Seeking Validation Is Your Greatest Superpower | Lucas Dickey | Fernish

Shamil Malachiyev: Hello everyone. Today's guest is Lucas Dickey, a founder who has pitched products to Jeff Bezos, launched Amazon's first cloud media services, co-founded VC-backed startups like Fernish and Deepcast, made meme games, AI satire and nerdware with his project Apes on Keys, and is now working on a stealth startup we're going to talk to him about today. He's one of those rare minds who blends startup rigor with artistic chaos, and I can't wait to figure out what drives him. Welcome to the show, Lucas.

Lucas Dickey: Thank you. I think that's the most glowing intro I've had today, and I'll definitely take it.

Shamil Malachiyev: So tell us, where does your story start?

Lucas Dickey: It's a good question. In the turbulent times we have today, in terms of geopolitics, which you're hyper-familiar with given your background, and the same could be said for many people in different countries throughout the world, I grew up as a military brat. My dad was in the US Air Force, and I had the benefit of growing up both in the United States and in Europe. I grew up in Europe in the nineties, when we had two Gulf Wars from a US perspective, the USS Cole was bombed, and I had frequent experiences in school where I couldn't go in because there were terrorist threats on the military base, and my school was on the base. It gives you a perspective many Americans don't have, that mobility plus living in different, extreme conditions, plus the idea of practicing good opsec, operational security, being very aware of your surroundings, who you're speaking to, and how to articulate things differently. What I'm getting at is having multiple perspectives and being able to be empathetic and step into a bunch of different shoes.

Shamil Malachiyev: It's not the most cushy childhood.

Lucas Dickey: Moving constantly, across geographies and different environments, definitely sets you up to smoothly, or at least competently, step into different environments. So the story probably begins there: moving a significant number of times between the United States and Germany until I was 18, and then professional and school life begins.

Shamil Malachiyev: Did you have a proper military upbringing, like they show in films?

Lucas Dickey: The military films make it look like it's boot camp for the kids, but it's not a military school. It's a Department of Defense school, a school system operated by the DOD. It's a standard school and class, but you still have those parents, so strict rules. Yes sir, yes ma'am. High degrees of being courteous and respectful, and those things are pretty important. My mom and dad are still yes sir and yes ma'am, and I'm 43 years old. Their parents both served in the military as well, so sir and ma'am was very common for me. That plays a part in who I am today. Whoever is on the opposite end of the camera, or the table, or the bar, could be service staff, could be Jeff Bezos, could be my mom, everybody gets treated with appropriate respect. The other thing from the military context is that you're in a very blended populace, because people come from everywhere. My high school in Germany had kids from like 50 states, including kids born overseas in Okinawa or Italy, because their parents happened to be stationed there. We're all thrown into this melting pot. If the US is a melting pot, these military bases are a hyper-concentrated melting pot. And being able to think about the folks around you, which, if you're a career product manager like I am, thinking about ideal customer personas and their motivations, the ability to switch rapidly between persona thinking is really helpful. I'm also a middle child, which means the neglected one, but the neglected one who will go along and get along in any context.

Shamil Malachiyev: The most neglected one.

Lucas Dickey: So I'm comfortable just fitting in, in a lot of different environments. In terms of upbringing shaping me, those definitely play a role. And if I were to attempt to be humble, ironic to say super and humble in one statement, post 30 or 40, I'm thinking a lot more about who I am and who I got to be, after years of therapy, like everybody else. Being a middle child, the ignored one, often meant that I did really well as a kid academically, so my parents mostly didn't pay attention. They were paying attention to my older and younger brother, who needed a bit more assistance. I was president of multiple clubs, captain of sports teams, and accelerated academically, which was easy enough to do at the university level and thereafter. Then it's, how do I get the validation I didn't think I was getting elsewhere, and how do I super-achieve to keep getting it? At Amazon, that means positioning myself in front of the most senior person to get the most done. Post Amazon, in a startup, the validation is going from one to 30 million in ARR. These days you might do that in a year; 10 years ago you'd do it over three years, which was pretty impressive. So the constant validation-seeking is also a constant driver. Every founder talks about the thing that makes them depressed or anxious. For me it's imposter syndrome, not having confidence in who you are in that position where you feel projected upon.

Shamil Malachiyev: Imposter syndrome.

Lucas Dickey: Thank you. It's definitely imposter syndrome. Even for the VC-backed businesses later, the imposter syndrome still feels real. Every VC pitch you step into, you have to justify who you are and what your existence is. If you live for validation and feel like you don't get it, that can be challenging. Or if you're pitching developers on a developer-centric product and they don't get it, that can feel invalidating. At this age, that definitely drove me through the first 20 years of my career. My wife would say it still drives me today, but now I'm trying to be a bit more conscious about my motivations independent of receiving the validation of others.

Shamil Malachiyev: It makes for the best life stories. If any founder asked me the secret to success, having spoken to so many successful founders, I'd say the main secret is to make sure you don't receive enough validation from your parents as a kid, and then seek it for the rest of your life.

Lucas Dickey: Because then you won't stop doing anything after that. This may be my trouble. I have a 12-year-old, a nine-year-old, and another on the way, and there's this hyper-concern around the coddling of the American mind, which is the name of a book by Jonathan Haidt. If, post tiger-parenting, you're hyper-involved and always telling your kids they're not doing good enough, which maybe drives the need for validation. Or you're always there, always paying attention, giving them everything they need, and they don't need more validation because you're validating them constantly, and then what happens when they enter the professional world and there's no one to stroke their ego? How do they stay motivated in that context? We're in this fun generational dynamic we're living through right now. Although you also see 22-year-old foundation-model developers building businesses. I don't know if you saw Scott Wu from Cognition Labs the other day, saying the team works seven days a week, sometimes past midnight, and that's the culture, and if it doesn't fit for you, maybe we're not the company for you. He's early twenties, so people are still finding validation, but there are multiple cultures at play. In the Bay Area, an immigrant-rich part of the country, there are cultural tendencies that aren't the norm of generic, dyed-in-the-wool white Americans who look like me and have been here for 250 years.

Shamil Malachiyev: I notice it a lot with immigrant parents and children, because when you're an immigrant you have to rely on yourself, and those are usually quite entrepreneurial families that demand entrepreneurial skill from the children as well, and that's the only time they praise them, when they show that capability to make something in the world.

Lucas Dickey: You brought home the extra bacon the family needed today. You can pick a bunch of American stereotypes, like the Korean immigration around corner-store groceries, very common on the west coast, in LA or in New York, but the kids all helped with stocking, or interfaced with vendors on deliveries, or you needed to help your sibling by going to do a babysitting job. In any of those scenarios you learned something. My wife and I always say, if only we could throw our kids into a much tougher environment so they'd develop resilience. Her upbringing was much harder than mine. Mine was more generic middle class, but both of us had very busy military parents, because my mom was in the military as well.

Shamil Malachiyev: How do you navigate the balance of raising your kids to be well-prepared for a world that's going to have a lot of uncertainty around AI, and at the same time making sure there's no resentment from their side when they grow up, that they still want to talk to you?

Lucas Dickey: It's a good question, and probably the perennial one from every generation with some variant. But we're living in this exponential age, and AI only accelerated it. I'm a big fan of Lenny's podcast, and anytime he has a guest on who's a father or mother, he asks what they're teaching their kids. Guillermo at Vercel, with five kids, what are you teaching them? Mike Krieger, CPO at Anthropic, what are you teaching your kids? A lot of them happen to have kids under five. So I'm thinking about it a lot. For me it's how do we instill, using an Amazon phrase, a bias for action, or just a desire to do the thing. It's interesting to have the thought, but can you follow the thought with action? So how do you engender curiosity in your children? A combination of curiosity and a bias for action is probably the most important thing you can have.

The more you understand the world at large, systems, chaos, the things that take part in whatever you're building, whether that's human emotion or business strategy, the more you benefit whatever you're working on. If we're moving toward a world of playing orchestrator, guiding multiple agents, having curiosity and the ability to think at a systems level, while also being the type that just knows I've got to get things done, you're probably going to be more successful in the next generation. You can have curiosity but no desire to get anything done, and then you're going to be a great consumer, sitting on the side of the table, but you're not going to bring net-new content, products or services into the world. So with our children it's very much curiosity and bias for action. I'll admit they have screens, which means we struggle with bias for action now, so we do things like accountability. My nine-year-old has to read, the ratio's always changing, but something like 20 minutes to get 40 minutes of screen time.

Shamil Malachiyev: Just like 20 minutes of TechCrunch and then you can play for a bit.

Lucas Dickey: Exactly, please jump on the information and summarize the VC news of the day, and then you're allowed to watch whatever you'd like. For both children, our daughter has to read a certain number of books this summer. You have to do it. Ideally it moves from us telling you to do it, to you saying that book was great, I want to read the sequel, or it was inspired by another thing and I want to read that. How can you instill that? With curiosity and application. And sometimes maybe trying too hard, as with most parents: you play video games, cool, how can I translate that into something commercially viable at some point in life, or just to expand knowledge, because there's someone designing that game, a storyteller, someone who did the branding, storyboarding, background art, physics engines, all this stuff behind the scenes, and that's just a game. So I'm trying to build a game with my nine-year-old. The initial ideation session was great. He loved it. We were chatting with advanced voice mode in ChatGPT, three-way rapport going on. The curiosity phase is great. But then he and I go to sit down at, in our case, Claude Code, and maybe it would have been easier if we'd used v0 or Lovable or Bolt, something immediately visual, like Replit. I was trying to teach him a little about basic file-system stuff that'll be helpful in the long run, and getting him to sit still was impossible. So I don't know if the delta between curiosity and applying the idea to codifying it into functioning code is the interface, did I put the wrong interface in front of him, or is it just that it was fun to shoot the shit about it, dad, but when it comes time to actually do it, nah, I'll let you go ahead and do it. How do I convince him it's fun to be the person creating that thing, coming up with the ideas, and that you've got this machine facilitating it, so you get to do the cool stuff and decide how it works, while it does the actual work.

Shamil Malachiyev: I've also noticed curiosity is easy to achieve when there's a lack of information and it's hard to get to. For children today, all they need is one prompt and it tells them the whole knowledge of the internet. In our childhood we had to go to the library, find books, find people, and that was an adventure. Now it's 20 seconds, there you go.

Lucas Dickey: It's interesting the way you put it, because it's like we were working for curiosity before. Now, if curiosity is free, do I even make the effort, because it's going to be there when I want it, in which case it atrophies, because you're not doing the thing. It's one area where I like that ChatGPT released study mode, and the concept is percolating across other chat clients: rather than giving you the immediate response, it uses a Socratic method to drive you to the final answer, in which case you're legitimately learning. As you respond, it continues to open the aperture of the thing you're trying to learn about, instead of showing you the entire globe, it shows you bits at a time. That's the way we learned, progressively and iteratively. We weren't thrown the entire answer. So long story short, it's some amount of hustle, grit, bias for action, and a lot of curiosity, because if the world keeps changing at the pace it does, it's the curious who benefit the most. They're not going to feel burnt out, they're going to say, holy shit, that thing's cool, how do I dig into it, instead of, there are 18 other things I could be paying attention to, and let it all wash over me. It's the ADHD curiosity that's going to benefit those people really well, and both my kids have ADHD, so I'm trying to nurture it for good instead of having it work against them.

Shamil Malachiyev: Let's go back to your early days at Amazon. What was that like? Were you much different back then?

Lucas Dickey: Me or the company? Those who've known me throughout most of my professional career would say I haven't really changed all that much. The exposure to more things has changed, and I've always been a bit of a curious sponge. If I can put myself in a position to learn more or do more beyond what was asked of me, for better or worse, I would. There are frequent points in my younger career where I wanted to do more than was asked, and had peers say, dude, you're making us look bad, we have this responsibility, we're all doing it, why are you creating more work for us? And I'd say, I already blew through my work, I'm ready for the next thing.

I did a company before Amazon. When I joined Amazon, I didn't join as a product manager. I joined on a catalog-operations team. For background, I was at Amazon MP3, which is now just Amazon Music. As with most Amazon products, because you need to define this new thing, Amazon just names it exactly as it is, so an MP3 file made it Amazon MP3. I think the first to break that mold was the Kindle. Because we were doing non-DRM-wrapped media, unlike your iTunes where you could only sync between five devices, we were launching Amazon MP3. In the catalog-ops team, user-facing features had a lot of engineering and product support, but the catalog-facing side had significantly less, so some of the activities were very manual. That could include debugging an XML file visually as a non-engineer. I happened to be a non-engineer with technical acuity, so I wrote bash scripts to do XSLT, a lot like JSON parsing, but at the time XML parsing, to find malformed things and fix them programmatically. All of a sudden I'm doing 20 times the number of records everybody else is. A couple of people didn't love that. There was one guy I had a bit of beef with, because even in typing, if he'd just used the tab key to move through the fields instead of clicking every field, he'd be faster, and I'd say, I know what I'm doing, don't tell me what to do. That's the challenge I sometimes have with coworkers: impatience. I've struggled with impatience for a long time. I just don't get why you're not here yet. I have to take a deep breath and figure out how to rearticulate the thing.

Because I have this drive, I have to recognize that not everyone has the same drive or is motivated by the same things. I wanted to raise the ranks at Amazon, I wanted the next level of responsibility. The individual I was referring to was perfectly happy having the job he had, because he had other things in his life that motivated him differently. He was there collecting a check and that was fine for him. In my case that wasn't enough, because I wanted to do more, which meant I had to outperform. So there was a rat race at Amazon, a highly competitive environment. I was a good fit for that era of the company, and contention was fine for me. Amazon really liked rich intellectual discourse, fighting both sides of the battle, what would now be called steel-manning arguments. That was natural for me.

Shamil Malachiyev: How much life was there in your work-life balance back then?

Lucas Dickey: In my Amazon era I was probably working 70 to 80 hours a week. I was maybe top 5 percent in terms of my work behaviors, because I really enjoyed what I was doing. It's why startup life called to me after Amazon, because working seven days a week on things I was really excited about was fine for me. My ex-wife and I used to talk about, do you live to work or work to live? In my case, despite having children, there's a big part of me that lives to work, because I like doing the things I'm working on. If I were 70 or 80 years old and had multiple massive liquidity events and were absurdly wealthy, I'd probably still want to be working on something, because I really enjoy the satisfaction of starting a thing, or working through to find the solution. I do have hobbies. I make music, I do plenty of activities with my children. But I find a lot of value in building and doing work.

Shamil Malachiyev: So you were this guy working in a technological mecca of the time, one of the largest companies. How do you go into the furniture business without any clue about how furniture is made? Does that happen?

Lucas Dickey: How does that happen? There were plenty of other things in between. My LinkedIn profile says something like aspiring polymath. I've always had an interest in a lot of things. My undergraduate degree was liberal arts, poli sci, philosophy and comparative religion, but I worked a technical job all the way through college, providing account technical support to students, staff and faculty at the University of Washington. I stood up a LAMP-stack blog platform for myself in the mid-nineties because I was tired of manually writing HTML and CSS each time. Little did I know that thing could have been called Blogger, and I could have been Ev Williams and sold it to Google for $100 million. I happened to be in Europe, and entrepreneurship wasn't what it was like in the Valley in the late nineties, let alone today, so I wouldn't have been aware of something like that. But having the technical acuity, and interest in lots of things.

In terms of getting from Amazon doing digital goods, there's an irony that I worked at Amazon, the foremost logistics company on the planet, Walmart might disagree, and I never worked in physical logistics there. Post Amazon I went to another digital-media company, doubleTwist, sort of the iTunes for Android, a hub for a music ecosystem if you wanted a non-Apple device. Then I did ad tech, buying and selling inventory programmatically through demand- and supply-side systems. Then a movie-ticketing company, where I was building their ads platform for full-page takeovers and live videos. Then event-access ticketing, a company trying to take on Ticketmaster, where I was working on biometric access and rotating QR codes, at the beginning of facial recognition in a commercial context. So I jumped between verticals very rapidly, and for me it goes back to mental models and intellectual curiosity. There's a new thing here, I have to deeply immerse myself in who the stakeholders are, what motivates them, the customers, the vendors, every player in the stack. That's been particularly interesting to me.

Shamil Malachiyev: Do you lose interest once you understand the whole stack? Is that what it is?

Lucas Dickey: No one's ever asked that as point-blank before. Four years into Fernish, the furniture-rental business, I did hit a point where I recognized that everything I could do within the budget available to us, I had done. So I ended up having serious conversations with my co-founder, our CEO and president, about whether it made sense for me to continue at the business. This wasn't them suggesting it, this was me suggesting it. I recognized that I'm highly motivated by things with a sufficient number of challenges, and I really like multiple concurrent challenges. When there are five fires burning and I'm expected to deliver, that's when I operate at my peak: consuming new knowledge, high ambiguity, forced to express a bias for action. That's a perfect environment for me. On the other hand, where we'd gotten with that business, the inventory management, catalog management, the e-commerce interface, the billing interface, all of it had been done, and we were in more iterative, marginal improvement. It was hitting a lot of optimization. I've often done zero-to-one and one-to-10, or one to 30 million ARR, but as soon as you hit that point, a lot of it becomes optimization.

Unless you're fortunate to make enough that you can create a new business unit. At Amazon, when I did digital music and then got to pitch Bezos, Jassy and Wilke on a new thing, that new thing became a new business unit, because Amazon is a big enough entity with the financial resources to do that. At Fernish we weren't in that position. We were actively trying to raise our next round, and Lucas was the second most expensive person on payroll. I had a really good VP of engineering we'd hired as employee number six, and a really good director of product who was promoted into VP of product and had done e-commerce her entire adult life. We've got two great people. Do I even need to be here anymore? If there's no net-new ambiguous, messy, hairy problem, it's not optimal for me, which tends to be why I excel at zero-to-one, because the whole thing is greenfield, blue sky, a big hairy audacious goal all rolled into one. So I occasionally struggle with that. The best way I dealt with it in companies before was to find a new thing within it. At Amazon, a year or two and a half in, I was like, is there something else we can be doing, and that was when I discovered Spotify live in Sweden only, before it made it to the US, and I was using a VPN to do an access model instead of an ownership model. That's partially the thing I ended up pitching to Bezos and crew, finding a thing that motivated me that I got to spend the next two and a half years working on. So it's a constant struggle to keep myself excited and intellectually driven. And this age is perfect for me. Last night I had a terminal open, using Warp and Claude Code, working on four separate applications in parallel, writing requirements as I went, clarifying technical details, jumping around doing design and logos, all on four applications at the same time. I love it, because each is an independent problem I'm trying to solve, with a little load-balancing between them depending on complexity.

Shamil Malachiyev: And then having four separate investor meetings for four different apps, and finding four co-founders.

Lucas Dickey: That's what I should be doing. In this particular case the company I'm working on is Prompt Yield, and I'm working on reference applications as well. If we're this horizontal service, what are the different verticals that could plug into it? So while Martin, our founding engineer, is working on the core service, I'm doing a lot of applications that plug into it, hypothesizing what it would be like to be an application developer building on top of a platform like this. I'm working on three of those, and the fourth is a developer-experience tool for people vibe coding. I'm calling it a repo assist. It takes screenshots at the point you go to do commits every time, so if you need to review historical code, you can also review historical images visually. Let's say I moved through two days of work and realize my data model is wrong, or I implemented authorization wrong, something backend-oriented. If I'm a pure vibe coder, my instinct is to roll back four days, or two days, of activity. Now I can build the backend the way I want, but what happened to all those visual tweaks? The idea is it's taking screenshots of the application as you build, so you can always go back and say, cool, I dropped all this, but take a look at these two reference screenshots for the home screen, the settings screen, and a detail page. The data model exists, the underlying services are all there, so here's a visual representation for the model to quickly infer where I'm headed without having to work my way iteratively back. This is only possible because of the age we're in. So I love that you asked what happens if you run out of intellectual curiosity, and a lot of it is having to stay self-motivated and find new ways to do that. Sometimes it works and sometimes it doesn't.

Shamil Malachiyev: I found a similar thing with myself. About four or five years ago I realized that whenever I start a company, I need to make sure that within a year and a half I have somebody managing the operational side, because once it gets boring, just standard growth-mode things, I lose interest.

Lucas Dickey: It's an interesting question, because I know other folks whose curiosity peaks when you hit scale and optimization. There's a reason Google's SREs, site reliability engineers, live for uptime, live for reducing latency by mere milliseconds. Can I shave off three milliseconds? Because they know in search that three milliseconds can make a material difference in the time things get on screen and the conversion that follows. SREs love that optimization, or anyone who fine-tunes a model for three years straight. That's just not your personality or mine. I would prefer to do 80 or 90 percent of the work, get it to the viability stage, and then allow others to continue to optimize. It's obviously critical to continue to the latter if you can have massive success, but the beginning tends to be the most fun.

Shamil Malachiyev: That's something I do with hiring, because I believe there are two types of people, entrepreneurial and administrative. Administrative people love optimizing, processes, structure, making sure everything fits the book. Entrepreneurial people are creative, they love chaos. The easiest way to spot them is to look at their work table: the creative entrepreneurial ones have a mess.

Lucas Dickey: It works its way back. You can take the entrepreneurial side and, instead of calling it entrepreneurial, call it zero-to-one obsession. I love hiring the first set of people to do things. I was pitching a VC earlier today, and we were talking about how we'll compose the team I'll ultimately get financing to hire. We were talking about a mutual friend who's a legitimate polymath, advanced math at Oxford, loves philosophy, very well read. Near the end they said, it's clear you know a lot about a lot, the idea you're working on is amazing, you're clearly very smart, but not everyone can understand you, because you have your API layer, your surface area that they interact with, and my UX is harder in some ways for some people. If you're in my ilk, we vibe, but if not. So I think about that. When it came to recognizing where I might have shortcomings, having humility, or it could be arrogance, recognizing I'm going to win this thing, but the only way I win it is if I find the right team players. With Fernish, my co-founder actually recruited me to be his co-founder. He had a background in leverage finance and mergers and acquisitions for JP Morgan. A furniture-rental business is a float-based finance business, so you have to understand appreciation and depreciation schedules, cash versus accrual, finance and accounting.

Shamil Malachiyev: Because the furniture is paid in installments and you need a model to calculate and make it work.

Lucas Dickey: Correct, and ours was complex. You could rent for two months, for 12, or for 12 and then month to month after that at a different rate. The physical goods are being financed as well, so where does the customer's lifetime value and payback period of individual goods reflect versus the payback period against the loan we took to get the good? We weren't buying them at an atomic level, we'd have tranches of cash, but it still had to play out so they'd align correctly, so we'd ideally have paid off the good multiple times over by the time the loan was fully called, so it was profitable. I couldn't have talked about any of this circa 2017, but Michael brought me along, and I had a software product-management background and had been at multiple startups, within the first five employees multiple times. He hadn't. So we were good yin and yang. When we started hitting scale issues, moving beyond only operating in LA and Seattle to more sophisticated fulfillment networks, hub-and-spoke models, could we support San Diego from LA, Tacoma from Seattle, we brought on a chief operating officer who was my former boss from Amazon and had only worked in digital media for two years. The rest of her career was physical supply chain. She was a PM in charge of launching two-day delivery with Prime circa 2004 or 2005, used to open fulfillment centers with Jeff Wilke, who led North American operations and was ultimately consumer CEO of Amazon. She was a hitter, and we recognized we needed someone with that skill set that the two of us lacked. But in the zero-to-one stage, Michael and I loved trying to figure out logistics, supply chain, fulfillment, warehouse refurbishment ourselves. We're both driven by intellectual curiosity, but also recognize when what we can do is insufficient and bring in the right person. Hiring's a lot of fun for us. The same is true for organizational design. I suck at doing the thing myself, but have a ton of fun thinking about how to structure things appropriately. I'm a great coach and a lot worse of a player in some sense. I gave feedback to a senior engineering leader once about how he presented himself to his team, and said, you have to occasionally stop, because you're sucking up all the oxygen in the room, which means the rest of the team doesn't get to contribute unique, novel ideas that you're circumventing by laying the entire playing field out. And I say that despite the fact that I can talk 45 minutes out of a 50-minute meeting, so there's a little hypocrisy in it, but recognizing it. I have a ton of fun at the zero-to-one and one-to-10 stage with almost anything. It isn't just the building or ideation of the business, it's building businesses, people, and all the rest.

Shamil Malachiyev: When you and your co-founder were coming together, and you'd already had experience working in those startup teams of under five people, what advantage or knowledge did you have that he didn't, that you think was important back then?

Lucas Dickey: These days it's encapsulated pithily by everyone on Twitter with, you can just do things, you can just ship software. Until the last couple of years, people didn't fully wrap their heads around that. A lot of times, starting companies, there was an intimidation factor: we're starting this thing, now what do we do? There's no boss telling you what to do, the business plan is being created by you, every step is non-deterministic, and you're the one who has to shape the direction. If Michael chose to do this business again, he could have done it without me. We accelerated many things because it was the both of us together. I immediately had the network to bring on an engineer who wasn't interested in full-time employment but could get us up and going and help with the initial three engineers we onboarded, because I'm not a software engineer by training. You're not going to have me doing the technical whiteboard session, but I can talk about data models. Being able to bring that person on, and say, Michael, we don't need a fully fleshed website to bring on first customers, all we really need is a catalog, prospective customers we can talk to, some way to present it, and a way to implement what we're talking about simplistically. Duct tape and baling wire, a truly lean MVP, which we did. We said, maybe we're not going to have wholesale relationships with furniture providers just yet, and we don't necessarily have an opinion for the type of furniture we're going to carry. Furniture stores have a style: Ikea is clean Scandinavian, Restoration Hardware is high-end and cushy, brands have a representation of who they are. Did we need one just yet? What we ended up doing was Google spreadsheets for products Michael identified. I put together PDFs with price and image, then emailed them to prospective customers and asked, among these, are any the style and price point you're looking for? That's V1. Then I'd come back with a revised version. In the meantime we built a very basic checkout pipeline where they put in their payment credentials through Stripe and a place for us to deliver, so we captured their address. That was the software. If you're not a startup person, you might assume, per the Reid Hoffman-ism, if it's perfect, it's too late, but we had our first 20 paying customers on monthly recurring or annual contracts with minimal code written, and we learned a lot. We started developing an opinion on the furniture we needed to carry, the appropriate margin profile of the items, and the customers actually going to use our product. I had certain customers I went through 20 turns of PDF generation with, who still didn't pull the trigger, and I told Michael, that's never going to be our customer. We are not someone who uses an interior designer to go through their living room 18,000 times. Our customer is a little more strapped for time, and admittedly a little strapped for money, because our target at the time was younger millennials and older Gen Z who were moving every 12 to 18 months and wanted nicer furniture but didn't want to pay more for it, because if they were going to move again in 12 months, it would cost $5,000 to move stuff from A to B, and the goods themselves didn't even cost $5,000. So how do we set up a world for folks who wanted things like West Elm and CB2, domestic US brands, but weren't ready to pay $1,800 for the sofa? They could get it through us. Now we have a furniture type that our prospective ideal customer persona wants. What does Michael get out of Lucas? A person who steps in and says, we can get started on some of these things without a fully fleshed product. You could burn through millions of dollars before getting any validation from customers. Is there a way not to do that? And we did. By the time we went to raise money, we could say, we have 60 customers, and here are all the things we've learned, and the validation points we need to grow the business further. That is what I brought to the table that he didn't necessarily have: I've been there when the email addresses haven't even been created yet, when you have to pick a payroll provider, when it matters whether we're incorporated or not when we first start. I was also a big champion for Michael, knowing he had every capability, and that there'd be a world where I wouldn't really be necessary for him anymore, because the company, e-commerce and reverse logistics, is challenging but known, and it was a business where he could continue to sink his teeth into those operational things. Unlike you and me, Michael loves operations. He wants to get the business bigger and figure out how to optimize and grow that margin profile, whether it's negotiations with vendors, pricing strategy with customers, or the way you hold the money so there's overnight treasury being generated. That thinking about capital is exciting for him in a way it wasn't for me. Now it is a little bit, because I went through an experience as a rental-tech chief product officer as well, where we did things with overnight flow, sitting on other people's money and making interest against it, so I have an appreciation for compounding interest against other people's money. But the zero-to-five-people stage helped us get going faster. And going back to imposter syndrome, it probably never really goes away, but if you've done anything like it, being within the first 10 employees, whether at Amazon or a VC-backed startup, then going to be the first two people, I at least had the purview of seeing what was done elsewhere, some of the mistakes made, or a network we could leverage through other people who'd been in the same position.

Shamil Malachiyev: Why Deepcast? Where did that passion come from?

Lucas Dickey: Deepcast was interesting. When I was at the rental-tech company, which I won't name, I was just miserable. By the time I joined it was post series B, which maybe goes to my point about zero-to-one versus beyond. Most of the formative decisions had already been made, and material decisions for the next year had been made. So for me to slug in as a C-suite level with a roadmap already established, and skeletons already in closets, and things we needed to wind down, it just wasn't a lot of fun. Deciding what I wanted to do next, I was contemplating a venture role, an operating partner or platform team member helping those making the investment decisions and supporting portfolio companies. I found a firm I was familiar with, and a partner there said, hey man, instead of coming to VC, why don't you be an operator again, I think that's who you are, and I have this idea, discovery is really horrible in the world of podcasting, and has been for the last 15 years, can we do something about that? It feels like the technology is here now if it wasn't before. That was the inkling for Deepcast. There were other things at play. With the release of GPT-3.5 and ChatGPT, I wanted to spend my career going forward working in generative AI or anything vaguely in the world of AI, because in terms of curiosity it's a space that's constantly moving and I can constantly be learning. I'm not a physics or applied-math background, so it forces me to level up my learning and be a bit more academic.

Shamil Malachiyev: So you spend most of your time listening to Andrej Karpathy.

Lucas Dickey: Whatever his speeches of the given day. Half my podcasts are admittedly pragmatic engineering and how-I-AI, from the application layer all the way to a three-and-a-half-hour interview with the founders of Windsurf or Cursor. I love those conversations, and I take notes and think, they talked about this fine-tuning technique, I don't know what that means, I'll go do a little deep research on it, and then I feel more well-versed. I liked what it meant for range generalists to take these tools and do more with them. In our case with Deepcast, it was how do we solve this problem of search and discovery, partly because the only thing to work with from an indexable perspective was the metadata attached to an episode, and as a podcaster you have roughly 1,000 to 2,000 characters. That goes into Spotify or Apple or whatever people are using, with Spotify, Apple and YouTube accounting for 90 percent of playback, so you're optimizing your metadata presentation there, but you don't entirely know how their search algorithm works. Let's say it's a DoorDash episode and someone talks for three and a half hours, or it's Lex and he decides to go crazy and talk for eight hours straight. How do you encapsulate that in 1,000 characters? With the range of conversations happening, and elections being moved by podcast conversations, international diplomacy being moved by them, business conversations moving markets, we said, can we take advantage of the fact that speech-to-text and automated speech recognition has improved in efficacy and come down in price, and now with generative AI and LLMs we can process summarization, taxonomy and takeaways way faster than you ever could at human scale before? People had tried summarization before Deepcast, completely human, mechanical-Turk-driven, and they couldn't scale, they couldn't get past a couple hundred podcasts, because they had to employ an army of people or maintain a volunteer army, which is really challenging. Now, all of a sudden, we could do summarizations of thousands of episodes per day and do all this data extraction. So the motivation was, I'm a podcast junkie. Yesterday I was listening to TBPN. Are you familiar with TBPN?

Shamil Malachiyev: Or what the acronym stands for.

Lucas Dickey: Honestly, I don't know, but it's on YouTube. They're kind of like an MSNBC. It's two guys, early thirties, Jordi and I'm blanking on the other host name, live-streaming for six hours straight, whatever's happening in a given day. Scott Kupor was on, who's now serving as head of the Office of Personnel Management, OPM, but Scott Kupor was a GP at Andreessen Horowitz for 25 years. He was talking, and I thought, what if I wanted to dig more into what Scott Kupor was talking about, or his history? You couldn't do that in any podcast platform before. In ours you could search, and beyond finding the episode and seeing the summarization of the takeaways of what he said, and sharing that with friends, I could use him as a node and jump to other episodes discussing him. If I'm a public-market investor curious about the IPO of Figma, and then the 250 percent pop, and what that means for the next week, it'd be great if I could jump between episodes mentioning Figma or Dylan Field. You can't do that in Spotify, Apple or YouTube. So we accomplished those things. Unfortunately, there's a user-behavior change. You had to say, this is a second screen, it doesn't exist within your primary screen to use, like the IMDb-to-Netflix example. Consumers just weren't interested in that approach. There was a lot of power-user interest, but getting mass-user interest, which we'd need for venture returns, wasn't there. I'm willing to bet that almost everything we worked on will manifest in every Spotify, Apple and YouTube, and they're doing it iteratively. I brought up topical analysis and entities to an executive at another podcast company about three months after we launched it, and they had their version live six months later. As a consumer and someone who wants the best through the sparring of capitalism, I'm happy they introduced that feature, because now it's better for everybody. It's a bummer we couldn't get consumers to use it on our platform. Deepcast was very much driven by my own idea, mine and my co-founder's, around wanting to discover new content, especially mid- and long-tail content we might otherwise not discover, and the knowledge-retention side, taking away parts of an episode to store in Obsidian or Notion. But that got me the next step toward what I'm working on now.

Shamil Malachiyev: Can you tell us maybe some hints on what you're working on?

Lucas Dickey: The new thing? I don't mind. I'm stealth-moding it on LinkedIn, but on AI Twitter I'm being fairly open about it. Ultimately, one of the things we discovered at Deepcast, I was alluding to bouncing between episodes based on people, places and products. Those are called named entities. A non-named entity could be a shoe, but a Nike dunk is a named entity, a proper-noun entity. We used those as points of discovery between episodes, and we were also linking out from those to the open web. So if Elon Musk was a point of connectivity for episodes, you could click on it and it would take you to his Wikipedia page. It was kind of an anti-pattern: a lot of product managers try to capture everyone on your website and never let them leave, whereas I wanted to give them an opportunity to discover more of the thing they've expressed interest in. We never really got there with that feature. When we first rolled it out, it was the era of LLM training sets always being four months trailing the current date, no fine-tuning to keep the data fresh, no web search built in, and the LLMs through their APIs didn't have access to it either, so the corpus of known entities and associated URLs was fixed. On top of that, the sycophancy of trying to satisfy humans was arguably higher and more pernicious, because it was more subtle. This is actually something it still has a problem with today, and as I'm pitching VCs I keep using this example. If you go to ChatGPT and say, here's a list of my 20 favorite books, can you return Amazon links so I can go buy them? I ran that experiment last week. Ten of the 20 were completely hallucinated Amazon URLs. They were close enough that at a human glance they'd look right. Amazon follows a very distinct structure: amazon.com, hyphen-separated title of book and author, slash, then dp, slash, then an Amazon Standard Identification Number, or ASIN, that hasn't changed since 1996. So if you're an LLM doing next-token prediction, you can predict that the URL you mean is this title and author, because you already know that from your generalized knowledge, and that little alphanumeric string, you can probably predict what it is. That's me anthropomorphizing what the LLM is doing, but that was happening a lot, and it's still happening.

The thing that occurred to me, in a meaningful way, was that there are literally millions of product mentions generated daily, and none of them were being monetized. And not only were none of them being monetized, but, maybe even more important for consumers, I had to go through all this effort to get to a thing I was trying to find. The example of, can you give me the links to the books I want, and it still didn't work, so I'd have to manually look those up. The reason I was trying to get those 20 books was that I was putting them on my website and wanted to link to each. ChatGPT wasn't able to do it programmatically out of the box, so I had to manually look them up on Amazon one at a time. On the building side, we're early days, the company was incorporated only three or four weeks ago, but we're moving at this exponential pace that is AI. I have a full-time co-founder, founding engineer Martin, building a fine-tuned model, an RL fine-tuned model, that takes into account being able to disambiguate different entities. It's one thing to note a dunk from a product-entity perspective, there's a type of cookie called a dunk, there's also a Nike shoe called a dunk. Knowing the context within which the entity is mentioned, we'll do a better job identifying which entity was intended. So entity disambiguation is happening there. The other part is figuring out whether you're driving them to the right merchant destination to buy it. And the third part, without giving away all the secret sauce, is whether it's a valid URL, doing URL validation to make sure the URL you pass someone is actually the right URL. Then, on top of the model, a service abstraction layer, so application developers of any vertical type can plug into our platform and get back high-quality URLs associated with named entities, specifically commercial named entities, and ideally affiliate links so they can generate revenue. So in your case, if you had a corresponding newsletter or blog, when we summarize this episode, I've probably already mentioned 50 products over the course of this conversation. If you had affiliate links for all 50, you didn't have to negotiate a direct sponsorship. If it's web-based, it's AdSense but without the clunky visual ads that don't fit your style. It's in-situation, in-context, in-natural-language, embedded with the rest of the copy. It could be an SMS-based chatbot, a chatbot embedded within a website or native app, or someone's recipe site hosted on Squarespace. In any textually based one that mentions products and is looking to monetize, those are all prospective customers. On the flip side, merchants are looking for a bigger surface area for people to discover their products, and there's too much channel consolidation between Google, Meta and TikTok, such that paying programmatically in each is expensive, so they're all looking for net-new ingress points into their stores. There's fun tech at play with RL, where we can give a bunch of information back to merchants and say, we've recognized that for a shoe type you sell, it performs better within application type A, with user type B, when your marketing copy is this way, because the same shoe is sold across four different merchants but merchant A is always winning and you're merchant B and never winning, and the only difference is your marketing copy. Any product detail page with insufficient information is less likely to lead to a conversion. It could be pricing, it could be any number of things. So it will behoove merchants to be on our platform, because we'll give them good feedback and surface their purchase opportunities through a network of sorts. I'm loath to call it a network, because we're a service layer upon which other people could build their own network. We keep calling it an AI monetization primitive, a layer above inference, search, memory, compute. It is affiliate, but I prefer to call it natural-language or contextual advertising, because it's around the context of the thing. It used to be called native advertising, Taboola and Outbrain did this in the late teens, but it was a very manual process by comparison. Now we're in the era of generative AI and can do more with the technology we have today.

Shamil Malachiyev: So once ChatGPT starts bombarding me with ad links, I'll be like, thanks, Lucas.

Lucas Dickey: It's interesting, ChatGPT is probably one of the few places that may or may not use what we're building. This was a VC question as well. Perplexity is not going to use us, because it's core to Perplexity that they get these URLs, since they're positioning themselves not just as a knowledge engine but as a commerce engine, doing as much around agentic commerce, booking flights and hotels, and the Stripe integration that lets them do more direct sales without leaving Perplexity. OpenAI, on the other hand, has the foundation models and Sora and Operator and Codex, and ChatGPT is whatever it is now, like 600 million users, many of them consumers, and whether they choose to optimize for commerce is a bit of a TBD. They're doing experimentation, but not to the extent they're focusing on many other things, like improving the experience for developers, because that seems to be working really well for Anthropic and its enterprise business. So they have a lot to tackle. I was also sharing thoughts with a VC around OpenAI and Anthropic, and the front-end model providers who are all using Surge or Scale for data labeling. Could they have been doing data labeling with humans themselves? Yes, but they opted to have their researchers work on the core technology and bought the labeled data elsewhere. So with ChatGPT and potential links with us generating, it's possible they say, these guys at Prompt Yield have optimized everything to get high-quality URLs that aren't link rot, aren't 404, are exactly the entity the user is looking for, and have given us an interface to leverage in a very low-latency, high-speed fashion, maybe they end up using this. Or, knock on wood, we could end up being part of one of these companies' ecosystems, at the pace of their acquisitions. But we're not targeting this one acquisition. For most folks, fundraising is not something you can talk about while actively doing it, because there are regulatory rules around it. So let's just assume I'm fundraising right now, and these are the sort of conversations I would be having, hypothetically.

Shamil Malachiyev: Yeah, hypotheticals.

Lucas Dickey: As much as Fernish I was excited about, and furniture still is a compelling business, Berkshire Hathaway owns a company called CORT, a furniture-rental business in the United States valued at about five and a half billion dollars privately, part of the Berkshire Hathaway holding company. They're doing it at an 80 percent gross margin. They own the railway the furniture goes on, they self-insure through Travelers and GEICO and a bunch of insurance providers, they own Nebraska Furniture Mart, the third-largest furniture manufacturer globally, which is mostly white label. So they're fully vertically and horizontally integrated, a phenomenal business, but still valued at five and a half billion. That's a great brick-and-mortar traditional business. On the other hand, Prompt Yield in my head is the AdSense size of market. If commerce is going to be happening in more and more native ways, integrated more tightly into the rest of our web or mobile-native experience, because there are agents in the mix, Prompt Yield represents a significantly larger opportunity. So there's an element of excitement thinking this could be really big, and that motivates me. A lot of these zero-to-one times, it's, can this be bigger than the last thing? It probably goes back to the validation thing again, can I get validation from more people because I'm doing this thing that provides them value? Talking to another VC, he said, Fernish, I didn't realize you were the co-founder of that business, I'm in New York, I actually used Fernish three years ago, it was great. When I hear those, that's the validation you're looking for. The equivalent may happen online. Even just knowing people are using it, and maybe the experience could have been improved, there's validation there. They believed in it enough to give it a chance, and now the question is, can you earn their trust back?

Shamil Malachiyev: What would you advise business founders who want to make sure they don't miss the moment where they're able to advertise their services and products on LLMs? How do you prepare for that, and how do you spot the right moment?

Lucas Dickey: That question can be taken a bunch of ways. One is, we're at the advent of agentic commerce and you're an emergent company, what should you be doing? Interesting enough, today Shopify just announced agentic commerce in early access. For Shopify that's one of three separate services. One is Shopify catalog, for their partner network, effectively their affiliate network, to get better access to their catalog through an MCP interface, which is semantic, so instead of needing deterministic names, you can describe types of things and it can surface. Shopify is a great endpoint for us to connect with, because they're thinking about the world in a similar way, making that catalog layer easily accessible and high quality. They also said stores themselves should have the option for an MCP layer, so a store can make itself like a custom GPT, and now that MCP layer has access to their in-stock inventory and the ability to facilitate a purchase, because the Shopify MCP store is behind it. So if you're a merchant, or you develop products, CPG, DTC, staying abreast of that stuff is important. You're either giving over trust to the merchant and hoping they figure it out, or taking some ownership yourself, but either way, make sure you or your service provider are leaning into the fact that there are going to be multiple different endpoints to buy the stuff that are very different than before. What does that mean for me, do incentives change? If it's affiliate-based, am I comfortable doing it at higher percentages on newer platforms to drive faster adoption, and then renegotiate the affiliate rates later? The same could be the case for agentic advertising, a bid-based model more like traditional blue links with Google. If I'm a merchant, what am I doing across every different vector, geo, or AEO, or whatever they're calling AI engine optimization? What do you need on your page so that when the bot scrapes it, it pulls the right semantic information to better present you within a chat client? I still think it's TBD whether that's working as expected, but there are certainly 20-plus companies and a lot of money going into that space. Regardless of whether it's the final one, you need to be cognizant of it, which means, as with everything, things are getting more complex. Doing what you did before is not enough. Spending into channels you already know are too expensive with paid search or paid social, you want other channels anyway, so you should be exploring ways for your merchant interface to be accessible through purely agent-driven, or human-driven with agents in the mix. In our case, the recipe-website example, that's a Web 2.0 service, a publisher putting together recipes, but if I'm a merchant, maybe I'm Amazon Fresh, what do I do to make sure those recipe apps can connect the recipe to the store? Prompt Yield, ideally through an interface, will be a good way to do that if we play our cards right. So the point is, they have to be cognizant of the changing landscape of having agents in the mix and humans supported by agents.

Shamil Malachiyev: I love it. I'd say the summary of our session today is what Steve Jobs used to say: stay hungry, stay foolish. Stay curious is something we added.

Lucas Dickey: For sure. Curious, foolish and hungry probably are me to a T, and I'm okay with being wrong, so the foolishness. But I'm not good with being inactive or insufficiently curious. The biggest insult for me would be being called boring or ignorant. Those would bother me far more than many other things one could say about me.

Shamil Malachiyev: Thanks so much for coming onto the show today. This has been great.

Lucas Dickey: I very much appreciate you having me on and letting me suck up the oxygen in the room and talk about myself a little bit.

Shamil Malachiyev: Thanks so much. That's a wrap.

Lucas Dickey: All right. Thank you.