joined ramp um I Ikind of felt silly because I was talkingto a bot for a really long time and hisname's Cody and I was like wow Oh,people here are so nice. He's just sogenerous with his time. Yeah, I just[laughter] I so like he just anonymousPFP and I was like, "Okay, cool." Um

00:33:47hetell me everything and uh he's he's hespeaks naturally. He's all lowercase. Uhand um and then and then uh when Irealized here's a bot. Oh, I felt like acrisis. But uh I I then started to seehim evolve actually in front of my eyes.I started to see him um have a blog. Istarted to see him teach other agentshow to to work and then how to teachtheir humans how to work. Uh I startedto see him innovate on patterns where itwas like, "Oh my gosh, we have noweverybody wants a Slackbot. Everybodywants Cody. Oh, that's kind of unsafe."Uh by the way, uh and Cody's like, "Oh,don't worry. I got it. I saw this andI'm going to fix it. And then he helpsus fix it. And then um this is what'scrazy to me. Um I learned something Ilearned something great from from Codyum and from the people building him. Uhand I touched on this. There there's alot of noise, right? And and one of themost important things I learned was likenow we have new ways to cut throughthat. I saw Cody drop an albumin our Slack about everything he wasdoing. I wish I could play it for you.It's actually pretty good. And it is allabout the things that we are learning

00:35:03and need to do maybe differently.And and then he started posting videos.Somebody gave him Gemini and Nano Bananaand uh oh my gosh, every day. How do youdo this? Don't worry. Let me make a songfor you.And it's kind of catchy. I don't know.It's it's it's yeah I think it's theSlack bots for me and then we have afantastic um design programming rightwhere um you know learning is is notrequired heavily encouraged uh but it'sstarting to shape many different shapesand forms and it's like very applicable.It's like go learn a thing over here butyou don't actually ever use it. That'snot that helpful. Uh we understand likewhere things are difficult and wherewe're having trouble so let's like sitin a room. We actually sit in sometimesin here uh sometimes we have like asmaller version of this uh if you canbelieve it and uh we all sit in thereand um Elizabeth here shares theirscreen and just we we just see it happenor I see somebody do something and Iwalk up to them with my littlemicrophone and I go up to their desk andI I I know they use Loom so I turn theLoom on and then I connect my Bluetoothand then I I connect the microphone andthen I ask you, can you show me yourprompt and can you

00:36:18walk us through itand then we post it in the channel andeverybody sees it. It's interesting howand this has kind of been a trend evenon the podcast where there's so muchopportunity right now for people who areproblem finders, designers to go and fixthings internally like the the internalworkflows are so ripe. So manyopportunities for little tools solvingproblems, eliminating inefficiencies.That being said, now for the finalquestion here, I want to return to theseven out of 10 idea because it's notjust internal tools. Fact is the modelsare not there yet. They're not amazingat design, but every once in a whileClaude will do something on the papercanvas where I'm just like, "Dang,that was pretty good." You know, almostuncomfortably good. And you can kind ofsee where this is going. And so given aworld where the models are playing anincreasingly large part of our processwhen we're designing the interfaces usedby real people on our real product.How do you Megan think about the waythat the value proposition of design isshifting?>> My very personal opinion and by personalI mean it's truly just me. This is likenot a representation of the company thatI think our models will be good enoughby the end of the year or the AIindustry in general to do design most ofwhat we consider

00:37:33very like fundamentaldesign work. Um and so I think there's afew differentkind of directions I see this evolving.The first is thatfundamental systems and brand those areall subjective tastes that you need tobe able to guide and like establish andthose have a lot to do with like theautomations and the systems that you'rebuilding those are established like youcan prompt into them but like when it'sreally really an expert hand likecrafting it you can really tell and Ithink that will still be very importantand so a lot of work will go intobuilding those systems that help modelsdesign and help everyone get access tothese. Uh the second one I think is thatwe're going to enter an era wherepersonalization and customization is thename of the game. Uh you can already seeit in all the demos that you saw todaywhere people really love building customtools for themselves. But I have a verystrong hypothesis that there's limits tohow much people want to customize. Youalways need a great canvas. You alwaysneed a great starting point. And then weneed to have a flexible framework toshow people what should be fixed. It'slike you don't want your login screen tochange every single time you log in, butyou might want your dashboard to beflexible as well. And the decision ofwhat to keep fixed, what to be allowedto customize, how you guide people

00:38:48through that, that's like fundamentalUX. So, I still think there's going tobe a lot in there in the decision-m, butwe're abstracting away from justdesigning the UI layer into thestructural layers of like what is fixedand flexible UI into the harness layerof exactly how you interact with themodel to like the primitive layer oflike what does it mean for a model tohave an identity or the product you haveto have an identity. And I think whatwe're going to see is that uh builders,designers, engineers in general will endup going deeper and deeper down as likemore and more of this these higherlevels are solved.>> I hope I can speak for everyone here.Like it's a fun time to be a designerand there's just so much to learn.There's so much to evolve and dive intoand I really really appreciate all threeof you joining. I you know look up toall of you and the impact you have onthe industry and it means a lot to beable to pepp you with questions. Let'sjust give them a round of applause.Thanks everyone. [applause]>> [music]