Matt Mazzeo, former CAA agent turned venture investor (Lowercase Capital), joins Sam and Shaan for a wide-ranging conversation. The opening insight: AI makes proactive sales scalable — you do the work before the customer asks. Matt explains how he’s using AI color theory to style himself and others, then demonstrates the broader principle. The conversation covers the Mario Kart theory (multiplayer is always better), why taste won’t save anyone from AI, Apple’s podcast fumble, agent platforms as the next frontier, and Matt’s investments in Supabase and Replit. The second half is story time: Peter Thiel vs Matt for the Replit deal, David Bonderman’s railroad theory, the Derek Jeter truffle incident, and a full retelling of taking Kobe Bryant to visit Twitter HQ.
Speakers: Sam Parr (host), Shaan Puri (host), Matt Mazzeo (guest, former CAA, Lowercase Capital, venture investor)
AI as Proactive Sales [00:00:00]
Shaan: I think about Matt Mazzeo and the word “early” comes to mind. You were early at CAA to the internet trend — first to sign YouTuber talent at a traditional Hollywood agency. Now you’re early on AI. You’ve been texting me photos of myself and my house redesigned using an AI style agent. Can we talk about this?
Matt: It starts with my wife making fun of me. She said: your entire closet is black and white, you don’t buy clothes. So we spent a few months in Japan and Korea, and every person on every corner looked like they were out of a magazine. I realized I’d just never taken the time. So I decided to train o3 — which is spectacular now — as a Korean color theorist. Have you seen color theory on TikTok? They hold up swatches: this one brings out your skin, this one clashes. It’s like an allergy test for fashion. People spend hundreds to thousands of dollars on this.
So I ran color theory on myself. The model told me I’m a soft autumn. It gave me a list of primary colors, what to avoid, how to build capsule collections — brand by brand, measurement by measurement. My wife and I started building it for each other. Then for friends.
So I sent you an AI version of you — dressed up way better than you actually dress. And then there’s a button that can buy the outfit. That’s where we’re going.
Shaan: This is the golden nugget I took from you in the last few months. AI is not just a product — AI can be the go-to-market. Instead of a customer finding you, deciding to try it, signing up, paying, and then you do the work — you grabbed photos off my Instagram, grabbed my house off Zillow, showed me what could be. You did the work before I ever asked.
Matt: That’s exactly it. The cost of work dropped so dramatically that instead of selling the promise, you can sell the finished work. You sell the proof.
Think about headshot apps. It costs maybe two or three cents to generate a high-res headshot now. There’s an addressable market of ten million people on LinkedIn with terrible headshots. You can audit LinkedIn, generate a low-res sample for a penny, send it directly. Don’t ask “would you like a new headshot?” Just give them a new headshot and use that as the entry point to a much bigger service: your whole LinkedIn optimized, your whole profile rebuilt.
Shaan: My brain kind of exploded. Any business, any business, can now flip the sales model on its head. Instead of “would you like me to maybe do this for you if you hire me,” you say: “Check this out — I made this for you. Want to work with me?” Obviously that’s a better pitch. But before AI it was not scalable. Now it is.
Matt: And you string it together with tools like Clay. Real estate agents, for example. I know who the agents are, I know their listings — it’s all public. You run an AI agent to scrape their listings, stitch their photos into a social-ready reel, and email them: “Hey Steve, saw your listing at 410 Montgomery. Built you a reel for it. Here’s the link. Suggested changes welcome.” Steve responds, you say: “Want me to do this for all your listings? Here’s how to sign up.” The sales process is what a dealer does versus what a salesperson does.
This is like the friends who were doing Google ads in 2005 buying clicks for a penny, or doing Facebook e-commerce at a dollar CPM. It’s a distribution unlock that’s less tapped right now. The window exists.
The Mario Kart Theory [00:18:00]
Matt: Here’s one AI framework I keep thinking about: the Mario Kart theory. If you look at the last wave — SaaS, mobile — you could have made a lot of money just following one rule: everything is better in multiplayer. Figma, Notion, Airtable, Slack. Take enterprise tooling and make it real-time and cloud-based. Those things had network effects baked in.
AI today is all siloed. I’m having an individual conversation. Where’s group ChatGPT? You and I are in conversation all the time — why isn’t GPT in there with us? Why do I never see your outputs in my feed? There’s a lot we’d each be fine sharing.
Midjourney had this right — and it’s almost lost to history — but they launched on Discord. You joined a server, hopped into a channel, and there were a thousand people generating images. You’d see someone using it for T-shirt designs, someone doing anime. You’d watch them change their prompt and the image got better. “Oh, I never thought to add that. Let me copy paste.” We were all teaching each other.
It was bad product design on paper. But the one thing it was great at — putting you around other people doing the same thing — outweighed every downside. They could focus entirely on making the model better. The destination site they eventually built feels less social to me, frankly. The feed is there, but it’s not the same.
Shaan: Right — the style idea would probably be better as a social environment. You and I are different color types, similar body types. Maybe we have the same channel. You post a look, it actually works, you upload a photo — I see it work. Your success inspires me, teaches me, triggers me to use the tool. That’s a fundamentally better experience than me doing it alone.
Matt: Almost all of these AI experiences — outside of health or legal — would benefit from a free social tier. We collectively forgot a bunch of web 2.0 lessons. Social by default made everything better. The iOS apps all went social. Now we’ve forgotten. That’s a big market gap.
Taste vs. Volume [00:30:00]
Matt: Everyone in my circles right now who feels threatened by AI is saying: “taste is salvation.” But I think the taste argument is going to hit a wall — literally the consumer.
Do you know the Lisa Dol story? A decade ago, AlphaGo beat Lee Sedol at Go. Everybody said Go was safe — there are more possibilities than atoms in the universe. AlphaGo makes what looks like a mistake. 78 moves later, it wins. Lee Sedol retires. The quote: “I never truly scratched the surface of Go. I feel helpless.”
Everyone who’s saying “taste is my value” is going to have their Lisa Dol moment. Because taste is in the eye of the consumer. The consumer is the arbiter. You might win the award, but you don’t always win the choice. Billions of people prefer a McDonald’s burger.
Shaan: We made these mistakes already in content. Remember when it was: “what’s going to happen to the truck drivers?” And then it turned out the first wave of AI hit graphic designers and copywriters. The fake Joe Rogan / Steve Jobs podcast — I was like, “Oh, podcasters. That’s gone eventually.”
Matt: And the taste protection argument already failed in video. TikTok’s algorithm beat years of curated TV programming. The algo is literally the definition of taste — and it already proved it dominates. Apple started the entire podcast movement, owned every launch, and then decided the future was taste: great shows, no slop, HBO. On the other side, a million random creators with no tools, no budgets, no features, no data — built audiences that now sway elections. And where’s that consumption? YouTube and Spotify. Apple fumbled it because they didn’t think the creators were tasteful enough.
YouTube did the same thing. Half a billion dollars for originals — all the agencies pitched $5M channels. And the only things left standing were native creators who didn’t match the advertiser taste profile but made stuff people actually wanted.
Supabase and Replit: Building to Understand [00:44:00]
Shaan: Tell me the Supabase story. How did that investment happen?
Matt: During COVID, I was building a music co-listening product called Roadtrip with my friend Brian Wagner, one of the best product guys I’ve ever met. It’s like: you and I are on a phone call, two minutes of silence is excruciating. But put music on and we can sit in a room for hours. The music creates a substrate for conversation.
We couldn’t figure out retention — live, synchronous products are really hard — but in the process we kept hitting the same wall: Firebase sucks. Every social product runs into it. You’re trapped in the Google universe, migration is a nightmare, and they weren’t upgrading.
Brian said: “If you ever see a company that’s like an open-source Postgres Firebase alternative, that’s the ticket.” A week later I’m ripping through YC batch companies and I see: Supabase. Open-source Firebase alternative. I’m talking to my partner, I’m talking to Ben Tossell who made the connection, and we ended up leading the seed round at Lowercase and the Series A.
I never would have had that conviction without experiencing the pain firsthand. The outside of the problem looks like “Firebase does this, probably fine.” Only when you’re building do you know.
Shaan: So when you go into these experiments — Roadtrip, the Korean beauty TikTok shop — is it “I’m doing this to learn” or “I think this is the next big thing and the learning is the consolation prize”?
Matt: It’s curiosity. Sometimes I have an idea nagging at me and I just have to build it to get it off my brain. Sometimes I think this is going to be a huge category and I have to understand it from the inside. When I was building the video shopping product, I’d been spending time with the team at ByteDance, sharing notes on what I saw coming in the US and what they saw from China. Video commerce was massive in China. Nobody had cracked it in the US. I went and built the whole stack — found wholesale, spent hours at Korean beauty shops, built the TikTok store, did the fulfillment myself.
With Replit: when I got the opportunity to invest, I was already using the product obsessively. The first meeting I ever had with Amjad, I learned enough Python to build a small app in Replit — just to schedule the meeting. I sent him an app where he could pick a time. VC courtship.
Shaan: What stood out about Amjad?
Matt: The origin story. He didn’t have access to computers growing up in Jordan, so he’d work out of PC cafes just to get time on a machine. Same energy as the early Silicon Valley researchers who had to schedule time at research labs. He was obsessed with spreading computing access to everyone — one of the first people at Codecademy. True believer, technical enough to build it, clear on the pain because he’d lived it, and willing to be an endurance athlete about it. Years of building tools for teachers, years of infrastructure — before it was sexy. That foundation was why Replit was positioned perfectly for the AI wave. The vibe coding era made it look like a roller coaster, but it was built on a decade of patient infrastructure work.
Matt: My three AI investments post-COVID all follow the same framework: agent platforms built by world-class practitioners who are productizing their dark gift.
Cognition — the maker of Devon. Scott is a mathematical genius. I sometimes joke that Devon might just be Scott answering coding questions in the background.
Augment — Harish Abbott. He worked at Amazon fulfillment, then built Deliver (third-party logistics), sold it to Shopify for a couple billion. He is the ultimate logistics mind. No detail escapes him. His brain runs on routing optimization. And now he’s built an agent that embeds that expertise at token prices.
The underlying logic: when social media emerged, the best teachers stopped teaching in schools and started teaching on Instagram and YouTube. The best teacher should have millions of students. The internet vaporized geography. Agents are the same thing for labor. You’re no longer limited to who you can hire locally. The best employee in the world — globally — can now work for millions of companies simultaneously, just as the best teachers reached millions of students. The reinforcement training was done by the world expert. Now it’s priced at tokens.
Stories: Jeter, Kobe, Bondo [01:10:00]
Shaan: Let’s do some stories. Pick a card.
Matt: I got Peter Thiel.
The only time I ever encountered Peter Thiel was competing for the Replit investment. When you go against Peter Thiel, you plan to lose. I didn’t. I think it’s because any given Sunday, if you really love the product, if you’ve done the work, if you’ve obsessed over it, you can pull out the win. I was sending Amjad product thoughts at all hours — shower thoughts about the UI, edge cases, everything. Not because it was the art of the VC deal. Because I genuinely loved it. And I think there was a moment where he was like: I can have one of the best investors of all time, or I can have Matt. Matt loves me. That was his choice.
Shaan: David Bonderman.
Matt: David Bonderman passed recently. He was the founder of TPG — he helped build the Bass family investment strategy in Texas, made his first big exit on Continental Airlines, and built TPG into one of the titans of private equity.
I was on sabbatical and I had a list of people I’d always admired and wanted to spend time with. David was on it. I called Rick Hess, a close friend and mentor who was partnering with David on a new strategy. Rick said: “Dave’s really interested in AI. Will you put together some thoughts?” So I got a three-hour session with Bondo.
I showed him Midjourney, ChatGPT, walked through the new products. He was mostly quiet, riffing, asking a question here and there. At the end he said: “So, what are you going to do?”
I said: I’m going to invest in a couple friends who are smart and building stuff.
He said: “I’d buy railroads.”
I thought he’d gone crazy. He explained: there are two ways to play it. Everything in AI is moving so fast that any product could be obsolete in six months. That’s shifting sand. You can play toward the bleeding edge, but that’s where everyone is competing because it’s shiny. Or you can buy a railroad. Find the industries that aren’t going anywhere no matter what happens, and apply these tools there.
It was like talking to business Yoda. I ended up meeting him in the middle — building in an old, dirty industry that will exist forever, but applying the tools available today. He could say “invest in trains” and have it be the best business advice I’d ever gotten.
Shaan: Derek Jeter.
Matt: The most absurd Jeter story. At a talent agency you’re around A-list stars constantly. But something embeds in your brain with sports stars from childhood that makes them different.
I was a young VC at Lowercase. Jeter and his partner were starting a sports media company. We go to dinner at a nice restaurant in Beverly Hills. Waiter walks up — thickest New York accent you’ve ever heard. He sees Jeter and his face melts. He reverse-ages into an eight-year-old. “Mr. Jeter, Mr. Jeter, I’m a big fan.” Goes through the whole thing.
We all order the special — truffle pasta. He comes back with fresh truffle. He gives me a little. He gives Jeter’s manager a little. And then he gets to Jeter.
His brain freezes. His hand doesn’t.
He keeps grating. And grating. And grating. We start laughing while it’s still happening because it is a mountain of truffle that won’t stop. The entire plate is black. Hundreds of dollars of fresh truffle on Derek Jeter’s pasta. He’s just looking at Jeter, hand still moving, and says: “I think he got it.”
It transports you. It reminds you that these people exist on a different plane for a certain generation.
Shaan: Kobe Bryant.
Matt: I grew up a die-hard Lakers fan in LA during the Shaq-Kobe years. Christmas was Lakers on TV. Kobe was that person for me.
I was at Lowercase when my friend Betsy Skolnick reached out. Kobe was injured, thinking about life after basketball, obsessed with tech and venture. Would I come have dinner and teach him about VC?
My Jeter-truffle moment. I said: “Now? I can come now.”
We go down, have a three-hour dinner. What surprised me was his second gear of curiosity — he didn’t just ask the surface question. He asked every relevant follow-up, every “but how?” By the end he said: “This was great. Will you send me things to read?”
I put together the essential reading list — every blog post I’d ever loved. Thinking he’d never read it.
2 a.m., I get a text: “I read that post.”
We’d have conversations in the middle of the night. Eventually he said he wanted to come to San Francisco and meet startups. Again thinking: he’ll never actually do it. He said: “When are we going?”
We took him to early-stage companies, then to Twitter. Twitter had just bought Vine. We’re doing the tour, walking through the engineering floor. There’s a Ping-Pong room. There’s an engineer in there who is absolutely a savant — both-handing, in the zone, full matrix mode.
Kobe looks in and says: “I got next.”
Every engineer in the building materializes. They’re all on Vine immediately, filming this. Kobe goes — bang bang bang bang — and hits 80. The engineer goes — bang bang bang — and hits 82. The room erupts. The guy has beaten Kobe at Ping-Pong. He has never been more alive. He’s going to tell that story for the rest of his life.
And I remember Kobe’s face. Not fake pissed. Genuinely pissed. “I’m buying one of these. This never happened. I will destroy you the next time I see you.”
He walked out. We eventually put together a side fund with Kobe — made two investments. One of them turned out to be Stripe. But his business partner tried to renegotiate the terms, and we don’t do those kinds of things. That was the end of it.
Sam: Matt, this is amazing. Thanks for doing it.
Matt: Super fun. Thanks for having me.