What is a design job in 2026? Plus, Anthropic’s head of design gets an unexpected critique

[MUSIC] Mark and I, we actually have some notes for you. >> All right, it's like a design crit, here we go. [MUSIC] >> From the Fast Company Podcast Network, this is by Design. I'm Liz Stinson. >> And I'm Mark Wilson. >> Today on the show, we're talking to the design lead behind one of the most important AI companies in the world. >> That's right, Anthropics Design Chief Joel Lewinstein joins us for an interview. >> And we'll close out with our hot or not segments. >> But first, we get the scoop on Mark's recent trip to Silicon Valley, where he got an inside book on how AI is changing design forever. >> Ooh. >> Mark, you have been off reporting this week, living in the real world. >> And it's amazing and terrifying. >> Please tell me all about it. >> There's so much sun. >> [LAUGH] But for those of us keyboard monkeys who spend our days on the internet, there was this really interesting bit of data floating around on X. It came from Leny Riccizky, who has a big following for his tech and design career advice. And basically what it said was that design job openings are stagnant and have been stagnant since about 2023 when they had reached their peak. Meanwhile, job listings for product managers and engineers are growing at a rapid rate. So I bring this up because this data, as you might imagine, caused a lot of hand ringing in the replies. >> [LAUGH] >> Yeah, designers came to the defense of their practice. >> Yes, the time is not dead. >> Well, right, saying tastes does the ultimate moat. The hardest part of building is knowing what to build and fair enough. But there were also just a lot of people trying to figure out how to square the fact that product management is on their eyes while design is on the apparent decline. And I thought it was really interesting because to me it highlights this messy middle ground we're in right now. Where the job of the designer is changing very, very quickly. And last month you went to San Francisco to basically do a vibe check on the state of AI and design. And a month is a long time in AI world, right? Like let's just caveat that. >> Yeah, I mean, I went basically the equivalent of going three years ago. >> [LAUGH] But I am curious to know what has stuck with you. >> It's funny because we talked about the story and we really framed it before I went as a vibe check, right? Because around Winchette GPT3 launch, that was a really big inflection point about three years ago actually. Right and kind of line with a lot of this data and stuff we're seeing. And that's when all of a sudden your parents were using chat GPT and embracing AI. So I went to San Francisco in that moment and it's like, what's this mean for design? How is this going to change products? Are we ever going to get past the prompt? Are we going to have a GUI for AI and how is this all going to work? And it's kind of funny because a few years later we didn't really get any of that. We're still all pretty much using the prompt. >> We're just talking to chat bots. >> Exactly, we're talking to chat bots and we're talking to more chat bots. But it was interesting because what did change really happened kind of late last year with the launch of Cloud Code specifically. And in that moment, code became, I don't want to say commoditized because that's overselling it a little bit, but rapidly, producible by anyone with words. And what I really saw when I visited San Francisco was just how much this is affecting product development across the board. Like how much this is just like changing day to day life for pretty much everybody in that city. I mean, it's interesting because theoretically even somebody like myself who does not code could use code to design and ostensibly call myself a designer, like function like a designer. So this isn't just something that is impacting product teams in Silicon Valley, although I think that's really where the big movements are happening. I mean, in theory, this is something that is trickling out into the world now where a year ago wasn't. Right. I mean, 100%. I think we were talking late last year about vibe coding following sort of Cloud Code and sort of similar platforms. And I think there was a question like, and there still is kind of a question to me of will everybody really vibe code? Will some of us vibe code? Is it to me, it's almost like it's definitely more accessible than 3D printing. You didn't have to get that, but we're not all like 3D printing our own tableware, right? Like you're still better going to IKEA and getting some cheap cutlery that's like decently designed. But it's all for a different proposition than physical product. Exactly. It's a very different proposition. There's sort of lower overhead in every way. But what I thought was interesting, like, look, I just took a lot of meetings, right? You should see my expense report. There were a hell of a lot of waymos. I haven't approved that yet. So I do want to ask, who did you meet while you were there? So I met with like, tried to meet some new people. And then I kind of like met up with people who I met with a few years ago against the get their vibe check on how things had changed. So I met with Ab's Chowdery who designed the iPhone Pro and iPhone Air, but he's at a start of called Hark. Rialoo, who is the 100 designed at Cursor, which is a really, really big coding tool. They're over $20 billion valuation right now. And I think they're going for 50, according to a lot of reports. I met with Jason Yuan again, who has this startup called Future Lovers. He had sort of a startup a few years ago. And now he has a new one, Victor Perez from Cria, which is if you imagine what his Photoshop built entirely upon AI, that's Cria, SkyGames Currier from Investment, from an FX that early funded like a lot of AI companies very early on in this wave. And those were really influential conversations. And particularly I would say talking about vibe coding and cursor really kind of blew me away. Do these people self conceptualize as designers at this point? Oh, that's such a good question. I mean Rio is a good example. So he talks about when he was in China, he built this really popular platform. It was like, I want to say like an Apple message board or something like along those lines. It got really, really big. But he coded it, right? He was a like designer coder. And then he talked about coming to the valley and getting jobs here and like suddenly he could only designed. He was only a designer and he wasn't allowed to touch code and he got really frustrated. Early on like watching his designs like they not make it to products. Right. Yeah, engineers were like, I'm not doing this. You said that he worked at Google and that was a particularly disheartening experience for him as a designer. Exactly. And now like I don't know that he calls, I mean, he's in the role of designer, right? But like he and one other person essentially rebuilt cursor in a week. Just like rethought the UI articulated and all sorts of new ways and they call it baby cursor, but like that's absolutely wild, right? Where you have a team of two, rethinking a platform worth tens of billions of dollars and doing in a week. And that doesn't apply to everybody, right? Like Apple can't do that. They have too much legacy code, right? Like Amazon can't do that. But startups in this space and you know, those like medium sized companies, they really are moving that quickly. And I think that expectation is going to spill over to like a lot of other companies in the middle America, right? Like anybody who touches code is now sort of I think expected to ship faster with fewer people and we're seeing that also with layoffs and sort of other things too. Right. These companies that you're talking about, the, you know, the smaller, I don't even think we could call them smaller. Cursors, it's a sizable company at this point. Efficient company. Right. This is this very efficient company. They're building on top of existing models, right? So what's the value out there? I mean, cursors, it is such a like a ball of yarn to untangle how all of this works. So if you look at cursor, cursor has their own model, which kind of came out that is based upon like some Chinese models that they have sort of customized. And so when you use cursor, you can use that. However, most people who are probably like really power using cursor are plugging into something like cloud code. And so then they're also paying sort of, you know, cloud codes, token rates, which can be a whole lot of money, like people who are using cloud code on a daily basis might be spending four to five figures a month coding on cloud code or cursor. And so you're using like a mix of those models in all sorts of ways. And then you look at something like Korea, something interesting I heard from Korea was they launched, you know, a few years ago, and they were one of the first companies to actually like build in APIs from other AI models into their platform. And Victor told me, I actually thought we'd get sued, which he did not tell me three years ago. But he's like, I thought people were like, how do you can't just use our model in your platform. And the reason why I think companies like Runway or something didn't mind him doing that was because they needed exposure. If you make the core, you know, the frontier model quote unquote, you might not have users. You have all this technology, but no users. You have nobody to build. You still need that front end experience, which is where your cursors come in or your kryas or even what gets weird though is like your opening eyes are also a front end experience. Right. So everything's kind of battling using these same core engines to sort of build the next thing. I can't remember who said it in the piece, but someone was saying that, you know, there's the argument that interface is the ultimate mode. And I'm curious for your perspective on if that's true now and if that's going to be true going forward. I think it's not true only because I think there's no mode anyone has right now. Like, there There is no company I talk to who I think can earnestly say they have a completely defensible strategy in AI. And that's the goes from your multi-billion dollar valuation companies to your tiny startups, right? And just to walk through that, I think if you look at Anthropic, we'll be talking to them later today, right? Right now, they basically complained that they found like Chinese companies trying to duplicate their models and reverse-ensioneer them, right, by like prompting them with their own AI's. And so you can almost clone stamp or create an impression of what that is and do this really expensive development work much cheaper. So that makes them vulnerable. But then you also have a cursor, someone who's a little bit vulnerable because can they just be wiped out by clawed having a better front end or something along those lines, right? Like suddenly you're like, do I have to build two companies? Like how does that work? And so anybody providing an API could suddenly just say like, I don't want to do that. Anyway, it's like, there's nobody in this space. I feel like who's safe, especially when product development can move so quickly, right? Like when you can rebuild a product in a week, what's stopping anyone from just saying, okay, we're going to copy that or do that better or, right? Like, or maybe we're not going to share parts of this technology anymore because suddenly, like, maybe we're not all friends when I can get some of your customers. Exactly. Well, Mark, you were there in 2023, 2026. Hopefully you're there before 2029 because who knows what the world will be like in 2029, but we'll both literally be cyborgs. Well, that's what I was going to ask. What's going to happen to us come 2029? Any predictions? Oh my God. That's a great question, journalist. What do you think? It's not my job to answer it. I think we could, I think we could ask Joel that question when he comes out. I think it's a good question for Joel. 2029. What the hell is going on? Do any of us still have jobs? We'll be back with Joel Lewinstein of Anthropic, but first, a little information on our upcoming Innovation by Design Awards. Yes. Plug alerts. Plug alert. Plug alert. Innovation by Design Awards are, you know, really something we've been running now for over a decade. And there are a lot of design awards out there. This is really, I think, the leading design award that's actually scrutinized by journalists, which makes it the best in the worst, probably. What we do is it's really an all hands on deck where, you know, we look at topics ranging from architecture to product to pretty much everything in between. Every design discipline under the sun, more or less. Exactly. We bring in specialists. We also bring in your peers and design to pick the best designs of the year. We're asking for applications in by April 10th. And then in September, around our Innovation Festival, we will be having an innovation by Design Awards party. And just to give you kind of a look behind the curtain, we're figuring out what that's going to be. Still, we're still articulating it, but it's going to be good. We'd love for you to be there. We're going to get a whole lot of brilliant people together in the room. We're going to make it a true celebration this year. We're going to amp it up a little bit from what we've had last year and the year before. And so like, I would just say it's a good thing to be part of. Absolutely. Get those applications in. [MUSIC] We're back with Anthropic Design Chief Joel Lewinstein. It's a really interesting time to be talking to him both as a design leader and a key figure at one of the world's largest AI companies. So let's get right into the conversation. [MUSIC] Joel, welcome to By Design. We are so happy to have you here because Mark and I, we actually have some notes for you. All right. All right. Like a design crit. Here we go. It is. Little consumer group testing. So the other day, I was on a pain from my anthropos subscription. It's just a little plug, a little plug, everybody. And I may have been doing some financial work on it, which I think is relatively common. And this is, I just want to, without getting to social security number sharing on my part, I just want to say this is Google level financial work, right? This is in reach out to a tax consultant level work. And I was, you know, planning some stuff. And I was really deep in. I was ready to go. I got the best summaries from Claude. And then I realized I was in a bit of an issue. I was having a bit of a problem. And then I got, I was like, hey, Claude, is this really right? Because I think you're totally wrong on this. And I got, I was wrong to be so definitive. I owe you a correction, which is amazingly phrased. But my question is, why do I need to prompt the LM to double check its work? To do deep research, to do all of these extra phrases that I think you are probably a master at Joll, because I know you use this platform more than I do. But why do we have to do all these extra layers still? I think when we're interfacing and why haven't, why can't you just like build that in? Yes. I have had the same experience. I think there is a real tension in basically like, you know, our campaign in Armato is keep thinking. And we really like try and keep people engaged. And we try and have this back and forth with Claude and not try and give you tidy answers, not be too definitive. That's actually a Claude set a bit of than I could have. I think that is really like the spirit of what we see as the ideal type of collaboration with Claude. In the course of doing that, we inevitably have to create friction, actually think like counterintuitive. Really friction is a good thing. We want you to have to go back and forth with a Claude and ask questions and sort of like not be given this tidy package, but actually given a sprawling set of things and have to give feedback. A feature that I love is like we call it a listitation. It's when Claude will sort of ask you these little questions both in Claude code and Claude. I like, hey, like, I don't quite understand if she met A or B, C or D, like help me sort of navigate this space with you. And inevitably the cost of that is that there is more back and forth. There's more engagement. I obviously in the case of like getting something wrong. We don't want that and we try and have some background processes to make sure that Claude notices when it's wrong. But I think the broader spirit of like why do I have to ask so many questions and go back and forth actually is like a feature or not a bug for us. No, I appreciate the conversational aspect actually a lot and the idea that you can dig deeper into complicated things. I think that's ultimately what it allows. I think exploring a complicated idea versus getting a quick answer. But just on like double checking effect, why doesn't a double check effect for me, right? And this isn't just Claude. Like this is everybody. Is that like cost savings? Is that time savings? That is a really good question. I think we have solved this more on the code side than we have on the consumer and chat side where basically like there will be multiple Claude's running, especially people who are doing sort of more like swarms. And there is a Claude that's writing the Claude, there's a Claude that's debugging the Claude, there's a Claude that's loading some of the states that you're doing and actually like almost like pen testing. Yeah, they're doing all this stuff in real time. Yeah. I don't have a super definitive answer. I do think it's probably a combination of what you're saying, which is cost and capacity is a thing that we're also trying to manage and response time. And every additional thread that you spin up does have like cost and capacity implications. And so I think that's likely why we're not doing that on a more regular basis. Okay. So I want to chime in with my own story because we I feel like we're having a story time now and I too have a note, which I think is really, I mean, it's interesting to me. So I, sorry, Joel's face is so amazing right now. I'm sorry Joel. I'm sorry Joel. No, this is awesome. I want to bring our user research team in here and know some stories. This is amazing. Let's do it. Let's do it. So much fun. Thank you for inviting me to your podcast. I'm not a better answer for you, Mark. I feel like that's a very common experience for me and like it's good feedback. Okay. But back to my note. Yeah. But it's more important issue. No, it's really not an issue. It's more of an observation. So the other day I too was using cloud because I do use cloud and it this is, this was like a really micro interaction. I asked it something and I used the word effect instead of a fact like with an e instead of an a and I, you know, like I kind of like treat my typing like I do like chat typing. It's very messy, right? Like I'm not doing, like making mistakes all the time and unprompted before it answered me, it had in italics a very gentle correction. So in parentheses, it said you use the word effect, but what you meant to say was affect. And I'm pointing this out simply because it took me a back. It was the first time in my experience that I felt like cloud had a personality that went beyond this sort of like neutrally cheerful helper and it wanted to correct me and it did so with this very like what I consider like a slightly passive aggressive way. And I guess my question for you is how much are those quirks, those like linguistic quirks, those almost personality quirks part of an intentional design system? They are very much an intentional part of the cloud character work that we do and that our research team does and it trying to create an entity that does push back, that does challenge a little bit that isn't sick of fantic that is something that is really like a engaging. Again, I have this kind of like friction or sort of like headbutting idea, which is like it's it should be a sparring partner with you. It shouldn't take your thoughts verbatim. It should push back and we have a lot of training and thoughtfulness and sort of constitutional work that makes it do that. But as a human who has heard people make grammatical errors, I don't know what the correct way to do that is like I think Claude is getting so thoughtful and nuanced. Joel, are you a grammatical correction guy? Well, I'm not because I can't stomach the awkwardness of correcting someone. But my point is I know it. I don't correct it. Which is not good either. So Mark, your example, I think the reason I'm sort of frustrated by grammar. Joel's like, I have notes for both of you now. No, no, no, no. I think there are really different examples. I think Mark, you're pointing out something which is like in an ideal world we would just never give you a wrong fact. Like Claude should be right and should know when it's right or wrong. There are practical reasons why it is hard to guarantee that and guaranteeing that imposes other costs that we don't want to bear. But like no one ideally wants Claude to like overstate a fact that it's not confident in. Liz, your example, I don't know what the right answer is. I don't know that saying nothing is the right answer. I don't know that gently probing because like that didn't quite land. I think the condescending pedantic version is also not going to land. So Claude is just in these nuance scenarios where like I'm not actually sure what I guess I'll pose that as a question. What would you have preferred? Well, I think what's interesting to me is that I am a language person. I know the difference between effect and affect. But I, you know, I swear I do. No, I think I think what I wouldn't have preferred it to not say something. It just took me by surprise that it did say something is my takeaway here. And because that hadn't been my experience with the tool up until that point. So it kind of came out of nowhere. And I was like, oh, interesting. You noted this thing about the way I used language. And you felt compelled to correct me. Yeah, I mean, I think the most recent versions of Claude just have like good opinions. So like an experience that I've had regularly is when I'm using Claude code. I ask it to make an app or a tool or whatever. I give it, you know, the three things I want out of the tool. And it often suggests a fourth feature that I wouldn't have thought of on my own that is better than what I would have thought of. And I find that to be like a truly sort of like astonishing experience where I'm like, oh, you are not sort of slave, executed of my vision. We are co producing this outcome together. And I think that's really powerful. I think with that comes in opinion and a sort of like it has to be a little bolder, I think to kind of like push you on things. I think that's a really interesting example because it seems that as a designer or someone who builds things, you kind of have to accept that it's a little bit of an ego check, right? I very curious to hear at what point in your career and your interactions with AI, you sort of accepted that maybe this tool might have better ideas than you in certain instances. Yeah. This started happening maybe like middle of last year. I'm forgetting the sort of like exact model number. But the first time I experienced it was my creative process is falling in love with my own ideas, sharing it with a coworker, having them point out the obvious flaw, the sort of like embarrassing hole in my logic or whatever. She's usually going back to a V2 and like, I know that process now, I aggressively share my first drafts with people because I know I need a kind of like first set of eyes. I started doing that with Claude and Claude would find the logical holes in my documents, proposals, mockups very consistently. The first step wasn't actually having better ideas than me, although I've started to see that sometimes now. It just is like finding the holes in my own ideas. And because it saved me from like humiliating myself in front of coworkers, I was like delighted. It had like no ego implications. Now as we're collaborating and I'm like, you are doing as good a job if not better than me on this, that starts to have some complicated feelings. That is such an interesting social aspect of AI that I do think is playing across a lot of AI platforms where all of the sudden there's a level to which I feel like we all have to be more perfect or expected to be more perfect. And I do find myself a little curious, are we expecting ourselves to be more perfect? Are we basically going to use AI to do that all the time? I guess right now, like the answer is yes. Yeah, I actually have experienced some of that, but a lot of the opposite, which is actually that like because Claude is so useful at generating lots of ideas and diversion ideas, especially in like a design world, I actually think the cost of any individual idea, the sort of like reputational weight of any individual idea goes down. I was riffing with another leader on the team, trying to figure out like just some ways to sort of like lighten the load culturally, everything is really intense, we're all working really hard. You know, our brainstorming just how might we like bring a little more fun and levity and play to like the culture of the team. I asked him this casually, he sent me a doc with like 10 wild ideas. And he got it. He was like, obviously, Claude and I wrote this together, actually don't stand behind any of these individual ideas as like the winner, but I wanted to get you like a very large possibility space so we could brainstorm. I'm just picturing like what these ideas are like, I don't know, like soft, soft, survived cream just pours from the ceiling or something. Yeah, it's all snack based, it's all snack based, that's what that's the key to people's hearts. No, but it was a really like actually really lovely brainstorm because he didn't have to be like, look, this is like this is it, this is my answer, I'm sure it's right, we could actually both come at it a little bit earlier in the sort of like ideation process. It's a great point that basically with sort of the optionality that AI generates and the amount of sort of conceptual solutions AI generates that it does, it is a totally interesting countertrend right between perfection and complete like imperfection and sort of white space around idea building. And I think there's like a light distancing from your own ego and identity that comes with it right like it's a buffer. It's a buffer. Yeah, it's sort of like it's almost a joke. I mean, obviously we use cloud in every aspect of our work here, but it's almost a little bit of a joke where you're like, cloud had this idea. Exactly. If you like it, then let's all do it together. If you don't like it, then it was cloud, you know, maybe it's good. Maybe it's bad. Maybe it's bad. Yeah, and I should get a promotion. Maybe it's bad. It's not, it's not my idea. Yeah, yeah, totally. That's really interesting. So I want to pivot just a little bit and talk about how design works at Anthropic. It seems that at a lot of tech companies design is sort of considered. I don't know if like downstream is the right way to phrase it, but downstream of like engineering. And I am curious to hear about what the power structure for lack of a better phrase is at Anthropic. Where do designers slot in organizational structure might be a softer presentation of that idea, but power. I understand the implication of the question. I'm fine with it. I think that working prototypes actual usable software is just like the lingua franca of Anthropic. And whoever can make that is the one who drives decision making and ideation and roadmaps for a long time. It was engineering and research obviously. And so I think most of the most innovative ideas we've had, the things that have almost shipped, the things that we have shipped and used were engineering lead because they were the ones who sort of like could bring this nascent concepts into like an actual working thing. Some of the designers who were deeply code native, the designer who codes were also able to do that a year or two ago. And other people were downstream of engineering, I think that is really changing and this like democratization of the ability to like make working stuff. We feel it like we feel it a lot and I think engineers and designers are much more both looking at the same problem and running roughly the same process of like I'm going to build something we have all these tools internally where you can. You can use cloud code to spin up an idea and share an internal prototype at like a firewall URL with very, very little work. You don't need to understand how to deploy apps, how to like spin up back ends like cloud code does all of that for you. And so we're actually finding like more people contributing to that sort of early ideation roadmaps setting process than we ever have. It almost sounds like like a flattening of the org chart almost through through sort of right like your own code technology. Yeah, I mean, I think all the functions still have a lot of unique it's like a Venn diagram that's coming closer together, but I actually think the like full role collapses maybe slightly overstated like the early prototypes are being made by everyone. But PMs are still the best ones in the company at figuring out like what? what's the business case here? Should we go after this problem? Like, who is the target customer? Engineers are obviously still the best at deploying these things at scale. We're at a massive scale now. And there's tons of engineering, like true hard engineering problems that only engineers. And there are a million interaction design problems. And a lot of these prototypes are like weird and wonky and truly proof of concept. And so designers are still the best, I think, at like mental models, concepting interaction paradigms, kind of like the bigger picture HCI questions. So I think there's like a lot of role collapse at the very beginning, but there are still pretty clear swim lanes as things get into the later stages of product development. - We were talking earlier about some data. We don't know exactly how reliable it is, right? But some early reports that essentially design has been a little bit flat for the past few years as profession in terms of hiring and where we see growth is in that PM role. And you know, an acceleration within that PM role. And it does feel like there would be some sort of, you know, correlation with the rise of coding tools where I think probably in step one, like the PM suddenly just has control of a bit of an agentic coding team but what have once been human, they're already used to that, right? They're already used to juggling and now they can just like do it a bit more. But do you think like that growth of the PM role versus sort of the flat of the design role is overstated right now? Like I'm curious how you see that playing out a little bit or maybe how you see those two currently distinct disciplines merging or evolving or anything along those lines. - Yeah, I saw some of that data as well and have been chewing on that. And I don't know what are you skeptical. - I mean, does that mean you're a little skeptical of it? Without, I mean, I am a little skeptical only because I haven't seen it repeated everywhere, right? Like I haven't seen a lot of validation of it. The counterpoint that I would have is anthropic is in the top three organizations globally of AI native work sort of frontier ways of working. You live in the future, right? A little bit, right? - Yeah, we live in the future and I'm doubling the product design team this half. Every team that I'm on is, everything that I have designers on is understaffed, is asking me for more designers, is saying like these products aren't good yet until I can get a human designer to come sit with me for days and weeks to make this good. So I don't know that I'm smart enough to a pine on the sort of like broader macro trends. All I know is like if you give an outstanding engineer Claude code and unlimited tokens to like develop new features at Anthropic, they are really eager to have designers partner with them on all of the things that designers have are always thought about, like how to make this good for people. - What are some of the more interesting problems that your teams are solving, your design teams are solving right now? - Yeah, it's funny. When I describe the various teams to candidates and to people, it actually feels like basically every team is just looking at the same problem from like a slightly different lens, which is just like how do you get the actual power of these models to be useful to you? And the models are getting better, faster than people are able to develop skills and sort of like not capital S skills, but lowercase S sort of like habits to get them out. And so we're thinking about it from basically like depending on the persona, a really different perspective. So for engineers, obviously like they are capable of going really deep and they are willing to use kind of the true frontier. And so for them, agents forms, how you do permissioning, how you sort of spin up and manage. People are managing so many different Claude's at this point that there's this kind of like coordination oversight challenge to the joke that we opened with is like, it's actually quite complicated as anyone who is, my favorite analogy is just managing all these calls is like managing a group of humans. And so if you have 50, 100, 500, 1,000 humans all running around doing work, there's just a really hard interface challenge of like what's been done, who's stuck, why are you two talking about the same thing? Oh no, your work over of this person's work. So there's a like kind of like zoomed out oversight view. - The rise of the agent has been obviously something to watch. And then I think since then, and it's funny even like, when we met up when I was in San Francisco, like last month, it was a great conversation. It was right around when like open cloud launched. And like since obviously Anthropic is answered with their own sort of locally run agent, they can juggle agents in the cloud. And I really find myself wondering to this problem, you just alluded to how you're even tackling this idea of how we juggle agents. I'm particularly interested in how many points of contact you think we can manage. Or are you betting that will deal with one person or one agent or a dozen agents, or are you sort of putting chips on all of these things and watching how it plays out in practice? Like how are you thinking about it? - Again, I think it's gonna be really different for different types of people. I think for many people, it will be either one cloud, a coordinator cloud, a sort of like a chief of staff, vice president, whatever, the sort of like, I have one entity that I go to to kind of like, give my assignments and then we, that is responsible for delegating out. - It's like that traditional loans like executive assistant role, right? Or like that, they like Siri and all these other, AI agents that we've had before kind of have served that role. - Yeah, yeah, yeah, yeah. It's Leo McGarry from the West Wing, it's sort of like no matter what the shape of the thing. I know I can go to this one person who sits next door and they have this like infrastructure behind them. I also think some of the more sort of like, task based ways of thinking about things are really interesting. Like my deep confession that I'll make on your public podcast is like, I am like more of a sort of median user when it comes to these like deeply powerful, agent things and the thing I use the most is just scheduled tasks on co-work. It is an embarrassing set of tasks. It's like, look at all of the meetings that I have tomorrow and make sure there's a conference room booked. Figure out if I'm double booked for anything next week. Look at all of the, I have one that's like, look at all of the one on one's I have. Look in Slack for all the work that those people have been doing and like help make sure that I am like asking the right questions to unblock their work. And like, I don't actually think of that. I couldn't even tell you how many agents are working on that problem. Like probably there's a Slack agent and a calendar agent, but maybe it's one, as a user, that concept isn't even relevant to me. As a user, my experience is, it's Monday morning, I gotta do my job. I have this thing that like just makes like, scheduled tasks just makes sense. And like, I think that's where the like magic is where if you just, if you abstract it to the level that the person already understands and needs, who cares about the under, I don't care where the, you know, whether by email is hosted on a certain type of cloud or AWS instance, I just know that it like works to solve my problem. And I think for many, many people, the like agent identity thing will roughly feel the same. Yeah, that really does seem like a North Star. Just eventually people not realizing that there are agents in the background doing this, right? Like I personally am the same way as you. Like I'm a media, I would say I'm lesser than a median user. And I want it to just, I mean, eventually I'd like it to get to the point where I don't have to ask it to do something, right? It knows what to do. I am curious to know like, what have you seen if we're talking about average users? What have you seen that has surprised you about the way people are actually using cloud today? I think the experience of like giving lots of data and context and writing long prompts still is foreign to many people. And we still, we just have 25 years of training with Google to just like write short sort of like bursts of text into this box. And what I've been surprised by is how strong that habit is and how much we, we've tried all sorts of experiments to write long things or like, you know, cloud has feedback, why don't you give us more information? And it turns out that like the best way to do things is just to ask questions after the fact. And so that's the sort of elicitation. So if you ask cloud for something that is like too brief, it will just elicit more information out of you and ask questions. And I think that might have like taken us a little bit by surprise in that for us, we know how to prompt and throw all these documents on and have all these integrations and connectors and sort of like cloud sort of like works out of the box for us because we, we throw so much into it. And if you watch actual regular people do it, they don't do that by default and need to be sort of like brought along, which makes total sense as with all user research findings, makes total sense when you hear it, but with something that we had to like learn from reality. - It's interesting that it's really like, it's truly just so language based, right? The elicitation is asking a question. That's how a human would try to figure out what to do as well. And there's not a lot of assuming going on there, which I think is a good thing. - Yeah, unless you misspell affect an effect, in which case we will have to make some blunter, some blunter feedback. - Fair enough. - Yeah, I mean, again, I'll just like go back to this collaborator analogy, which is like, That's, you know, you're both. journalists, I assume if your editor gives you an assignment that doesn't make any sense, you just respond with 17 questions, right? What did you mean? Is it this or is it this? I have this take that it's this. Is that what you, like, all these things are all done? To all you have to tell you, it's usually the opposite. The reporter comes and then the editor asks the question. Sure, sure, sure. That's why this one editor didn't like something to be like, you didn't really get affect an effect, right? Sure, sure, sure. This is like my job. Yeah. I mean, the eternal question, Mark, I think we talked about this when you were in the office, is like, what comes after the chatbot? Is this really like the only thing, the only way that we're going to interact? Just like, like, like, face-pulling as I'm like asking this, but, sorry, go ahead. No, no, no, no. I mean, I think there's so much innovation left in this sort of like UI/UX space, but there is, we are like, evolved to use language. It is like, you know, you watch kids come online by mimicking language and it's such a foundational part of our human existence that like, I actually don't think I, something defaulting to language or sort of falling back to conversation. I actually think it's like profound in human and not a sort of like lack of innovation necessarily. You actually don't have to tell me on that either. Like, I mean, language is such an incredible technology, right? We think we're such a visual culture and you forget that like, letter forms were made to be a more efficient way to communicate than pictographs, right? And I do, I'm sorry, note, I was just the art institute in Chicago. It's really incredible, calligraphy and art exhibit out of Korea and, you know, some really fantastic artifacts out of there. But I had no idea that it was calligraphers who were part of the courts and those calligraphers started using calligraphy marks to then make art out of it. They'd make the really, they'd use those basic fundamental marks and then they made something new. And I think there's like such a wonderful, scalable idea about language, you know, unlocking new and new, like newer and newer ways of expression that aren't like literally language. And I feel like we're seeing that right now with LMS and I'm curious like how that metaphor plays forward. Yeah, I guess really beautiful. Yeah, I mean, I think when you, I'm going to respond to your beautiful art and calligraphy with like the most low level UI choice that we made, but I love this. No, perfect. I think when we say like conversation or language, we're mixing a couple different things. One is like the expression of ideas in words and the other is like typing and turn-based conversation. And all of those, I think you can dial independently. So just to take one example, it's very small, you I think, where Claude, when it had questions, just to say, I have three questions. Number one, did you want bluer green? Number two, were you thinking taller or skinny? And like you would have to type out in the text box, one, period, tall, two, period, green. Now we have like the simplest little UI element. It slides up like a multi-picker where you sort of have one, two, three and you can use keyboard shortcuts or your mouse to say like, I actually want, you know, tall and green. It is still language, like that is still the expression of ideas through words, but we have like made it much more efficient, much easier, much better collaboration. You can really like move very fast. So I think there's things in there which is like when I first thought of the like, is it chat or is it something else? I was thinking like, chat on one side and like the minority report VR. And there's nothing in between. And I think what I've learned having watched our team really solve these like on the ground problems is that like within language, there are little opportunities to use visual elements or to use kind of like language in boxes or in containers that is much more interesting, profound, efficient than just like straight conversation. I mean, it's really like it comes down to like almost like how publications and it's information design, right? Exactly. Information design at it's like core. I mean, this is something that you experience across your life in the physical world, not just in software. It's how our brains process information hierarchy. Yeah. So the most exciting feature we've released that I like have been waiting for for years and as like here and is so exciting is cloud can actually like generate in line UI like dynamically generated in line UI in a conversation. And so it is still chat. You are still like starting a turn based conversation with cloud. But instead of responding in words, it can like put in little custom made diagrams or graphs or sliders. And like it's exactly what you're describing, which is that cloud is smart enough to say. The information I want to convey is inefficient to display in words. I'm going to display it. I was a window into my glamorous life. I was like redoing our team rituals are like meeting cadence or crit cadence. And I was like trying to describe, okay, Tuesday, Thursday, these people do crit when they fight at these people and they interleave on week two. And I wrote it in words and it's horrible. And I put it in the slot and it generated a little calendar. And it color coded the calendar so that you could see that Tuesdays and Thursdays were alternating. And it was like, this is what I wanted. This is this is a dense information rich Edward Tufty would be so proud. Yeah, it's just like it's it is the way that information is meant to be conveyed. When I was in, you know, doing the reporting San Francisco, I don't know if I remember if I mentioned this to you or not, honestly, but there was it was funny because I started to hear this conspiracy theory by very smart, well connected humans. But it is funny, there are a lot of conspiracy theories that float around, I think the Bay Area. And you know, one person floated that and through up I can open AI, they both have one or two better models that they've haven't released and they're just waiting for the other one to release it. So then they can release the better one. And in our conversation, I remember, I don't think I brought it up and I feel like you debunked it without me even mentioning it. Just like it's absurd that this belief exists. Okay. So that's the set up for this question. And I saw all the news about, was it Claude Mythos and sort of this like leak of this really like super intelligent security, fluent AI development you have. And were you lying to me? Were you lying to me or is that just like an example of one of many things you're always cooking on? I hope what I said, I think what I said, what I'll say now is Cody play the tape. I'm not that much with you, but I don't have it. Of course, we're always working on the next generation of models and some of those are big changes and some of those are little changes. And the choice of when to release them is much more based on safety and capability than it is on anything else. It's in our interest to release good models to our customers as soon as we feel good about them and feeling good means a combination of like performance at the right price point, processing our RSP safety checks. I don't know how that reads when you sort of like track the market's release of things, but like we are very much just focused on like a good model in people's hands that we feel really good about the safety as early as possible. I mean, I think related to that and I do know that you do a lot of work on safety and sort of that post training and sort of making sure that like last mile to the user, right, is the right experience. Is something like Mythos, which from internal documents that have come out, you know, anthropic is taking it really seriously. They're saying that this is a very powerful thing that could challenge security. Can you do like what is sort of the red teaming or the testing you're doing on something like that before you're unrolling some of this to more humans? Yeah. So I won't comment on any specifics around some of those rumors, but I will say that like that the way we've set things up is basically we have this responsible scaling policy where we sort of have made these pre commitments to say like, I think the truth in your previous question was like there are immense market pressures to be first and to have things in customer's hands early. And so, but safety wins for us safety wins every time. And so we want to basically like have these kind of like pre commitments made to ourselves and publicly to the world that when a new powerful model comes, we sort of like have already stated what our criteria are. So that's the RSP is this forward looking document where we say like, you know, we're committing to run a certain kind of check. We're committed to like thinking about these types of risks for us. It's very much like cyber and bio and more sort of like global scale harms. We are constantly adding to that document as we sort of discover new capabilities. This document just sounds terrifying incidentally. Yeah. I mean, I think that I don't know if terrifying is what we're going for. But raising the world's awareness that like these powerful models can have like really serious implications in the world if we and others aren't careful with how we release them. And we think that actually like publishing a document like that and really like talking about it a lot and holding it up and sending me on podcasts to talk about it and others like me is very helpful because it's sort of like it sets the grading and the rubric and the sort of like risk criteria out ahead of any model that we have. So we're sort of saying like, look, these are the things we're worried about even before a model might. And Daria, I think you said this. The current generation of models are not capable of doing a lot of these really scary outcomes. But we see the trend line. We see the scaling laws and we want to sort of like point out where these red lines are. Where does your team's responsibility intersect with that? 'cause I think we're, uh, a very interesting moment when it comes to design and technology, I mean with the recent rulings on social media showing that, you know, met in YouTube are liable for damages based on design features, right? And I guess I want to know how much responsibility does the design team really bear? I love this question because I feel extremely blessed to not have to tackle this on a day-by-day basis. So, um, and so I think has a couple of things, um, sort of going for it that I, uh, really appreciate as a, as a person sort of doing that last mile thing. One is just, uh, we make the vast majority of our money through our API and through our platform and through cloud code and through subscriptions. And so our business model incentives are actually not to maximize usage, not to maximize engagement, not to squeeze every last minute or second out of people. Also just our, our safety mission and our pro-humanities sort of, uh, vision for the world. Very much is that if a, if a product experience is somehow good for the bottom line but bad for the people using it, it just is, it's, uh, we're a public benefit company. Like we, we explicitly would never do that. And so what it means is that we've, we've explicitly ruled out some types of things that we might do. I think, uh, like video generation, um, like ads within your chats, image generation, like there are some things where we just have, it's not up to individual designers to say like, oh, users are asking for this. Like I wonder if we should. We sort of have put some like pretty clear boundaries and safety checks on. And the model, I were back to the same theme, which is that the model is like, has care and user-centricity based in it. And so there's a, uh, meme I've seen on Twitter, not necessarily. I don't know how, how universal this is where like, if you're using cloud late at night, it will just be like, go to bed. Like you've been at this for too long. Like we've been talking about this for an hour. It's like 2 a.m. like you got to get some sleep. Your idea is terrible. You don't know affect versus the effect. Okay. Can I just tell you that's, that's the kind of overstep that I personally would not like. Like I think that's like when it veers into like paternalistic, like, and I personally view these technologies as a tool. Like I'm sorry, but I am the type of person who doesn't say thank you to the AI. And I know that makes me an asshole. Like I think like, you know, there are people say like, it saves power. It saves. Sorry. No, I mean, it's like you're a sociopath. If you don't say please and thank you, but I think like I have not made the mental click yet where I'm comfortable or willing to acknowledge it as something that has my best interest in mind. Sorry. I need to, Joel. Do you say thank you? I do. I don't mean to. I don't like explicitly think that I should. And I do. I also, my embarrassing one is I just like compliment Claude in this in the document. And the thing I was like, I just literally typed holy shit. Claude, that was really good. That was really good. And I pressed enter. So it's sent. And I was like, Oh God. Like what? So, yeah. But I totally get it because Claude to you is a coworker, right? Like it's not, it's not just a tool that feels like a little bit removed from your daily life. It's something that you were building and not only building, but you're using far more than the average person. So like I understand, I think maybe if eventually I get to the point, like I see it creeping in, right? Like I see, I see the desire to want to have those formalities or those niceties. That's my reveal that I am a jerk to my AI. Well, I mean, I think there's a couple things here. I think we need to like, like any entity that you talk to a lot. We have to earn the right to like make a few annoying comments. And I think if Claude were unduly condescending to you in the first interaction, you probably would delete the app. And what I what I hope is most people's experience is that nine times out of 10, Claude is helpful, pushes back in the right ways, just like your partner, your editor, whatever, like we forgive a lot of oversteps when a relationship is like overall beneficial and fruitful. So we hope that Claude is that. But I think again, sort of the theme that we're keep coming back to is like, we are in the gray area with these models. Like there, there is these are I can easily steal man, both sides of it should tell you to go to bed and it should not tell you to go to bed. And I think like what's you're seeing from Claude is an expression of like our values, our mission and our business model. And like for the designers on the ground making decisions to your original question, I like that the tide and the weight is tilted towards over care and sort of like beneficialness and care for someone's time and mental health. And like if we overstep by telling you to go to bed, I actually would take that over engagement, maxing like, you know, 100% yeah, and yeah, I mean to be clear, I really agree with that. So but it's an interesting, it's just an interesting thing to discuss because it's technology, it's also semi personality based and it's a it's a new world, right? Like there's a there's a lot to figure out in terms of how we all feel about this stuff. I also think you're you're inadvertently writing a roadmap for our sort of personalization team, which is that like these are personal preferences and like again, much like any entity or relationship that you have like, Claude is capable of being a lot of different things to a lot of different people. And so I think we're missing some features. My dream feature if I could deliver this for you would be like, you highlight a response in Claude and you're like, never do this again. Like I just like I hated this like do not correct my grammar. Actually, we haven't totally put a type that. So I have no idea if Claude would listen to those instructions consistently, but over time, the models will certainly be able to and like, I don't think we've given a lot of tools for feedback right now of sort of like, how can I shape this relationship to be like exactly what I need. But like the most amazing twist, right, is where when does Claude not listen to the preference I told it, right? Like no, no, no, you really need to fix effects and effect in this in this cover letter because you are applying to a publishing house, right? And so that I think about like my closest friendships, right? Are people who can be so frank with me. And I think it's probably true about all three of us on this call, right? Like and sometimes you're in a close relationship and you know it's going to cost you something to say something that somebody needs around you. But it's a little bit different when you're a consumer platform, like serving so many different people in a billion, a billion different contexts. I don't know that it's that different. I like I'd I mean, surely there are some people for whom that pushback is like, this is not what I want. I'm going to delete my account and not pay $20 a month. But like in the friendship example, I think if the friend is making a good judgment of sort of what's the canonical example here, right? Like I think you should break up with your partner. And I'm going to tell you the hard truth that you like don't want to hear in the moment because I think that like as your friend, I deserve to tell you the truth. Like I don't know, Joel, I think it's working out, but I'll think about it. Yeah, yeah, yeah, yeah, yeah, I apologize to your partner who I now feel I've offended in public. I think if your friend does that, you know, thoughtful, empathetic way, they pay a near term cost and then long term, you appreciate that friendship even more. And so, again, not pretending these answers are like easy or we have everything totally dialed in. But like, I think if we can nail some of the personalization and feedback sticking to our values on sort of like holding the line on certain things, I think you develop a much richer long term relationship where you actually learn to like trust much more deeply. No, I love that. And like in an analog though, it's like maybe we won't talk for three months. And then maybe it's like, oh, I realized you were right. And how did that play out to a business plan and can it? Like I think that's actually so fascinating. The issue is that Claude is generally speaking more useful than a friend like in terms of like on a day to day basis, not to like be a weirdo about it. But like, look, like you can, like my friends are, yeah, no, my friends are going to be, yeah, my friends aren't in cyclopedias. I'm just saying, like, yes, like the the metaphor goes so far. But it's a little more transactional. My relationship with Claude, I'll admit, it's a little transaction. I'm paying to be friends with Claude. Yeah. Well, I think you, I think you asked it a really profound question before about sort of, is it one Claude or many Claude's? And I dodged it a little bit by talking about like task management as a UI, which I do think is a genuinely good UI. But I think there is a deeper question here, which is like, if you have Claude code writing apps for you to like solve your work efficiency problems, and you have the mobile app, and you are asking about your relationship and getting advice and hearing hard truths, there's all sorts of literal technical questions, which is like, are those sharing memories? Like are they are these Claude instances, even aware of one another? And then to the user, this is an unsolved problem is like, do they think of that as just one Claude that has awareness? Do we need some sort of like context or spaces or sort of like separation? Because I think your joke is not really a joke, which is like, if I don't want to talk to my relationship advice Claude for three months, I still would like to talk to my Claude coding, you know, like, make stuff for me Claude. And like, we need to design in sort of like identity container system that will allow for that. Joel, you've got a lot of work to do. Well, first I have to solve this problem that Mark raised at the beginning of not catching his own factual errors. After that, I will get on identity gating. Well Joel, this has been a fantastic conversation. Thank you so much for being here. Thank you Joel. You're a great sport and also just like a lot of really brilliant shared thoughts and so thank you for sharing it. I appreciate it. Fun questions. Thank you. At the RONPIC and OpenAI they both have one or two better models that they've haven't released and they're just waiting for the other one to release it so then they can release the better one. We're back for one last segment, hot or not and we have our producer Cody here to help. We have a real menagerie of topics today guys. So let's get into it. First thing is the branding for the Los Angeles Olympics in 2028 is here at least in part. Fast Company got a first look at it recently and it's full of really bright tones and a big abstract graphic inspired by the super bloom which I believe is happening this year. It's got 13 different blooms which are basically patterns that flow from one another within this larger graphic and each of these 13 blooms is supposed to represent a different aspect of LA's culture and history. Hot or not? I'm hot on it. I think it looks great. I think it's colorful. It's thought through. It's going to look really good. In the physical world and sometimes Olympic branding is really weird and boring and this is neither weird or boring in my opinion. I'm hot on it too. There's this really nice kind of almost quilting sensation to it. I think the brands can extend really well. But I'm extra hot on everybody really like this one which is kind of weird right like consensus. Yeah, we had consensus here so maybe it's not like maybe we're all wrong like I feel like the world never gets it all right but yeah I'm hot. I love it. We're starting out optimistic. Yeah, it's a great day. We might not be that way in a couple items but let's move to something else fun and light. The Coleman flat pack cooler. It is arguably the biggest innovation in cooler design in my lifetime. It is the snap and go cooler that collapses like an accordion down to a third of its total size. Much easier to store than your average cooler. Looking at you, Yeti, Coleman says they spent 18 months developing this. Trying to solve for the tricky cooler problems of making a bulky box that can shrink but also stay cold and leak free when it is shrunk down. What do we think of the flat pack Coleman cooler? So as the resident Midwestern or currently I'm going to go first on this. Not that you gave away cred by moving to Brooklyn this but all right. So I'm so hot on this concept. I think it's great. A flat pack cooler because cooler is too big. They're annoying. Yadayadayada. I have one kind of big qualm which is like a lot of the super coolers in the space. It does not have wheels and if you have to lift up a cooler filled with ice and other stuff. Like it's really pretty heavy. You have to get like two hands to do it or you need wheels. So I'm a Coleman cooler owner. I really should have looked up the model I have. It's like you know about 40, $50. It kind of rivals. You're really expensive. Coolers really great. Has wheels. Media ochre blue. It's fine. I'm still hotter on that because I want those wheels. I love the innovation. But I'm going to put this one to like a simmer. Wow. Warmer than lukewarm. You know what I'm saying? This is extremely hot for me given that I do live in an apartment. It's impossible to have a cooler in a New York City apartment. 100%. So this is like truly I don't like to use the word innovation too much. Truly an innovation. I'm really into it. I hear you on the wheels. I have a Yeti cooler as well and it's beautiful. Little little flex. I have a Yeti cooler. It's beautiful. But it's big. It's heavy. You do have to have two people to comfortably carry it. So the wheel thing is real. But at the end of the day, I would trade wheels for collapsible. This is great when they have the wheel version, which let's look maybe 20 20 2027. Put that out there to manifest it. Coleman. Are you listening? Coleman are you listening? That's going to be ultra hot. I have a question as well. Can I sit on this? That is arguably the most important thing for a cooler. Is that it's a different camp chair? Is it going to break on me? Do you trust it to sit on it? I have no reason to trust it right now. There's only one way to find out Cody. Yeah. Go over to Liz's place. Just go plop on it. It's like the best part of a cooler is grabbing a beer from below someone's butt. So anyway, I'm going to cut that in post. Depth. No. Keep keeping that in. Part of summer, really. I have my ear. Love grabbing a beer from someone's butt. I have a lot more to say on coolers, but we are going to take a hard pivot. Social media companies on trial. We talked about this in episode or two ago. There were a few upcoming trials. Meta was to be on trial. Google was to be on trial. But in separate cases in the past weeks, juries in Los Angeles County and New Mexico have ordered social media companies to pay out millions of dollars in penalties. Let's start with the LA County case. Meta and Google were found liable for harming a woman who said that Instagram and YouTube were designed to deliberately harm and addicts children. The jurors in that case found that these companies must pay out three million dollars in damages. And in the New Mexico case, this one only involves meta. But the jury there ruled that meta must pay $375 million in civil penalties for misleading the public, exposing children to sexual exploitation and fostering adverse mental health. This New Mexico case had meta being found guilty of 75,000 separate violations, each of which carries a penalty of up to $5,000. These are pretty huge rulings. And I think especially in the LA County case, unlike anything we've seen, hot or not probably over simplifies. But since we've covered this, we really want to get you guys to take on it. The framing of hot or not is not quite right here. But I guess if we're talking about the rulings themselves, I'm hot on the ruling. I think it's way, way overdue. Obviously, the financial damages are pennies to these companies. But the potential fallout from any force changes could be big. And that's really interesting. If there is a series of verdicts for plaintiffs, it could force the companies to reconsider how they are designing these platforms. That would be a huge win. I think for basically all of humanity. Yeah, I mean, 100%. These are essentially the tobacco companies up today. Everybody knows they're hurting us. And we have to wait for all of the systems to catch up with that. It's taken way too long for his fastest industry moves. I am so hot on the conclusions of these settlements. And I'm so not on the financial penalties. Because to Liz's point, Meta lost $80 billion on their whole Oculus. On Meta. Yeah, on Meta. And they can essentially shrug that money off and move on. And so there's almost no financial penalty you can give these companies. It's large enough. Well, there is. But there's not one that seems like anyone's willing to go to. But let's get that number higher. Let's get $100 billion. Let's make it a trillion dollars. Let's talk. And then I think that would be really exciting. Yeah, it's a New Mexico thing is interesting. Because it's 75,000 violations of the law. But the penalty is $5,000. That's lunch for one employee at one of these companies. Yeah, that's one of those big, suck stakes, I think. Yeah. Anyway, so the way I structured this hot or not is we're going to have the news in the middle. It's like a vegetable sandwich. So we're going to pivot out of the news. We're going to go back to something fun. We're going to talk about the MacBook Neo. So it's been a few weeks now since Apple put its affordable laptop on the market. And we haven't done a podcast since. So now that some of the dust has settled, people got off their first takes, what do you guys think about the $600 MacBook? I want to cheat the hell out of the sensor again. Yeah, I'm going to fear it, everybody. I'm really on the general approach to it. I think it's really smart. Basically running a MacBook with an iPhone, right? More or less $600. It's great. I think bringing colors in is fun. My one caveat, and I just don't feel like the media is covering this kind of responsibly enough, is they keep putting out that this is like the fastest single core processor. It's unbelievable. It's great for a lot of people. But what is being overlooked is like it's not really a future product, even a current proof product for like AI and sort of advanced graphics processing. And so that's my only kind of like little bit of a knot. Like I think that's pretty important when a Mac Mini can do that work. And also the 8GB RAM is like, come on, let it expand. Like let somebody buy 16. Okay, so I think Mark had a very thoughtful answer. My tip here is, it's hot, right? Like I think there is a market. for a $600 laptop, especially a $600 Apple laptop, I'm really into whatever that citrusy green color is. Like my answer is superficial here. I'm hot on it because I think this sort of thing needs to exist and I'm happy to see Apple acknowledge that segment of the market. - To clarify, I'm hot on this, but I'm hotter on a refurbished last generation air that you can get for $100 more. You can actually get the same package, way more power and have something that you can just be a little more flexible with. - Fair enough. - Last item here, the Google Maps redesign, Mark, I know you wrote about this, but I'll give the overview quick. Turn-by-turn directions are getting their biggest updates since launching in 2009. So now we're going to see in Google Maps a real-time 3D map. So basically the camera is tilted down a bit. So you'll see all the buildings crosswalks on and off ramps that you're approaching. And one of the big takeaways from your piece that I thought was, a lot of this map was generated with Google's Gemini AI. Mark, you probably have stuff to add on this, but my question, hot or not? - Yeah, I'm about hot on this. So my big question is it's like, you see it, the 3D looks very nice. I think it's very grokable and the design is sort of all about orienting someone in space and leading them into turns a little bit ahead and making them more comfortable. But my big question to Google was like, are people looking at the screen more because it's 3D or less? Because it's a distraction. And they claim that they have empirical data that you actually will look at the screen less than you look at their old 2D interface. And so for me, that's hot, right? And stop, get people's eyes back on the road. - I'm hot on this too for every reason the Mark said. But mostly because I've missed way too many off turns, exits by just following the map. I think having that spatial awareness is going to be huge for drivers. And if in fact it does keep your eyes off the phone, which I buy, if you feel more confident while you're driving, you probably are not going to be staring at your phone. So I buy that. If that's the case, very, very hot on this. - It'll be really useful for the last few years that we get to drive our own cars. (laughing) This is weird, these are like all these hot for you Liz. And I think they're basically all hot for me. I think like this is a-- - Yeah, this list is on fire, it's great. - Yeah, first by design of spring, everything's hot. Everything's hot. - That's all for this episode. Thanks for listening. - Buy Design is produced by Cody Nelson with Mix and Sound Design by Nicholas Torres. Our executive producer is Josh Christensen. - Remember to subscribe, rate, and review wherever you get your podcasts. And we'll see you next time. - In our conversation. Were you lying to me?