Taste Labs founder Thais Castello raises $18.5M to build AI for aesthetic taste and content quality

Speaker 2: Sorry for keeping you waiting. Thank you so much for taking the time to come chat with us. How are you doing?

Speaker 1: What's happening?

Speaker 10: I'm doing great. How are you guys? Thanks for having me.

Speaker 2: Thanks for hopping on. We talked about Taste Labs, the launch, the back and forth a little bit yesterday. I would put

Speaker 1: Did you mean to rage bait everyone with the name?

Speaker 10: I I knew there was gonna be some debate about the name, but it's actually it's actually funny. We registered the company like a year ago with that name. And so

Speaker 1: before the

Speaker 2: partner popped up. Okay. Okay. Yeah. You gotta for sure. What about this? What about if I boil down the core debate, it feels like a lot of people are very optimistic about your business. Like, there are just a lot of models that where the outputs even if you just put aside taste, like they're just not satisfactory to the user and providing more data, more signal there, that's valuable. Great. Good business. But then there's this other discussion over like, can taste ever be something handled by a machine? Is it so human? We'll never touch it with the most magical godlike superintelligence. Those are the two angles. How reasonable is that as a framing as is that of a framing for this discussion?

Speaker 10: I think that's very reasonable. But by the way, I almost went to even do the exercise of not even using the word taste to describe what we're doing because I think people got very caught up in the terminology. But, ultimately, I think one thing that everyone can agree on is slop is a problem. Slop is everywhere. It's not going anywhere. It's just getting worse.

Speaker 2: Yep.

Speaker 10: And so that's really what we're ultimately trying to solve is how do we raise this bar of quality first and how do we just allow us to even be able to create better things with AI. But it's not a substitute for human taste, and it's not trying to determine, like, one taste, if that makes sense.

Speaker 2: Yes. Yes. Yes. I think another thing people are worried about is, like, the idea of monoculture, which happens in Silicon Valley every once in a while where someone says, like, serif fonts are tasteful. Sans serif is no longer tasteful. Therefore, everyone must use sans serif or vice versa. And we're all like, wait, everything looks the same. This is actually bland. This doesn't feel as rich or tasteful as what you'd see in a thriving community that actually has a ton of culture to it. I'm interested to know though about the actual process. Congratulations on the funding round. But how are you thinking about imbuing models with taste? How much of this is about data labeling? How much of this is bringing on experts with particular backgrounds? Is this a very high skill ceiling effort where you're finding, you know, artists and and people with art history backgrounds? Like, who who is the ground truth? Where is the ground truth coming from?

Speaker 10: Very good question. So I think first, on the mono monoculture piece, I think that's exactly what we're trying

Speaker 6: to solve.

Speaker 10: Problem with STOP is this repetition that you see. Right? Every site now looks the same. Every site looks looks the same. No one wants that. And so part of what we're doing is building this community of experts that is at the heart of everything that we do that are a lot of designers, front end engineers, art people that can exactly, like, bring this whole range. I think we don't even as a team want the company to be a reflection of our own, like, personal preference. Like, think it's much more how do we map this whole range of preferences that exist and select the best people in the world at this so that they can bring to the table all their opinions and discussions of like when is something good, when is something bad, because also none of this can really happen in isolation. Right? I think Yeah. Yes, first, have to really raise this floor and figure out what's the minimum bar of quality. But then I think there is this multitude of preferences, and that's always going to be the case.

Speaker 2: Yeah.

Speaker 10: So think the case maker is kind of

Speaker 2: I I always think of in text the it's not this, it's that. It's the parallelism. Antithetical parallelism is the name of that term. And by itself, I probably read that exact phrase in actually, somebody pulled up an old Paul Graham essay that had that exact phrase, the forbidden AI slop phrase from a decade ago. And it's probably been in New Yorker articles and and famous books. But it was only when it became so repeated that it was in every paragraph of everything that was ever generated that it became a tell and it became somewhat I don't even know if it's less tasteful, but just annoying. And I think that Yeah. I'm interested to know, like, how much of how much of taste in your mind or just the, like, the quality output is driven just by variation.

Speaker 10: I think that's a very good point. Like, the feeling of repetition, it will lead something to be less tasteful over time. Right? That's why you have these, like, cycles of trends because you have kind of the people that are early, then the soon that it becomes too ubiquitous, it kind of loses its power. So it's this moving target. And I think that understanding how do you force diversity in an intentional way. It's Right? Not just, for example, turning up the temperature of a bottle and having it produce more randomness because that wouldn't be, like, intentional. Mhmm. And so it's like, how do you understand, yes, how to force that diversity but in a way that that's more intentional.

Speaker 2: Mhmm.

Speaker 1: Time what are your aesthetic super intelligence timelines? Do you think we're we can one day Yes.

Speaker 2: It feels like it feels like we're we're I mean, even just the latest revision of the models, the like, if you interact with the Frontier, the it's not this, it's that, like, is actually gone. I haven't seen it in a long time. Like and a lot of these things, like, fingers just went away, and no one talked about it because it was only fun to talk about when it was sloppy and terrible. And now people people are like, wow. Like, the victory lap was like two days in tech Twitter and then and then everyone And moved they and they found some new bone to pick, which is maybe the human condition will always find something to complain about. But, yeah, what what do you have like some end state, or or do you think this is a this is a life's work? Is this something that I

Speaker 10: I do think it's a bit of a life's work Okay. Situation because I think people have this phase now, like, to your point, how you fix the problem of six fingers, that's equivalent in design. Right? Yeah. That's equivalent in writing. Yeah. But once you're over that, then I think it's like, how do you actually raise the ceiling of that too? And that's a moving target. So I think part of this that's very different than training a model to be amazing at something like coding or math is that it's only if can train that capability once. It's something that you need to consistently retrain. I think a lot of the things the model right now is bad at is even these fundamentals, like things like vision, alignment, typography. But once that's solved, then I think it becomes much more of this problem of personalization, moving trends, localization too. Right? So Yeah. Yeah. Honestly, I think within, like, a year, a lot of the capability problem will be solved in terms of what we can actually produce.

Speaker 4: Sure.

Speaker 10: But then the the bar of ceiling, I think, is a much, much longer endeavor.

Speaker 2: Talk about transfer learning in taste, where taste applies more broadly. We were just talking to the CEO of Arm Holdings, chip design firm. In labs, they talk about research taste. I imagine that there's a lot of chip design that can be done just with deterministic like Verilog model, Verilog fine tuned model. But I would imagine if I actually went and talked to the best, you know, designers at ARM, they would say, yeah, there is a lot of taste in what we do. And I'm interested to know if you think there will be knock on effects As everyone probably goes to images, front end design, the visual taste of a visual designer. But I'm wondering if you see yourself as doing something that is in the path of solving challenges outside of visual taste and maybe there will be implications in things that maybe people outside the industry can't see the taste that goes into the design of a CPU, but in fact there is.

Speaker 10: Totally. So I think it's funny it's funny you say this because whenever I'm even explaining the company, a lot of people are like, there's space in code. Right? Like, I Yeah. I am very opinionated about the way that I write code. So I think how can we develop really a framework and almost this, like, infrastructure to be able to turn something that is so objective and something that's more objective or that can be codified. And that logic, I think, will apply to many things

Speaker 6: Mhmm.

Speaker 10: Way beyond just aesthetics. I think aesthetics is a super important one that we care deeply a lot about as a team, which is why we started there. But creative writing is another one. True point. Like, anything really where there's multitude of opinions Mhmm. Applies the same logic. So

Speaker 1: Yeah. What what does demand look like just this week? Sounds like you started the company around a year ago or at least incorporated it. I imagine that the haters would be in shambles if they if they got a look at your pipeline just given how much attention, attention the launch got. But what what's what's demand been like?

Speaker 10: The the inbound of the last forty eight hours has been absolutely insane. Don't get me wrong. It's actually the reason we took so long to launch, the reason we took so long to launch is actually because we already had more demand than honestly I think we could serve before even the launch. And so it's been a very low sleep week for me trying to catch up with all the demand. But yeah, I think both on the side of the labs and of the Appler companies, it's been kind of of bonkers. Still trying to wrap my my head around it.

Speaker 1: And then and then meeting that demand, creatives creatives I I've I've had the pleasure of working with so many amazing designers, photographers, creative directors over the years. And oftentimes, I have found that many of them are not not don't Oh, we are. Their life is not guided as commercially as many other kind of domains. Mhmm. Right? Like software engineers, lawyers, accountants, etcetera, where, you know, if account if an accountant has an opportunity to work with a company for a great salary and the company is doing something, you know, ethical or or not even just not illegal, they're like, cool. Sign me up. Whereas like create create creatives that I've worked with, I've I've at some points just offered to make an intro. Hey, there's this company and I'll just be like Vibes off. Vibes

Speaker 2: off. Yeah. There's no price to work with that fringe brand.

Speaker 1: How has it been talking with creatives who many of them are are already like met plenty creatives that love AI. They're like, this is amazing. This is like a new tool. But certainly many of them are also, maybe they wouldn't even admit that they enjoy it as a tool. So how's it been building out the network?

Speaker 10: So I think there's still definitely a lot of, division maybe within the creative community. Like, people that are really leaning leaning into it and others that are very, like, anti. I think what's been really cool is actually finding these creatives that want to be that British. They're like, hey. Like, I I do agree that AI is inevitable, and why not make it better? Right? Like, why not be part of the the solution to make this actually something good and that we can also use? And so that I think has been really cool. But you're right that they're way more picky, which they should be. Our our founding designer actually had we've made her a hat that says picky because I I totally agree with this. It's like we have to create a community that actually people want to be a part of and are excited about. But, of course, I think you're always gonna have people that are anti AI, especially, and that will be against it. And I think we have to find the ones that are pro.

Speaker 2: How, predictions? How many years until we see someone affectionately and nostalgically embrace AI slop? Like the slop core aesthetic. They're bringing six fingers in. They're prompting the latest and greatest models, the nano banana seven, and they're like, I want you to add six fingers because that has a retro feel that takes me back to Dolly two. And I want Yeah. Yeah. Maybe they just go back to the model if they're really authentic. It's like running the running the the iPhone photo through the VHS recorder so you get the film grain in there. Something like that. How long until until Slot becomes affectionately loved and reminisced about?

Speaker 10: I think it's yeah. It's gonna be kind of like the the taking photos on a on a film camera because it's because it has, like, a specific vibe. I do think that people do gravitate to these specific aesthetics and vibes.

Speaker 2: Totally.

Speaker 10: And maybe in, like, five I think you need a little bit of distance so they're not traumatized by those topics more. So maybe, like, five years Yeah. People will be like, oh, I really miss those, like, ali fonts and purple gradients all over my my page.

Speaker 2: Yeah. I'm looking forward to time. We had so good in 2026. I just wanna go back to that I know.

Speaker 1: So good. Well, do you think last question. Anyone that has, you know, worked on building a brand or a product, I think early on in their career, they can have this, like, frustration around the gap between their ability and their taste level. Do you think that's going away already? Because I even noticed like historically if I wanted to try to create like let's say 10 ago, I wanted to create like some type of render for for a product Mhmm. Or design for a product. I could go into Figma and like make a version of it very very quickly. And now like with a Let's say I have an idea for a joke consumer product I wanna send to to John. Now I can basically one shot a version of it that's like pretty easy and it feels like that gap at least initially has closed. Do you think that's do you think that's the case or are people's expectations just like already much higher?

Speaker 10: Gonna rise. Yeah. I think so I think people that love this mastery of the craft of design and that are, like, the peak of taste makers, I think, will always want them to keep improving even their ability to create to, like, keep up with their taste.

Speaker 6: Mhmm.

Speaker 10: But I think what we really don't realize is, especially not with AI, any person on earth can suddenly create. Right? Like, you you Yeah. Anyone can prompt. Anyone can can build something. And I think that group is a very large group that is producing a lot of volume. And if we can, again, both actually raise the skill to match people's taste, but actually even help to maybe improve people's taste and, like, show them, understand their taste better, or improve the things that they can create and so their their bar will get higher, I think that would be awesome. I think the craftsmanship will never go away of the people that are, like, at the peak of that field. But I do agree with you that I think you're gonna see this ability for you to suddenly match the level of taste that you were gonna be had as a as a consumer even though you weren't, like, an expert or professional like before.

Speaker 2: Yeah. Guys paid top dollar.

Speaker 5: You having to

Speaker 1: convince people with taste to move to San Francisco or are you setting up satellite offices in New York and LA?

Speaker 2: And Borschwicke and Austin.

Speaker 10: I'm I'm trying to convince a few to move. There is an amazing concentration of creatives in New York, and I'm actually from Brazil originally. Brazil is also amazing, like, aesthetic city. So, yeah, we have we have a couple of people that are spread out, but convincing a few to move.

Speaker 1: That's great. Awesome.

Speaker 2: Thank you so Well,

Speaker 1: congrats on the on the launch.

Speaker 2: Congrats on the round. How much was it? How much did you raise?

Speaker 10: 18 and a half million. There

Speaker 1: we go.

Speaker 2: Congratulations. And thank you so much for taking the time. Sorry we couldn't do this yesterday, but I'm really glad we got to digest I'm the

Speaker 10: gonna have to convince Jordan now to like the name eventually. That is my my one mission.

Speaker 2: We'll get that. No. No. I think you won him over just with the fact that you trademarked it a year ago.

Speaker 1: Yeah. Yeah. Yeah. I thought it was entirely was a week ago.

Speaker 2: It's a discussion, but a year ago you

Speaker 1: And I also like I I like that I don't hate tasteful rage bait, you know? I just have to deliver on an actual product like the business But you have to deliver on what you're Yeah. And it has to be aligned. And I think that every lab, every AI company, they want tasteful outputs. And I think you're going to do quite well. So great to meet you.

Speaker 10: To raise the bar. Thank you for having me.

Speaker 2: It's so nice. Great one. We'll talk to you soon. Let me tell you about Figma. Jordy mentioned it. Agents, meet the canvas. Your AI agents can now create and modify your Figma files with design system context. And let me also tell you about Railway. Well, our guest sits down. Railway is the all in one intelligent cloud provider.