Matt Mazzeo

AI dropped the cost of work — so sell the finished proof, not the promise

Matt Mazzeo's core thesis: AI didn't ship a new distribution channel like mobile or the App Store did, but it collapsed the cost of doing work so far that you can now deliver the finished deliverable before a prospect ever buys, flipping the sales model from selling a promise to selling proof.

“But what did change was it let you do the work in advance. The cost of work dropped. The cost of work dropped so dramatically that instead of selling the promise, you can just sell the finished work, right?”

Steal thisBefore pitching a prospect, use AI to actually build the deliverable for them, then lead with the finished work instead of asking for a meeting.

Penny headshots: deliver the work unasked, then upsell the real service

Mazzeo's worked example: at ~2-3 cents per high-res image, audit LinkedIn for terrible headshots (an addressable market he pegs at ~10 million high-intent people), send each person a free improved low-res headshot unprompted, and use that as the entry point to sell a full LinkedIn optimization service.

“And you could literally just for like a penny, two pennies, like create a low-res version, right? Of it and sell it direct. Don't ask them, do you want to set up a whole new headshot? Just do the thing. Give them a new headshot and then use that as the entry point to a whole other part of the business.”

Steal thisGenerate the cheap version of the deliverable for free, send it cold, and treat the higher-margin second-order service as the actual product you're selling.

Mario Kart theory: everything is better in multiplayer

Mazzeo's investing heuristic: you could have made a lot in SaaS by only backing tools that are multiplayer by default — Figma, Notion, Airtable, Slack — because real-time, cloud-based, social-by-default products carry network effects. His knock on AI today is that it's all siloed, one-on-one, with no 'group chat GPT.'

“the Mario Kart theory is this is actually like you could have made a lot of money investing in tech over the last handful of years, um, or at least in the SaaS. If you just followed this one, if you just followed like everything's better in multiplayer, like, uh, like build things that are like multiplayer by default, like Figma, Notion, uh, Airtable, Slack”

Steal thisWhen evaluating or building a tool, ask whether it gets better when more people use it together; default to multiplayer.

Mazzeo found his Supabase investment by feeling Firebase's pain first

While building a failed co-listening music app during COVID, Mazzeo and his partner repeatedly hit the pain of being locked into Firebase. A week after his partner said an open-source Postgres/Firebase alternative would be 'our ticket,' he spotted Supabase in the YC batch and led its seed and Series A — a deal he says he'd never have understood without having felt the pain firsthand.

“I'm just ripping through it and I'm like, Supabase, open source Firebase alternative, right? And I'm like, okay, this is just— this feels like a great moment in time. And I'm talking to one of my partners. He's like, I love this idea. And Ben Tossel, who's like an incredible tinkerer, builder, investor out there, was an angel in the company. He made the connection and we ended up leading the seed round at Cotoo and did a Series A there as well.”

Get close to the metal: build the thing yourself to find the real pain points

Mazzeo's learning method: to understand why US video commerce hadn't worked, he built an actual Korean-beauty TikTok shop himself — sourcing wholesale, visiting Korean beauty shops, doing fulfillment by hand. Approaching a category out of curiosity is a win either way: maybe a business, at minimum firsthand knowledge of the pain points.

“Two, the only way to do it is get close to the metal of it. And if you just approach it out of curiosity, like it's just a win no matter what it— maybe it turns into an interesting business. And at the very least, you've like seen what the pain points there.”

Steal thisTo really understand a category, build a tiny version of it yourself end-to-end rather than studying it from the outside.

Taste won't save you — you'll get 'Lee Sedol'd' by consumer choice

Mazzeo's rebuttal to people who think taste is their AI-proof moat: like Lee Sedol losing to AlphaGo, taste-believers will face a moment where their taste is put against AI's in front of the consumer — and taste lives in the eye of the customer, who reliably picks the McDonald's burger over your fancy one.

“You think you're a tastemaker until you put like the burger in front of the person. Like, actually, I like this burger more. It's like, but that's a McDonald's burger. You're like, yeah, I like the McDonald's burger. Billions of people like a McDonald's burger more than your like, you know, Saint Charles— Fort Charles burger.”

The best operator productizes their 'dark gift' as an agent priced at tokens

Mazzeo's frame for his AI agent bets (Cognition/Devin, Augment): the world-class operator — like Harish Abbott on logistics from Amazon and Deliveroo — encodes their hard-won knowledge into an agentic platform, so any SMB can rent the best-in-the-world employee's brain on demand instead of hiring them.

“And so I think we're moving to this world where like, the best, the best at a task will be able to productize themselves as these agents and deliver.”

Steal thisIf you're elite at a narrow operational task, encode your playbook into an AI agent and sell it per-token to thousands of companies.

Any given Sunday: Mazzeo beat Peter Thiel for the Replit deal by genuinely loving the product

Competing against Peter Thiel for the Replit investment, Mazzeo won — he led the Series B — by authentically loving and obsessively using the product, sending founder Amjad every shower-thought product idea. The lesson: you won't win them all, but if you do the work and love it, on any given Sunday you can pull out the W against a heavyweight.

“And then I think there was a moment where like, where he was like, I can have a guy who's probably one of the best investors of all time, or I can have Matt. And Matt loves me. And, and, uh, and that was his choice.”

Buy railroads: in times of fastest change, own what won't change

David Bonderman's advice to Mazzeo when shown ChatGPT and Midjourney: you can chase the bleeding edge where everyone crowds in and software goes obsolete in six months, or you can 'buy railroads' — bet on old, durable industries that will be around forever and apply the new tools to them. Mazzeo split the difference and spent the next year applying AI to an old, dirty, stable industry.

“He goes, well, there's two ways to play it. One is this world is moving so fast that everything is changing. And it's like if you built software today or you built a product today, it feels like it's moving so fast that it could be obsolete in 6 months. And so that's shifting sand. You can play towards the bleeding edge, but that's where everyone is going to be playing because it's shiny and it's fun and it's distracting. Or you can buy a railroad”

Steal thisInstead of chasing the shiniest new tech, apply the new tools to a boring, durable industry that isn't going anywhere.