Loading summary
Transcript58 lines
- [00:00]
Kyle Turman
The guy goes, you know how to code, right? And I was like, yeah, yeah, I do. He's like, all right, just like, start building it. No process, no reviews, no, like, testing. So I start building it, and we end up deploying that feature, like, the following week.
- [00:15]
Rid
When you're the first designer of a product, whether you like it or not, like, there's a piece of your DNA that gets fossilized in that early, you know, brand and architecture and just how things feel.
- [00:27]
Kyle Turman
And these are, like, subtleties. And a lot of people will probably look at that and be like, oh, oh, that's just like one of those designers wearing a beret and drinking cappuccinos and, you know, doing that fancy designer thing. Like, they're just silly. But this is real. You know, there's science that backs this up.
- [00:42]
Rid
All of a sudden, you're designing a product that is inherently non deterministic. So I'm curious, how has that difference impacted the way that you practice design?
- [00:52]
Kyle Turman
Coming up to an AI model is like, you go into a cave and then there's this rope, and then like, 100ft off, there's this giant wizard sitting on a chair. And then there's a tiny little sign that says, say anything. Like, that's crazy. That's so intimidating, right?
- [01:08]
Rid
Welcome to Dive Club. My name is Rid. And this is where designers never stop learning. This week's episode is with Kyle Turman, who was the first full time design hire at Anthropic. So if you're interested in designing AI products, this is the episode. We're going to get a behind the scenes look at what it was like bringing Claude to life in the early days and all of the hats that Kyle had to wear during that time, we go deep into his process. What makes the anthropic design culture so special and all of the things that Kyle's had to learn while working with AI as a new material. So let's start this episode by getting to know Kyle a little bit and learning how he landed this opportunity.
- [01:51]
Kyle Turman
I worked at this place for six months and realized a few months in, like, this just wasn't gonna be the place for me. And recognizing this difference between this thing is bad and like, oh, you know, this place is terrible. And realizing, like, no, that's just not for me. And that's just like a mindset shift that I started to try and have over the past few years is like, even when I listen to a piece of music or I see a piece of art, oh, this thing isn't bad. It's Just not for me. And that's okay. And we're able to kind of say that and move on. But at the same time, I did feel like a massive failure in the sense that I took a big risk going to this startup, leaving a job that I really loved at Slack, and it didn't work out and I felt really terrible. And so I kind of freaked out right after I left. And I interviewed. At one point I was interviewing 12 different companies, which I would not recommend. I was very stressed, with great support from my partner. Kind of put pause on that for a second and just said, you know, what do I want to do right now? And the thing that kept coming back to me, I was interviewing for a lot of these director and VP of Design positions and I just started making stuff again and I had so much fun. And so I just started making little side projects. I was working on this little app, that theory. I never finished it because, you know, adhd, but I, I, I wanted to make an app to like let friends kind of hang out together. Serendipitously, there was a friend of mine who connected me with this other startup who was actually working on this concept and, and I was like, okay, cool, maybe I'll just do like a quick two or three week contract with them. So I did that. It was so much fun. I built out all this design. They didn't have a designer at the time. I built out all this design framework for them, product direction, like I just went, I went ham, like I went crazy with it. It was so much fun. And they were like, oh wow, we have like enough designs to work off for like six months now. This is, this is great. And through that process I recognized that I really liked freelancing, working with all these different companies, being able to like jump into different problems. And I was like, all right, this is it. I found my calling. I am going to be a freelancer. And I did this, you know, probably like 10 years ago and it was super fun. Only did it for a year and then I landed a job. I have a couple of clients over the summer. We're having a good time. There's a friend of mine, Julius Tarn, who was also freelancing at this place called Anthropic. And I read about this place and I had done some explorations with early GPT and a few other kind of AI tools. And the mission of Anthropic was just really fascinating to me because it was unabashedly saying, hey, there are real concerns that we need to address around AI. And where it's progressing and how it's progressing. And this was before everything was really as wild as it is today. You read that and you're like, yeah, that seems reasonable. Like, oh, it's interesting. These people are concerned now. I look back at that, I'm like, okay, yes, this was correct. And these things are getting better and these systems are improving quite rapidly. I was like, this seems like an interesting place. I start talking to them. And so I interview for a position there. Lo and behold, I didn't get the position. And this sounds kind of a little, a little weird, but like, that was the first time I had ever gotten rejected for a position that I actually wanted in my entire career. And so it was very humbling, a very humbling experience. But I was interviewing for a management position and I remember thinking, I just really loved this company. I loved everyone I talked to. I knew Julia's contract was ending and so I messaged the hiring manager and I was like, hey, I know that you said no to me for this role, but if you need someone to fill in as an IC for the next couple months as you find people like, I'm, I'm going to do a contract position with you. That was kind of like an interesting opportunity because I can still do my like freelance. I dream an idea, I'll do that for a couple months and then I'll go to the next thing, right? This is my new future. I make it a week into Anthropic and out of the blue they were just like, do you want to just like stay here forever? And. And I was like, well, I don't know about forever, but I'll definitely consider an offer maybe just because I was having so, so much fun and the people there were so amazing. And so I like to consider myself kind of like an anthropic foster fail, as it were, where they kept on to me for, for a while. And so I ended up becoming the first kind of full time product designer there in Anthropic.
- [06:12]
Rid
What was the state of Claude then?
- [06:14]
Kyle Turman
Oh, wow. So that was like Claude 2. And Claude 2 was pretty decent, but it was very rudimentary, right? Especially like the early, like GPT3 and stuff like that. You could do things that were interesting if you knew how to prompt. You could get it to do some analysis of data. You could definitely generate a lot of really interesting content. But things were a little wonky. It would hallucinate. The ability to follow what we call the constitutional AI for being helpful, harmless and honest was definitely a little Bit more too literal. And so when we came out with Cloud 2.1 shortly after, it was kind of known for refusing every single request because it was being a little too much of a Boy Scout. It felt interesting and it felt compelling, but it didn't feel transformative yet.
- [07:06]
Rid
Real quick message and then we can jump back into it. For a long time now, Raycast has been my portal to AI. I mean, all it takes is a quick keyboard shortcut and all of the models are instantly available at my fingertips. But the problem was you had to be on a Pro account. But that ends today because they just announced that Raycast AI is available to everyone. No subscription, no account needed, you get 50 messages to try, and you even get access to all of their extensions. I mean, I'll put it this way, if I'm going to get a new computer tomorrow, the very first thing that I'm going to install is Raycast. It is that good. And you can try Raycast AI for free. Today. Just head to Dive Club. Raycast. That's R A Y C A S T. So you know how I've been talking about how I use genway to do research with AI? Well, something surprising is happening. And just for context, I use generally in two different ways. One is contextual interviews. So I prompt the AI with what I'm hoping to learn and it has a dynamic conversation with each person. And the second is usability testing. So I upload a Figma prototype and their AI agent helps me test them with people all over the world. I mean, you should see the quality of the follow up questions. It's pretty crazy. But here's the surprising part. At the end of the interview, most people say they're more comfortable opening up to an AI agent than a real human. And it's just another reason why I'm hooked on the product. If you want to try it out, there's even a secret landing page just for Dive Club listeners, which gets you 2 months free and 10 credits. To recruit people, just head to Dive Club slash genway. That's G E N W A Y. Okay, now on to the episode. So what were some of those early design challenges that you were working on? Like kind of put us as a fly in the wall in those first few months.
- [08:57]
Kyle Turman
I walk in the door, there was no sign at all, you know, anywhere that said that this was anthropic. And have a small little onboarding group. I think with six people, we do a little, you know, presentation thing where they walk through security. I have like few hours left at the end of the day. And I talked to who the hiring manager was at the time, who was the engineering manager, because there was no, no other designers except for Everett, who worked on brand, who also was basically like the only design voice in the company really at the time. So I sit down and I'm like, hey, you know, what should I, what should I do? How can I help? How can I be helpful? They're like, okay, we're doing this new thing where we're adding payment to Claude. They had just launched Cloud AI and we wanted to be able to add like a pro subscription so that people could get more usage and we could also start kind of like monetizing understanding this platform. And so I started designing all the payment flows and I was having a ton of fun making some little animations for confirmation states, things like that. Knock it out in about four or five hours, send it. And I was like, okay, cool, like, what's the next step in the process here? Like, how do we get like feedback and approvals and stuff like that? And this was like 7, 8, 8 at night or something like that. And the guy goes, you know how to code, right? And I was like, yeah, yeah, I do. He's like, all right, just like start building it. No process, no reviews, no, like, testing. So I start building it and we end up deploying that feature like the following week. And so I just jumped right into the code base, worked on a lot of the front ends with an amazing engineer named Brian, and we built an entire kind of end to end payment processing flow along with like cancellation, changing your credit card, like all these different steps in about a week. And so.
- [10:39]
Rid
And that's processed some serious money too.
- [10:41]
Kyle Turman
It has, it has, yes. So definitely a hell of a starter project. That's kind of just how it worked for a while. Where shortly after Joel Lundstein joined, who was on this podcast, I think a few months ago as the head of product design. And eventually we hired a couple other product designers. But as the time went on in these early days, it was just like, we need something built. Design it, build it. There was no design system. I slowly built up a design system. Reusable components in figma, reusable components in React. We were only web at the time. If we wanted to launch something like, say, for Cloud Pro, not only did I do the front end code, the actual UI designs, I had to also do the marketing designs, the social assets, the like, strategy, the naming. Right. Like, we debated, what should we call this thing? Should we do something special and call it more than like Cloud plus or Cloud Extra or something like that. And we're just like no Cloud Pro. Okay, let's figure that out. Right? But I'm having these conversations and I'm also like literally the next second adjusting CSS pixels, you know, and so the like elevation of what you're having conversations and working every single day was just like all over the map. It was definitely stressful in some ways, but I think was also probably the most fun I think I've ever had because you just get to go for it and do all of the things that we talk about doing all the day. I've said this for a long time, like a design in Figma is never going to be seen by an end user unless you do one of the Figma file deep dives. But other than that, right, it's never going to be seen by the end user. And so I think we spend so much time like twiddling our thumbs and working on these small details in Figma and then by the time it makes it into production, it's totally different. You also learn so much more, especially with these tools that inherently are non deterministic. It's been really interesting kind of seeing that happen so fast where really I started doing more designs in Figma when we hired other designers so that we could work together in Figma. But most of the time when I was designing stuff, it would just be like, can I get to the first point to where I could get this into code? And then I would just finish it in code. And so a lot of the early designs, if you, if you saw them, are just like kind of half finished because they would just be just enough to get me into the code base because we were just moving so, so fast and growing so, so fast.
- [12:58]
Rid
It's kind of a theme I'm noticing, of the people who were some of the, like, early makers at these now like, you know, popular scaled AI products. Figma was only introduced as a core part of the process because they needed a source of truth in the design team and maybe everyone could code at that point. But like in the beginning you're, you're kind of starting with code and so it's like another notch. I'm like, okay, there's a clear trend here happening.
- [13:22]
Kyle Turman
It's just, it's just so much faster and sometimes speed is good and sometimes speed isn't what you're looking for. Right? We definitely spend more time in Figma because we try and validate ideas more, we try and get more feedback in the early days, there was no time for validation. It was like, we got to build this. This needs to exist. Please make it so. Right? And that type of work in that type of environment, just as much faster to do in code than it is in figma.
- [13:47]
Rid
For sure, it sounds like an incredible role because you have the speed component. You're wearing all of the hats, like, the spectrum of hats you're wearing is so wide. But then also it's pretty clear, like, you're given this level of autonomy where I'd imagine you were making thousands of micro decisions a week about literally everything.
- [14:03]
Kyle Turman
It's been really interesting and also humbling because now we have a product design team of about, I think, 12. And there's a lot of decisions that I made on the fly that I thought maybe are slightly better than a default decision. And so I would talk a lot with Everett and on our brand team, we really tried to make the Claude ethos be something that was more than just a tool. We wanted to embody kind of this aspect of humanity into what we were doing, because that's so core to our mission. And so we made a lot of smaller decisions that ended up becoming bigger decisions in terms of, like, the background color for Claude being slightly more beige and warm rather than kind of cold and white. The responses for Claude being in serif instead of in sans serif, because that's how a lot of news organizations and long form reading for thousands of years has kind of been introduced. And then only more recently have we been using a lot more sans serif for long form reading. And so these types of decisions were definitely made without a ton of committees and research and framework. It was definitely more from the hip. And I'm seeing now some of these decisions. We're refining them and questioning certain aspects a little bit more. But a lot of the core decisions still resonate with people. And I think about that too sometimes where, yes, we could have done so much market research and proven out all these things to be like, oh, yes, definitely, this is the way to do it. But because we had a high level of trust with each other and because we were able to really tap into what made this meaningful and had this mission that was driving us, I think it made it a lot easier to make those decisions. But there's a ton of other smaller decisions that people now are like, why does this look that way? And I was like, because I had 30 minutes on a Saturday. Like, that's why.
- [15:48]
Rid
Like, can you go a little deeper on what those, you know, like, guiding principles or North Stars were that you were working backward from as a designer who's making a ton of decisions every day. Like, what was guiding you?
- [16:01]
Kyle Turman
In the early days, when I joined, the product team was quite small. I think there was maybe like 5 people total working on product, but the company overall was around 200 people. A lot of that work was on the research side, obviously, but we also had a pretty strong policy arm and we also had a pretty decent trust and safety team who was there before we even started working on product, which is basically unheard of. Right. And so there's this high commitment to safety. And when I joined, I really wanted to understand what that meant in terms of like, you know, there's a lot of things that we're thinking in terms of like preventing long term harm and preventing current harm that these models could potentially do. And one of the values that we have as a company is kind of ignite a race to the top, which is really thinking about how do we show the world that you can still do really amazing things while also considering the potential shade as well as the light. So Dario, our CEO, has a really great post about this called Machines of Love and Grace where he talks about the reason that we think about the darkness, the potential bad things that could happen, is because we need to also hold the light, the potential good that can happen. Right. And so we can't just think of one, we have to think of both. And that includes, like, a lot of people who I think are very intent on, like, this thing just needs to evolve and get better. Like, you know, safety be damned. I don't think that's the right approach either because we're trying to consider both of these aspects. Right. Safety is not a verb. Right? Like, safety is the result of something. How do you make something safe? What is that emotion? And I think this is a thing that I really loved having Everett on board for early is because I love talking about emotions. As designers, we don't talk about them nearly as much. I think the only time we talk about emotions is when we're doing that typical journey map of like, user sad, user happy. Where are they at on the journey? No, every single thing that you interact with has feeling and emotion tied to it. Right. Like when you turn a kettle on in the morning and it just, just has this nice satisfying click to it that gives you the sense of calmness that you don't think about, but reverberates through your entire day. And all these kind of decisions we make as designers have all this emotional impact to it. Whether we think about it or not, it's there. And so I really loved that Everett and I would talk about, like, how do we want people to feel? What is the feeling of safety? And a lot of what we talked about is like, well, we want these models to feel human in the sense that they're not going to be something that replaces humans, they're going to be something that works with humans. So this kind of organic quality. And we also thought a lot about history in terms of how do we root something that is going to change the course of the entire rest of civilization for here on out in the way that civilization had been thought about before. And we worked with a really amazing branding agency called Geist, and they helped create the anthropic logo and then also created the Claude logo. And so it has this. This cute little spark. And we kind of made that to be like this spark of creativity, spark of imagination. And the entire vibe of that is very more serif. So it kind of anticipates a little bit more of that old world feeling. And that creates this sense of a deeper human connection and trust. And these are, like, subtleties. And a lot of people will probably look at that and be like, oh, that's just like one of those designers wearing a beret and drinking cappuccinos and doing that fancy designer thing. They're just silly. But this is real. There's science that backs this up, that thinks about different types of things, trigger memories and associations, and those are tools in your toolkit as a designer that a lot of designers just don't think about. A lot of our early discussions were really around how do we make people feel the way that we would like them to feel? And really leading up to this sense of the verb of safety being trust. So how do you make someone feel like they are safe in an entrusted environment? And there's a lot of design decisions that you can make in that. I think I was really happy that we made the time to make those and that we had the discussions about it without someone being like, oh, that's just hobbly, bobbly, weird stuff.
- [20:09]
Rid
You use the word subtle, and maybe in isolation, each decision is subtle, but the culmination of them really does feel different. Like, I'm like the early adopter, right? Like, I've used all of these different tools. And Claude kind of does sit in its own little camp esthetically, and a big, you know, supporter of everything you're talking about in terms of, like, rooting it in history and the type of feeling and Safety and everything. But, like, also strategically, there is just a unique factor that exists because of those decisions in a sea of sameness, you know?
- [20:40]
Kyle Turman
Totally. And I think that's the thing, too, is, like, not everyone vibes with that. We get tweets and emails and things of people being like, oh, the UI is terrible and it sucks, and I hate it. And, like, it burns my eyes. I don't know how many times I've designed something that someone has claimed has burnt their eyes. So I feel like I am the result of many people's cornea issues. And I'm sorry about that, but I think that that's kind of also the point is that we, I think sometimes as designers try and design for this, like, global mean where we just reduce everything, you know, to. To what is most palatable to most people. And I've often said that you can't design something that someone loves without a few people hating it. It's just not possible. Right? And so I think that that's, like, part of the way that we thought about this too, is that we wanted to be able to take enough risks that some people would really love it. And you have to give them something to attach to to make that happen, right? And that can't be the only thing, right? Like, we still have to think about business and strategy and other things like that. But it's an important tool that a lot of people don't consider or they consider to be worthless. But it has this emotional quality where you're like, oh, this is different, and it's different. And I want to associate myself with this, right? It's the same reason that I talk about a lot, like, why would someone wear vans instead of Nike, right? Like, they're pretty much the same shoe, especially, like Nike Skateboard or whatever. But you wear vans because it makes you feel a certain way, right? There's that kind of ineffable quality where people want to attach themselves to something that is meaningful. Especially in this world today, where every single person trying to broadcast themselves and sell themselves to become influencers on LinkedIn or whatever, right? And they'll say and do whatever it takes to be noticed. I think there's this craving for authenticity and this craving for something that is real. And whenever someone is authentic, I talk to a lot of students about this. When you're your authentic self, people will call you out for this and won't like it. And that's scary. And that's why a lot of people aren't their authentic selves often enough. But it Also attracts the people who you want to be attracted to, right? Like, if you are more yourself, you're going to attract the people who see that and value that. And so I think it's the same with brand, I think it's the same with design.
- [22:57]
Rid
As you're talking, it's reminding me of something that Tuan Kumar, who is upcoming episode, he's the head of design at Lumalab, and he said something that kind of stuck with me. He talked about how when you're the first designer of a product, whether you like it or not, like, there's a piece of your DNA that gets fossilized in that early, you know, brand and architecture and just how things feel, you know. And sure, there's a lot of strategy and you can look at anything and say that it's strategic, but also you can look at things and be like, you know, there's a piece of Kyle at the core of what Claude is.
- [23:30]
Kyle Turman
You know, I remember having this conversation years ago with someone about how they were adamant to say design is not art. It is not art. And they were really passionate about this. I think we've had this discussion a thousand times. This, to me, was the discussion before code became the question. Should designers. Code is like, is design art? The reality is it is and it isn't. There's a part of everything that you build, whether you try to or not, that has a bit of you in it, because if it doesn't, it's not good. We love things that are designed because of designers. Right? Like, things don't just naturally evolve to be designed, to be beautiful in this kind of physical environment. Right. They do in nature. And so there's, there's a bit of that that is interesting. But if you're creating something new, it requires someone to intentionally create it. Right. And it is impossible to not imbue a part of yourself into that. Now, that doesn't mean. I think there's also like the reaction to that is that I think back in the day there was a lot of designers who were very full of themselves, who were like, well, I'm going to design what I want to design, users be damned. It doesn't matter. Right? And then I think that's kind of the whole human centered design approach, like, let's get feedback from users. But I think we overcorrected on some of that where we have not really acknowledged that there is a piece of us in everything that we design. And I think that's also why it's like so awesome and hard to be A designer is because you put some of yourself in it. Even if it's a form on a admin page, like there's a bit of you in that because it's enjoyable, it makes it enjoyable. And I think when humans do work that they enjoy it is higher quality.
- [25:09]
Rid
Your background is very different in many ways. You know, like you're coming from more like leadership, big companies, Slack, Dropbox, and we've already kind of talked about some of the differences now that you're the only designer fast moving startup, kind of to an extent given a blank canvas of what the heck to do with this product. The other big difference though is all of a sudden you're designing a product that is inherently non deterministic. So I'm curious, how has that difference impacted the way that you practice design?
- [25:38]
Kyle Turman
I think I used to get really into the details in Figma and I realized that the details only matter if the users see them. Right. And it really changed my perception when I started thinking about that principle as it went further is like, how do I make sure that like the decisions that I'm making on a daily basis are going to be the highest value to users and that changes your framework. When you start thinking as kind of the only designer at a startup, right? Like your understanding of what is important kind of goes out the window because you have to kind of reframe like what's going to have the highest impact for the company, what's going to be the most beneficial to help users do the next thing that we want to do. We have another value at our company where we talk about we want people to be helpful, honest and harmless. And a word that I say a lot is helpful. Because I have really changed my perspective where I see design as being helpful essentially, right? Like we can do things that literally no one else can do. And there's something magical about that that we often forget where like we can take a sentence in a document and we can turn that into a real thing that actually has buttons and scrolls and interactions and all of those things. And so what I found myself doing a lot was as we're trying to figure out what this thing is, how it works, what are the principles for interaction. Obviously I think people maybe are coming back around to this idea now, but especially early last year there was this huge push for designers to be like, yeah, we need non chat ux. The chat UX is just terrible. I think we've scratched the surface for what's possible with AI. But I think that it's also really egotistical. To kind of say, like, oh, we need a new UX paradigm, without actually understanding why or what the problems are or what the benefits are. And the reality is, why chat works and why chat took off is because it is both a way of working that anyone can do, no matter who you are. Like, you can just type in a box, right, and get something. And I think that this is the thing that's really hard with AI models is that they can do anything. So it's really amazing that you have the ability to say anything, but it's also really intimidating. I said this before, where coming up to an AI model is like, you go into a cave and then there's this rope, and then like 100ft off, there's this giant wizard sitting on a chair. And then there's a tiny little sign that says, say anything. Like, that's crazy. That's so intimidating, right? Like, how do you overcome that? And so there's definitely a challenge with that. But the benefit is you can do anything. And this is what we see from our customers. I see reports of Claude helping people with homework. I see reports of people helping Claude with, like, relationships. I see people working on apps, you know, building out little components and stuff. Like, the variety of stuff you can do is so broad that I think that there's an inherent benefit to the way that kind of the chat ux, or just rather the input UX works. And then also I think the benefit of chat is iteration. These models aren't perfect. And so if you can get something in two or three turns that you maybe couldn't get in one, chat inherently makes that conversational, right? Where you can be like, no, more like this, or actually, can you do that instead? And that kind of way that it builds that context as you're talking, like, inherently makes it more iterative and collaborative. And so I've explored probably, I would say, hundreds of different UX patterns, right? Shown them to people. We've prototyped things, and design, I think, is a really good ability to kind of basically answer questions like that. Like, is there a better approach? Let's try it. Let's see. You know, like, let's have all these ideas. And I think as we continued, especially in the early days, we just found that there wasn't anything better than what we had. And we found different ways to improve it. We added multimodal. We considered kind of this artifacts pattern that we launched last year where you can iterate on a static piece of content while you're working on the left rail. And that Pattern has now been picked up by the biggest companies in the world, right?
- [29:53]
Rid
Yeah.
- [29:53]
Kyle Turman
And so there's these little things that you don't think are that transformative at the time, but end up becoming these things that build on each other. And so I think that it can be foolish sometimes to just throw things out because they seem like they're not novel enough. And really, I think the job of a designer is to explore and ask questions, but then kind of be a mirror to the organization and say, this is what it would look like if we did this. Is this good? Is this not good? And so I think that's a lot of what I did in those early days and maybe still what I do today as well.
- [30:22]
Rid
I love the hundred plus prototypes backstory, because that's another one of those trends where it's like, you know, last summer I'm having people come on and talk about CHAT as like, the core interface pattern for AI, and everyone was like, we have to push past it. We have to push past it. And it's like, yeah, the pendulum is swinging back. It's like, actually, this is pretty good. Maybe, like, maybe this really is the right foundation. And so to hear you push on that and maybe even feel a little tinge of pressure to push, arrive at the next thing and then actually come to the place of like, okay, you know, this is pretty good, is super, super interesting.
- [30:58]
Kyle Turman
I think pressure is an interesting word too, because I would go to, you know, meetups or parties or. And I would talk to other designers and they. They'd be like, why don't you just do this? Like, why don't you just do this? There was so much outside pressure to. To do more, to explore further, and that's just not what we were trying to do. I mean, obviously we do that internally and we try and understand what we can learn to improve the models. And I work really closely with researchers to understand the right ways to do pre training, fine tuning, those types of things. And a lot of that work is never seen until it actually finally launches or it's integrated in some other way. I think that there's a lot of expectation because of just the vibe of what AI is, where you're like, it can do anything. This is amazing. People expect it to be a flying car when in reality. And maybe the Waymo is just fine.
- [31:50]
Rid
You know, the interesting thing about CHAT is so many of the design decisions are kind of unseen because it's just in the quality and the nature of that interaction. So is there a story or an example for People who are curious, like, what's it like as a designer where part of the experience that you're thinking through is in this completely non deterministic back and forth and not in pixels at all.
- [32:16]
Kyle Turman
We do a ton of prompting. This is a thing also that I think is quite different in terms of we'll get a PRD or a kind of vision for something that we maybe want to explore, that leadership is interested in. And part of what we do on design is we work with research. We do this ourselves as well. Is this possible? We can design the perfect thing. I've done this before, right? I'm like, cool. Here is the exact thing, here's how it should work, here's the flow, here's all the animations. It's going to be perfect and then we get the model to see if it can do it and it's not quite there yet or like it. It can do it, but only in this way or something like that. Right. And so a way I talk about this a lot too is like, imagine you're working at a startup in the just normal SaaS business using like kind of a REST API and you work with an engineer and you say like, okay, cool, here's this design for this form and they're like, yeah, we're going to see if we can get that data back from that, the server and we're just going to ask it a few times and see if it gives us that data back. No one does that. That's not how it works. Right. And so you really have to kind of rewire your brain where like you have to think first from a model perspective and realize that actually sometimes like we don't need design features, we need prompt improvements. A big thing that we did I think last year was we worked with Amanda, who's on our research team, and also I think Lisa on our product team to add a small change to our system prompt where Claude asked questions after a response. And there's an aspect of that that is, can be annoying sometimes, but also can be really amazing in the sense that it makes it feel engaging and it also like helps you realize like what you can do. Right. I had to take my partner to the hospital last, last year and it was really scary. And I'm talking to Claude about this and saying, here's what's happening, you know, whatever symptoms you know, and here's what the doctor's saying, like, help me understand this. And Claude would ask follow up questions at the end and be like, would you like to know what will probably happen next or would you like to know some good questions to ask the doctor or things like that? Right. And it just made it feel totally different than just having a response that just ended. Right. And that's not a traditional, like design solution. It is a UX solution. Right. In the sense that your user experience experience is so different. So there's so many things like that where we're thinking about kind of how the model interacts with people and how we can take what we've learned from our research. We have an amazing UXR researcher named Jane who is really helping us understand what people want and how people react to these things and how we evolve the model to be better at that. Those things are kind of unseen and don't really have pixel value to them, but are still designed in some ways.
- [34:59]
Rid
Can I put you on the spot with a little hypothetical something that I'm working through right now in terms of problem engineering and you know, I have way less experience and I'm genuinely curious how you would approach this type of problem where I'm thinking through a lot of design feedback, tooling right now. And so like, let's say that I make a two minute video where I'm walking through some kind of a Figma prototype and I'm playing with the idea of, you know, how can I then give the model that transcript and then turn it into the right questions and points of alignment for the rest of the team to then give feedback on. Simple as that. Maybe a two, three minute video context. How do we get the right two to three questions out of that that are the Eigen questions, those tipping point questions that really move the needle, blank slate. How would you as a designer even start to think about turning that input into that output?
- [35:54]
Kyle Turman
So this is why I think designers are really good at this, is that when we look at a problem, we don't immediately jump to a solution. Ideally, some of us do. It's kind of looking at that as a problem and trying to understand what the right approach is based on the core qualities of how the model works. And a way that I've explained this a lot is if I were to go up to my teammate and I say very simple phrase, where are we at? Right. They're going to know that I'm probably asking about the project we're both working on and that I want an update and can give so much information exactly how I want it, right? If I go to a random stranger on the street and I say, where are we at? They're going to Be like, California, I don't know, like totally irrelevant answer. That's not what I want at all. Right, same phrase. What changed? Right. It's the context. And so I think a lot of people don't quite understand how to build context for a model. And a lot of it is very similar to how you build context for a human. So in that situation, it's like, what would a human need to know in order to give that type of feedback that you're looking for? How do you explain that? And what I do a lot is that I'll actually work with Claude to do some of this. So I'll generate a prompt or I'll write a prompt out, test it out, and I see that it's not quite working exactly how I want to. I'll take the prompt in and I'll ask Claude, like, here's what I'm thinking. This isn't working for me because of these reasons. I'd really like it to be more like this. And cloud will help me kind of transform the prompt and make it a little bit closer to that. But I can see directionally what it's trying to do. And I pick up on certain clues that Claude will be able to notice that maybe I don't. And then there's some times where you just have to get Claude to look at something differently. I was working on this other side project where I'm trying to generate kind of tips for parents as they raise their kids and things like that. I was really struggling because most of the tips just sounded like a little superfluous. And I told Claude, I was like, now put yourself in the shoes of a human being. This is the situation you're in. Boom, boom, boom. Like you're tired, like you don't have much attention. You're really wanting the best for this person. I'm like explaining this in the same way that I would explain it to maybe like a 13 year old who has 3 PhDs, but like also maybe is on the spectrum, you know, like, they need that context of like, this is what the human is experiencing right now. Being able to kind of frame it that way, like put yourself in their shoes and then think about it this way and it totally changed the result. And then you ask glad to help formulate that into a prompt. And it's very iterative, I think, where you're kind of going through these different stages of like, very similar to like you start with a wireframe and then you kind of work down into details and figure out like, okay, generally it's working. This part needs some work and you kind of refine it over time. But most of my prompts are probably multiple pages by the time I'm finished with them. When I'm doing really advanced work like.
- [38:44]
Rid
That, really, I want to just keep pressing on this because everything you're describing is like a whole new set of skills for designers. And, you know, you're playing with this more than probably anyone that I've talked to up to this point. So, like, what are some of the other ways that you are using Claude in your design process and any lessons that have come out of that back and forth that you're having on a.
- [39:05]
Kyle Turman
Regular basis once you start getting into a larger company like we are today? I think we're a little over a thousand people now, and our product team is, I think, maybe close to 100. We've grown a lot. And there's a thing that I say, which is we just came out with a really great tool, Claude code. I use Claude to code quite a bit, which is really helpful to do things that maybe either I couldn't do before. Like, I'm writing Python a lot more than I used to, and I'm not really a Python head, but Claude really helps me do that. And so there's some things like that where I'm able to, like, use cloud code and. And three, seven sonnet and really do a lot of things that I couldn't do before. That's pretty well known, I feel like. And that's like, a lot of things that people talk about where, like, oh, yeah, Cloud helped me do something I couldn't do before. I think the big unlock and the big transformation is realizing that a lot of the problems with efficiencies at companies are not, can I write enough code or can I make enough designs? It's, can I talk to people? Can we get alignment? Are we all working on the same direction? Do we have trust in each other? These are fundamental problems that probably make up at least 50% of all efficiency problems at any company. Right. This is one of the ways that I started using cloud recently, where I kind of have a few profiles that I've built up of people that I work with. I'll include those profiles and then I'll say, like, I'm really struggling to explain this concept to this person. Here's. Here's what I'm saying. I don't feel like they're resonating with these concepts. How do I explain this differently? Or how can I see their perspective better? That has been truly Transformative because it is actually rewiring my brain and I'm thinking differently. That's when I think these models are going to become really, really transformative, is when they really intercept not just like the things we do, but how we think and help us become better humans. There was actually a specific person I was working with that I was having a real challenge working with early on and I just could not figure out why we just couldn't get past something. And they just kept bringing up the same thing over and over again and didn't really trust me for some reason. And I talked to Claude about it, had a good 15 minute conversation and then we had a follow up meeting in an hour. And I said, okay, I'm meeting in an hour, what should I say? And I did this kind of test where I was just like, I basically nearly verbatim said what Claude said I should say immediately it was solved. And I think the problem was, is that I wasn't really seeing their side or understanding their point of view well enough. And just by talking with this model, which doesn't have that same kind of heightened emotion quality to it, helped me see that perspective and explain it in a way that made sense to them. Would it be nice if that person saw that perspective without me doing all that extra work? Yeah, that'd be great. But that's not how humans work and that's not how life works. Right. And so I think it's been really transformational for me to use it as a tool for life.
- [41:57]
Rid
Did you play with the Sesame.com conversational demo that came out recently?
- [42:01]
Kyle Turman
I did, yes.
- [42:02]
Rid
That was my experience there where all of a sudden I was like, wow, in the future, am I going to be able to embed little elements of a given person on my team's perspective as a voice agent that I can then just like brainstorm with or get someone's perspective on. I thought about it for days after having that demo. I talked with that demo for about an hour. Hour. I don't know. I don't know what it looks like, but it's like a. It gets a little bit weird, but also it's kind of amazing too.
- [42:32]
Kyle Turman
It's a very weird world we're in right now, for sure. We acknowledge that a lot. I often say that things are weird and only going to get weirder. And if you embrace the weird, then it becomes actually like a tool instead of an overwhelming thing. And so I hear things like that, I see things like that. I'm like, yeah, that's very weird. And that's. That's expected, right? That's going to keep happening. But I think if you think about it in that sense of, like, why is this weird? How does this feel weird? What is my reaction to that as a human being? It helps you as a designer better understand these things too, in the sense that the goal is not to have that problem be solved, in the sense that it's like, oh, okay, cool, I understand my teammates better. That's great. What challenges does that create by having this thing that I talk to instead of my teammates? Right. And so it's like, I think that sometimes we think about technology as solving all problems, but really, sometimes it's a bit of a whack, a mole where it solves one problem and then causes three other problems. And so that's why we also try and do this a lot. At Anthropic, we have a company value called Hold Light and Shade. And that's kind of what I was talking a little bit about with Dario's post Machines of Love and Grace, where I think you have to understand, oh, yeah, this is really cool. And that would be great to be able to, like, talk to my teammate and test a conversation. There's definitely a benefit to that, but, like, what is the cost? Right. And so as we think about designing these features, that's something I think a lot about, which is, like, how do we get as much benefit as possible and reduce the amount of potential weirdness? So if we design something like that, like, I would love to figure out. We talk a lot about this too. Of, like, I see Claude not as a robot butler who is waiting for you. I see it more as like, a nano suit that you wear that, like, makes you better. Right. By changing that perspective, it centers the human. Instead of thinking about our problems as just, like, things to be solved. Right.
- [44:26]
Rid
Real quick message, and then we can jump back into it. I'm a big believer in the power of video to explain my thinking as a designer. So when it's time to get feedback, I'll drop a loom link in Slack and. And another link to a Figma prototype, and feedback will be scattered everywhere. And, I mean, it's a mess. So I'm building the product that I've always wanted to exist, and it's called Inflight. You can kind of think of it like an async crit. It's an easy way to share a video walkthrough along with an interactive prototype or whatever you're designing. And then AI interviews the people on your team to get you the feedback that you need and organizes everything for you in a beautiful insights page. So right now I'm only giving access to Dive Club listeners. So if you want to be one of the first to use Inflight, head to Dive Club Inflight to claim your spot. Okay, now on to the episode. I want to talk about design at Anthropic more broadly because, you know, we've spent a lot of time talking about kind of early days. You're up to 12ish designers now. What are some of the elements of the design culture at Anthropic that have kind of taken shape over the last couple years?
- [45:32]
Kyle Turman
When you're the first designer, you set some tone in some ways.
- [45:36]
Rid
Yeah.
- [45:37]
Kyle Turman
And the tone I've tried to set is in that same category. Let's keep things weird. Let's embrace the weird. I would say that's something we try and do a bit on. Design is like, we. We are serious in a lot of ways, but we don't take ourselves too seriously. And I think if we did, if we were just, like, all super intense and super serious about everything all the time, then I think we would forget the weirdness that we're in and, like, forget to acknowledge that. And so that's one quality I really hope we maintain is I want to make sure that we have the ability and the psychological safety to continue to, you know, suggest weird things. What if we did this and this, like, flew in like this, and then it faded in? Or like, oh, what if this product that we're building is actually more like this thing, completely different and igniting? Kind of a curiosity to be able to say, yeah, that could work. That's interesting. That's a different perspective. That's something I think is very important. At the same time, we move very fast and still to this day, right. Like, there are things that ship in weeks of them coming to someone's mind. Right. I think we do a lot more targeted ships these days, and we're really thoughtful about what we put into the product because we just have millions of people using it. We didn't have to think about as much even literally a year ago, as we think about how we make things actually useful. A big phrase that we've been saying, which Jordan on our design team and Kim on our design team have been saying is, like, how do we meet people where they're at? And I think that it's a big thing that I'm seeing. You see this all over Twitter. You see this Reddit, wherever, when people are making AI tools They're so focused on capabilities of, like, here's what I got this model to do, right? Very rarely do we say, here's what this model did for me. And just changing that framework, right? That. That language is something that I think we try and do a lot. And so even in our critiques, a big thing that we've just, like, really gotten more specific on is, like, is this too complex? Is this more complex than it needs to be? Is this trying to be a feature for feature sake, or is this actually going to help people? And so I think we've pushed back and we've actually not shipped a lot of features because we realize that the, you know, it would probably introduce more complexity than it was needed and required, like, further thinking to figure out how we solve it holistically, rather than just like, shipping the next thing.
- [47:55]
Rid
It reminds me of something that Joel said to me, which I guess almost been a year now, and I'm curious to get your perspective on it. He talked about how working on a product like Claude, all of a sudden it kind of became a little bit okay to have solutions in search of problems because you had all these capabilities that you were getting hit with. So what's your take that?
- [48:16]
Kyle Turman
Oh, 100%. And I think that this is quite different in the sense that, I mean, you could argue that the Internet for a while was a solution in search of a problem, right? Like, we forgot that a lot of these things when they first came out were solutions in search of problems. The first apps that came out for the iOS store, right, were like a app that showed you drinking a beer and a fart app, right? What problem is that solving? It's novel. It's novelty. Right. And so I think we're in that stage where we're kind of transitioning from novelty into actual usefulness. That is where it starts to get a little tricky, where there's still kind of solutions that are in search of problems. And the key is actually understanding how that solution best fits into a specific problem. And so I think that changes the fundamental nature of how we even come about. A lot of times we'll even say, like, hey, this model has this new capability that we found. What does that do for people? And so sometimes we'll explore that. And this was actually how Artifacts got born last year was we were already working on this kind of, like, side panel to be able to, you know, iterate with Claude. But the thought was really mostly based on words on documents, where it's like, oh, I'm writing something with Claude and I want to see like how Claude changes it and it stays while the chat scrolls. And we found out when the sonnet and then 3.5 sonnet came out, it was really good at writing REACT and HTML and there were all these kind of internal demos of it writing code and basically one shotting entire little web apps and doodads and things from that. We were like, okay, well what if we made the artifact pane something that could actually render rich HTML? And so the team sprinted on that and we weren't even planning to launch that. We were just going to have it be code, right? But because it was so good, we saw that capability and we amplified it, right? And because of that you could actually see it be rendered. And it felt way more impressive and it felt more tangible and it felt more useful because you didn't have to copy it in your code editor. And the other benefit is like people who didn't know how to code could all of a sudden make all these weird little things and have a ton of value from it. And so I think that that's like one aspect where I think that really comes to a head, right, is like you find these things, but then you try to apply it into something that would be most useful. And I think that's the challenge.
- [50:46]
Rid
I love that example because it's working backwards from a capability. But it literally changed the industry, like in so many ways. Like, that's been the pattern. It created a whole new category of companies that are some of the fastest growing startups in the world.
- [50:59]
Kyle Turman
I mean, a lot of it is due to Michael on our team, basically like yoloing half the code himself. And it started with a demo that he. I mean, especially last year while we were still forming, most of the things we were doing was very bottoms up. And it did start with a demo that Michael posted. We each have Notebook channels in Slack. I really love this pattern. I think everyone should do this where you have a channel in Slack that is minus term in Notebook. And I post my work in progress there, I post ideas there, there's discussion, people have things. So Michael had posted this demo and based on a conversation with myself and another researcher, literally within three hours of posting that, Dario, our CEO, was like, we have to build this, we have to make this real. But that was before even the code thing. We're already wanting to build that. And then we figure out this new capability and we layer it in and it just makes it even better, right? And so there's all of that only happens because you create an environment of trust. That's really where that good creativity comes from that you just don't know about. Right. Like, I don't think any good feature ever comes from the CEO saying, we should do this feature, or them coming up with the idea. It's. I think the good mark of a good leader is knowing how to fan.
- [52:12]
Rid
The flames, especially when it's a primitive that can be used in so many different ways. Like, when you're building this, like, really horizontal product. It's like, the ultimate win.
- [52:20]
Kyle Turman
Absolutely. Yeah.
- [52:22]
Rid
All right, time is flying by, so before I let you go, I want to zoom all the way out and kind of talk big picture for a little bit. So how has your experience at Anthropic shaped the way that you think about the future of design as a discipline?
- [52:38]
Kyle Turman
I would say, like, the first year I was at Anthropic, I didn't really think about that at all because I was just. I think it's the hardest I've ever worked in my life. A lot of time was spent just, like, making things happen and making things work. And I'm so thankful now that we have a bigger team and we can all kind of step back and think more. And so, really, in the past few months, I've been thinking about it even more in the sense that I think that I mentioned a few things where, like, you know, I think designers are especially well equipped to think about large language models and how they interact with people, because that's what design is, right? It's. It's creating ways for humans to interact with machines. So we've been doing this forever, and we are really uniquely positioned to do that. I think a lot of ways that people think about, oh, yeah, this changes the design industry because all of a sudden, designers can code now, and I think that's great, too. Being able to code is definitely something that has helped my career a ton and the ability to both make things that I believe in without having to sell someone else and also get into the weeds and the fine details that are really hard to convince engineers to do. Those things are really helpful. I think it makes it easier to be more impactful, I think, as a designer. But I think the things that are going to change the most is when we look at kind of maybe let's say, five, six years ago, right before the pandemic happened, people were hiring designers like crazy, right? There were design boot camps. People were just, you know, getting hired left and right. And then when I look at portfolios, there are a lot of designers who worked at these really large companies. And because these companies had so much money at the time, the only thing that they worked on was like a little button in a settings menu, you know, four clicks deep that no one ever saw except for like 20 people. And they worked on it for like a year or two years, right? And it really makes me sad in the sense that, like, that's the only experience I think a lot of designers out of school had in that kind of category for a while, and they didn't really experience this quality of building something. And the thing that I really hope for is that we remind ourselves that design is not about buttons, it's not about pixels. Design is more than that. Right? People have been designing things for thousands of years before the Internet was even a concept. Right. As we consider what our role is as designers, there's an aspect of it where we have to think about taste as a part of this. And I've seen a lot of people try to define what taste is. And a lot of people think that they're fancy because they, they're like, well, I, I know what taste is and you don't. Ha. You know, that's not what taste is like. Taste is the ability essentially to just look at what the world is hoping for and be able to meet them there, right? Be able to say that something that you're making actually means something to someone and has intention and has quality to it. That's taste, right? And to me, if you look at like, you know, beautiful buildings or beautiful pieces of art that people will say are tastefully done, that's what they're doing, right? They're, they're inspiring people. They're. They're creating beauty not just to make money, they're creating beauty because beauty needs to exist in this world or else what are we doing here, Here, right? And so I think that this is a really big opportunity for design to bring that back and to recognize that our purpose is more than shuffling around little boxes on a screen. Our purpose is making things that are meaningful to people. And as AI becomes more and more part of our lives, it's going to be even more important. Right? I do think that there's going to be probably in the next 10 years, a renaissance of handmade items. It's fascinating to me. I worked at Etsy way back in the day whenever, like, handmade was kind of taking off and corporate was like, who wants to buy corporate? Meh. And now, like, there are apps that people are lauding that literally take a photo of a high quality item and help find like a TEMU version of that item for like hella cheap. And people are like, oh, this is great, like what happened? Why, why, why do we want all these knockoff things that just like make us feel like we're a part of a group without understanding the meaning and the purpose of that item that we're even buying, right? Where it comes from, how it's made. And so as things become cheaper to make in certain ways on the Internet, I think there's going to be more and more of people craving something that feels authentic, that feels worth using, that makes your life actually better. And I look at software today and most of it I don't like using. I use it because I have to, to whenever someone puts a little extra effort into a piece of software. And I don't mean this like I think people use the word delight a lot. And like there's some apps where like you complete a task and a little like flying unicorn grows across the screen and you're like, oh God, you know, okay, that, that feels very like mandatory fun type vibe. But when something works the way it should, whenever something feels considered and you're expecting something to be bad, but it actually is good, those magic moments are where like your ability as a designer will be very, very hard to ever replace as any sort of AI tool.
- [57:50]
Rid
One of the outcomes of software becoming cheaper that I'm hopeful for is that it becomes economically viable to make a heck of a lot more niche software where you can just deeply understand a set of problems that maybe never existed at the same scale required to build a venture backed business. But now all of a sudden it's worthwhile. And when you have 100 times more of that scale of problems, more designers then get the opportunity to do what you did, which was really steward something from the seed of an idea end to end to be this thing that creates value for a group of people. And that's the complete other side of the spectrum of operating in a big company and having your silo and your two pages of the app. And that gets me really excited.
- [58:34]
Kyle Turman
I look at some of these startups that have existed over the past 10 years and they hit a certain threshold where they probably should just stop. There are certain apps that don't need to be everything. They just do what they do really well and that's it. And that's fine, that needs to exist. But what happened is those companies took venture capitalist dollars and there's an expectation of return that is absolutely enormous. Right? And because of that expectation and that Pressure. They're forced to do things that make more money, even if they don't make sense or people don't need them. Right. And so I think one of the things that I'm hopeful for in that way, too, is that the software that gets created is not pressured to be more than it needs to be. There's a lot of apps that just do the job and that's it, right? Like be. Be the messenger app or be the file sharing app. Like, you don't need to do more than that. Like, you're done. Like, cool.
- [59:30]
Rid
We're a year or two away from someone being able to displace linear by just building a simple linear. You know, like, it's definitely possible we're caught in this inevitable bundling and unbundling cycle where it's like, yeah, you're right. Maybe it's going to be easier to just say, you know, we did it.
- [59:44]
Kyle Turman
That is one thing I actually do love about linear, though, is that I think it'll be really hard to replace because it feels so well crafted. Right? It feels like a good pair of shoes that you put on, and there's still a lot of value in that. But I also have really respected them for not growing more than they needed to grow. Right? Like, they're doing their job, they're doing what they believe in, and a lot of companies don't do that well. And so I'm hoping companies can stay smaller, they can take less VC funding, and they can kind of shape their own destiny. Because at the end of the day, isn't that what everyone wants? I've had to do a lot of design work that I did not want to do because I needed a paycheck. And that's real. And we have to acknowledge that, too. I talk a lot about fancy, cool design ideas, but. But, I mean, I grew up in the middle of nowhere on food stamps, and this has given me such a beautiful, wonderful life that I'm so thankful for. But I've also had to do a lot of stuff that I did not want to do. And I would love for more people to do things that they believe in and feel good about and feel proud of.
- [60:44]
Rid
Well, you're playing a big part in crafting that world, and I'm appreciative of all of your efforts, Kyle, and also just taking the time to come on here and talk about it and tell the story and share a little bit more about how you think. One of the things that's really stuck with me is just how you identify where emotions exist in almost all of the decisions that you're going through just as a designer. And I think that's really cool. I think it's really cool and it shows up in your work. And so thank you for taking the time today.
- [61:11]
Kyle Turman
Thank you. It was a ton of fun.
- [61:12]
Rid
Before I let you go, I want to take just one minute to run you through my favorite products because I'm constantly asked what's in my stack. Framer is how I build websites. Genway is how I do research. Granola is how I take notes during crit. Jitter is how I animate my designs. Lovable is how I build my ideas in code. Mobin is how I find design inspiration. Paper is how I design like a creative. And Raycast is my shortcut every step of the way. Now, I've hand selected these companies so that I can do these episodes full time. So by far the number one way to support the show is to check them out. You can find the full list at Dive Club Partners.