Transcript: ‘How Hugging Face's Investor Spots Winners'

The transcript of The Every Podcast with Matt Hartman is below. Watch on X or YouTube, or listen on Spotify or Apple Podcasts.

Timestamps

Introduction: 00:00:00
How Hugging Face went from a teen chatbot to the home of open-source AI: 00:05:06
The three decisions that set Hugging Face on its path: 00:09:52
Hartman’s approach to finding the newly technically possible: 00:12:49
Why Hartman backs technical founders: 00:18:10
How Factorial’s portfolio founders source its next deals: 00:21:10
Why faster iteration makes product insight the most valuable skill: 00:27:09
What Hartman learned building Ghost Pepper, an open-source voice-to-text app: 00:34:28
Hartman’s current theses, from non-language models to full-stack AI apps: 00:40:02
Why the best founders right now are experimenting 00:44:14

Transcript

(00:00:00)

Natalia Quintero

Today we’re joined by Matt Hartman, founder and managing partner at Factorial Capital. Matt spent nearly a decade at Betaworks, where he was a partner, and while he was there, he wrote the first check into a tiny chatbot company called Hugging Face, which Nvidia bought last month for $13 billion. In 2023, he left to start his own fund, Factorial, a venture fund with a very unique structure, which we’ll learn about today. Anytime I talk to Matt, I think of him as a builder and a tinkerer. Every time we catch up, he’s built something really interesting for himself or his fund or his sons. So welcome, Matt. So excited to have you on the show.

Matt Hartman

Thank you for having me. While I was at Betaworks, we made all of our investments together, so it w

s very much a team effort finding and coming up with a thesis for Hugging Face. Happy to talk about what that is—or at least, since I’m not there anymore, what it was like at the time.

Natalia Quintero

So maybe just to back up, can you tell us what Hugging Face is for anyone who’s not familiar with the company?

Matt Hartman

Sure. It’s the biggest community for AI developers. Some people call it GitHub for AI. I think of it almost as a marketplace between AI researchers who are doing actual research—writing papers that then turn into code—and, on the other side, developers who want to use that research, like me. I’m not the one training a model, but I like to use brand-new models. And then they have private accounts where you can pay to do training and run inference. So that’s how they make money.

Natalia Quintero

But in 2015, when you found them, they weren’t that yet. I believe they were a chatbot for teenagers. Is that right?

Matt Hartman

Yeah. And I should say it’s probably the largest place for open-source AI in particular, so it’s very open-source focused. So what happened was my partners and I at Betaworks were really interested in what we were calling conversational software at the time. This paper had come out in late 2015 called word2vec that let you take words and put them into vectors. I built a neural network in college. It was a handwriting recognition thing. And so I knew that you had 100 percent of the information in an image. That’s not true with words, right? If I say “king,” it might be the king of spades. It might be the king of England. It might be LeBron James. We don’t know which king I’m talking about. What word2vec did was propose a way to embed words within their context.

They would map these into what they call n-dimensional space, but picture 3D space. You would have “king” over here and “queen” over here, let’s say, and the vector points from king to queen. The craziest thing—I thought it was unbelievable—was when you moved the arrow over to where “prince” was, it pointed to “princess.” Same thing with cities. You would have Paris, and it would point to France, and you’d move it over to Rome and it would point to Italy. What that meant was that you could now take words and turn them into unique numbers that represented their meaning. And once you could do that, you could put them into neural networks.

So we had been tracking this stuff and were trying to figure out what kinds of products could be built. There were a bunch of companies that were, in my opinion, pivoting to be chatbot companies because VCs were talking about chatbots, and those were genuine new distribution channels. But what we decided to do was go earlier and find people who were rowing in the same direction, who had these same observations. The Hugging Face team had this observation that there’s something interesting to build here. What could you build now that this existed? At the time there was the Stanford question-answering dataset, SQuAD, and basic question answering was starting to become the core thing. So they said, what if you could have a chat with this other kind of intelligence that’s getting built? Maybe you talk about movies, maybe you talk about what’s going on this weekend. That was the core product they built initially. They were based in Paris, moved to New York, and worked out of our office, so I got to know them really well.

And then, in my view—I haven’t heard them talk about this much, so this is from my perspective—they had a consumer app that a lot of people were using, but the retention really wasn’t good. They were very honest with themselves about those metrics. When the Transformers paper came out, they quickly converted it into PyTorch. I think it came out in TensorFlow, and it needed to be in PyTorch for people doing Python jobs. They open-sourced that, and it went viral on GitHub. My recollection is that other people started reaching out to them and saying, “Hey, can you convert our paper?” Or they converted their own papers. And then they moved that over to the Hugging Face Hub.

They felt the demand from AI researchers who were trying to figure this stuff out. They themselves were developers and had implemented so much of this that they knew what was hard. So rather than just build a consumer app, maybe they could make it easier for anybody to use one of these open-source models and build their own products. That’s ultimately where the product and the business ended up.

(00:10:00)

Natalia Quintero

It’s just fascinating. When we were chatting in preparation for this podcast, you mentioned that Hugging Face and the team made a thousand great decisions. What would you say were maybe three pivotal decisions that led them to where they are today?

Matt Hartman

I think the first was being honest about the product. A lot of teams today go out and raise on whatever top-line metric they possibly can. I think the best founders—and this was true of Anchor, too—look at their retention and are really honest with themselves about what is genuinely working and what is genuinely not, over the long term, in terms of creating value.

So I think one thing they did was open source and Transformers. The second thing was moving everything off of GitHub—because it was on GitHub—and having a richer place for the weights to live. It was a more structured place for AI researchers to go.

And then another decision that comes to mind—I don’t know if it’s the most important—seems so obvious now, but I remember it being a question: whether you move outside of language models or not. Now we think about video models and robotics models and bio models and weather models. But language was really the first use case. And I remember wondering if everyone would end up using this for everything. What do they want to hold constant? Open source was always held constant, and then they were flexible on what kinds of models were trained.

Natalia Quintero

One of the things that’s really interesting to me about how you ended up as the first investor in Hugging Face is that it stemmed from a thesis you had, or a theme you were exploring. Maybe just to tug on this thread for a minute, I’d be curious to learn how you think about the theses or themes you’re exploring at any given time. You mentioned the one that led you to Hugging Face. How is that evolving over time?

Matt Hartman

Yeah, my goal is to find the newly technically possible. Sometimes it’s just tinkering with what’s available. Sometimes it’s reading the paper, as in that case. Often it’s somebody I think is really smart exploring something that surprises me. When I started Factorial Capital, one of the things I was having a conversation about with a number of people was whether large language models were working, and it was not obvious that was going to be the case. Does that apply only to language? Does it apply to totally other kinds of models, like large weather models or physics models? There are lots of different technologies people use now, but the underlying technology of transformers is really good at next-token prediction, or predicting the next thing in a time series, and that would be really useful.

So I started getting really interested in how this applies to robotics, to science, to all these other areas. That’s when I start to pull on the thread, and then I just ask as many smart people questions as possible. And often the thesis ends up 20 degrees off. I certainly don’t think I predicted what Hugging Face would become from what they started with. Or Anchor, which became the most popular tool to create podcasts. It was originally an audio social network.

With Anchor, I was getting a cup of coffee in 2014. My partners at Betaworks came at it from a different angle, but my angle was this: I was waiting for a cup of coffee, and I looked at my phone, and all the apps were competing with each other. Do I text? Do I send an email? Do I use Snapchat? Then I got my coffee, and I was walking back to the office and put my headphones in. And there was no competition for my ears. I had audiobooks a little bit. You could make a phone call, you could listen to music. There were podcasts, but they weren’t huge.

My thought was, maybe there can be something like a consumer Twitter-for-audio product. I looked for a year to try to find this. I would talk to anybody who would listen: “Have you seen things like this?” I would experiment with building things, and I would listen to a lot of podcasts. So my approach was seeing that gap, feeling that gap. When I met the team at Anchor, I’d met a bunch of other teams doing this, and in my view they didn’t quite have the right approach to the product. The apps didn’t know they were on a phone. And I remember the Anchor team said—I don’t know if they remember this, but I do—“There’s a bone in the back of your neck that amplifies the bass in your own head. That’s why you don’t like what you sound like when you listen to a recording. You don’t have that bass.” They were like, that’s psychological friction on why you wouldn’t post your own voice.

I think a team that is that deeply focused on why people do things is going to ask the right questions, even if it takes them somewhere unexpected. Ultimately, they had a very similar situation to Hugging Face. Their app went viral, but then the usage was okay. But if you looked at just the people who were posting, not the listeners, it was really good. At the time, Giphy was popular, and they were like, what if we were more like Giphy, where you don’t have to use Giphy to see GIFs? It’s available everywhere. Anchor would help you get your podcast out there. At the time, Alexa had podcasts, and you could post to Alexa. They really stripped out the demand side, focused on the supply side, and built a really, really good product.

So that’s the way I typically come to my theses: try as many things as possible. “What is newly programmable?” is another way to think about it. Bitcoin certainly is programmable money. Websites were programmable; they were little pieces of software. And now I think many of the teams doing a great job in robotics, which you can think about as a hardware thing, or in scientific models, which you can think about as research-heavy—the teams that are really doing well are ones with software development experience, thinking about them as a software problem, not necessarily a hardware problem.

Someone said a Tesla is an iPhone on wheels. I think that feels true. You can do an upgrade on a Tesla and it gets better over time. It’s really a software product as much as a hardware product. So I think much more today is becoming programmable. That’s the other way I approach it: What are the new tools that help us make new things?

Natalia Quintero

Yeah, it sounds like there’s a real commitment to the scientific method—finding where the edge of science is and who is thinking about it really critically. It reminds me of how in 2025 you told TechCrunch that your mantra for investing is that to invest in software, you need to know how software works. I’m curious how you came to that particular mantra.

Matt Hartman

I would say that’s true for me. There are obviously people who don’t code who are amazing investors, so there are probably multiple ways to do this. For me, I’m interested in things that are newly possible because of technology. And I think understanding what the stacks look like—understanding the technology itself—matters, even in an era of vibe coding where theoretically anybody can build anything. The questions you ask become more important. The question “How can I make this thing?” is different from “Hey, I’m seeing all these local AI models. They run on different kinds of computers. What’s the best way to optimize this for mine?” That’s more the question a technologist would ask, someone who understands the limitations of a local model and understands there are RAM constraints. Those are the kinds of products that resonate with me more.

And again, this is just my approach. There’s a very good way to back startups that is market-driven: you back executives who understand the industry. But I think a lot about the pivot foot in basketball. If you move both feet, it’s traveling. Large companies can just totally build something new, and that’s fine. But at a very early stage, I always assume the company is going to pivot. And the question is, are they going to pivot toward something that I, as a fellow traveler, also believe in, or are they going to have limitations on what the pivot foot could be?

(00:20:00)

If you backed a bunch of real estate executives to build an AI-native thing for real estate, they’re going to be fairly constrained. It’s more likely they’re going to pivot to something that’s still in that same market. Compare that to a team who might say, “The reason we’re going after this is because there’s a particular shape of the data in this industry.” Maybe it didn’t work, but it turns out that shape is right for the car industry or wherever, and they build a product that’s applicable to multiple industries. It becomes a horizontal product because they see the similarities across categories.

The companies I’ve been involved with that have been most successful are ones where the pivot foot was about the technology, not necessarily about market understanding of a particular vertical. So for me, backing technical founders gives them the ability to move with where the world goes.

Natalia Quintero

This is also so represented in the way you’ve structured your own venture fund and the partners you have as part of it. Can you describe a little of how the fund works? You have 12 investment partners, but you’re a pre-seed and seed fund. That is unusual. How does that work?

Matt Hartman

I had this idea in 2023-ish, leading up to starting Factorial Capital. I wanted to be involved with the technology that was changing. Not tech-enabled things like a brand sold online, but the actual technology itself. I was pretty technical, but it was moving so quickly. So my thought was, what if you set up a fund with a structure where you had venture partners who could help me really quickly and accurately? The goal is to come to very high-quality decisions very quickly at the early stage. Can they help me get really smart on where the technology is going?

So Clem, the CEO of Hugging Face, is thinking about what’s going on in the open-source world. Or Iqram, who cofounded Venmo, has spent a lot of time on developer tools, really understands this stuff, and is building new things all the time. It’s great for finding new companies, and it’s great for thought partnership on where things are going. So I set up the fund so that I could have an almost infinite number of venture partners. I set it up as a solo GP fund with venture partners. Instead of partners, there’s one person, but it’s almost like I’m partnering with every technical founder I’ve ever met.

At the beginning, that was really, really helpful. Over time, I thought maybe I would have cohorts of venture partners. What I found was that having a few very deep relationships was really good. Everyone who was a venture partner was somebody I had backed or who was part of the portfolio at some point, someone I knew really well.

Over time, what I found was that the portfolio of Factorial Capital itself became the referrals. I still rely on a small set of people for finding really interesting, new, crazy things and for helping me think through other companies. But as an example, Clem from Hugging Face introduced me to the Nomic team. They were an amazing team. We invested in them, and then they introduced me to Patronus. Patronus is a really interesting company. We did their seed. Then I got really close to them, and they raised a Series A, and they’re off to the races. Then they introduced me to another founder, of a company called Causal Labs. So we end up with—I don’t know whether it’s a flywheel or a family tree—a deep network of technical people. And the technology always moves forward, so there are always new people to get to know, work with, and learn from.

Natalia Quintero

Yeah, it sounds like the dream flywheel, right? You find a once-in-a-lifetime founder in someone like Clem, the founder of Hugging Face. Then, via him and other brilliant technical founders, you create a network of the most brilliant technical founders in their domains. They’re directly familiar with each other because they’re in the same ecosystem, or direct competitors, or building compelling products that they’re tracking and that inspire the way they think about the evolution of technology. And then they’re your partners, right? Clem is your partner. You also have one of the cofounders of Venmo, and Hilary Mason from Fast Forward Labs. You have this incredible set of partners helping you think about the future of technology. And the generations of technology are getting shorter and shorter every year, it seems. So the next generation of technologists becomes the next generation of your partners. It’s such a cool, very unique flywheel that you’ve built.

Matt Hartman

Yeah, it’s definitely fun to work with people and watch them build pretty amazing things. You have so little information when you start. Ideally, we’re the very first check into a company, so any other information you can get matters. Hugging Face started out of the Betaworks office. I was there. And when I started Factorial Capital, I started out of the Hugging Face office. So I was sitting across from the person who had built one of the large language models at Hugging Face, a super early employee. When he started a company, we were able to back him. There was a guy who was a big part of the robotics team. He started a company, and we backed him. The CTO of Factory was at Hugging Face.

And then there are the other founders. I work very closely with Clem. There are other people, like Thomas Wolf and Julien, who are so smart and interested in seeing other people succeed, and they’re often investing in and mentoring founders. So ideally you fan out from one core hub to meet more and more people, and you keep one thing consistent: very good technical people who want to learn, or who are doing it because they want to see something amazing built.

Natalia Quintero

So we’re hosting a conference in a few weeks, Thesis, and it’s about the future of work and the future of technology, really. Based on what you’ve learned as the first investor in Hugging Face and many other incredible companies, I’m curious what your thesis is today. We were talking before the pod about this discourse where some people say that in a future of AI, it makes sense to be more technical, and others say less technical. But you had a really interesting take, specifically for founders building a company, on what the core skill might be. Could you share a little about that?

Matt Hartman

Sure. And I want to add: The Hugging Face team has built something amazing, but there are so many other companies built by technical founders, like Factory and Modal, that are fascinating to me because the speed at which they’re growing is so fast. I’m thinking about why the speed is so quick. One possible thesis I’ve been reflecting on: If coding is becoming faster because we can use AI to help us code, then, as has been said—and I agree—what you build matters more. The job of a product manager or product person is really to see the product in the wild and intuit from user behavior, whether that’s the data or sitting looking over someone’s shoulder at how they’re using it, so that you can iterate.

If iteration is the most important part of finding product-market fit, and probably also of growing, then think about it this way. If it used to take two months to see the feedback, get the insight, put together the product plan, code it, and get it back in front of the user, it might take two days now. So the product person, whose job is to understand the user and translate that into the next version—whether by coding or by partnering with someone—is very valuable whether they code or not. And the iteration cycles are necessarily so much faster.

It’s fascinating to see that a company like Hugging Face took 10 years to get to the point it’s at. Then you have a company like Modal, which existed a little before I started Factorial, but not much, and Factory AI, which started at probably the same time. It’s only been a few years, and these companies have grown so quickly. I wonder if part of it is that the ability to build and iterate allows these very, very fast cycles.

(00:30:00)

So the result, in my mind, is that the ability to generate product insights—forget about what the job is called; it could be called an engineer, it could be called a product manager—the ability to synthesize how a user is using your product and then improve it has got to be exponentially more valuable. My guess is it was already the most valuable thing, and now it’s exponentially the most valuable thing. Obviously, distribution matters more than ever today as well. There’s a whole bunch of noise, and the easier it is to build, the more noise there is. But I suspect that’s a job whose value is going up tremendously right now, because of the speed of—

Natalia Quintero

Iteration, as the core variable to deliver speed.

Matt Hartman

Your cycle looks like this, and you compress it so your cycle time is like this. So you just grow much faster. Your work gets compounded as a function of time.

Natalia Quintero

I’m curious, how do you filter for that when you’re talking to a founder? When you’re meeting a compelling technical founder working on an interesting problem, with a pivot foot, how do you measure whether they can iterate quickly?

Matt Hartman

Usually when I’m meeting them, they don’t think they’re pivoting, right? They’re like, “No, this is definitely the thing. I’m raising money for it.” But I think that’s where the ability to iterate is probably more important than the thing you’re actually building. So usually I’m trying to understand what their insight is.

People have talked about earned insights, and I really like that framing. The way I think about it is: How many people were in a position to learn the thing that you did and then execute on it? I think that’s what “earned” means. If there’s some observation that we all have, there’s going to be more effort put toward solving that problem, which means more competition. If only 10 people in the world have come across this, because it’s a particular category or area of expertise, you’re probably going to have more time, because you’ll be ahead. You’ve also already spent a lot of time learning that problem, and you can go after it directly.

So I think an earned insight is important. And usually that goes hand in hand with the ability to generate new insights as you’re in market, which is another reason to have your product out there. There are people who can just divine what the right thing is. There’s a story—I don’t know if it’s true—that Figma was building alone for a year or two. That’s not the pattern I’ve seen. It’s more common that you put something out there, it’s pretty cool, something starts to take off, and you delete everything else and do that thing. Because you’ve been able to generate insights to get that first version, you know how to do it going forward. So often it comes down to the why. Why is this interesting? Why is this the problem to solve?

Maybe Anchor is a good example. Somebody who’s thinking that deeply about how your voice sounds in your own head is probably a person who’s going to really, really sit with their users and understand what problem they’re solving.

Natalia Quintero

And this is Mike Mignano. The best. So, great shout-out to Mike.

Matt Hartman

Yeah. I don’t remember which one of them said it. I’ve talked about it since—I don’t know if they remember saying it—but I remember thinking, wow. One thing they did that I’m positive about is that the onboarding was different depending on whether you had headphones in or not. Think about whether your volume is on. It’s different now, but pre-AirPods, if you didn’t have headphones in, you’d shut down an app if it just started playing audio. So how can you lower the friction when someone’s signing up, letting them do it with or without their headphones in? It was also a very cute way they showed the headphones. That kind of care pays dividends, certainly in a consumer product.

Natalia Quintero

Yeah. Something I often think about with really great companies is that they end up being the greatest expression of someone’s best self. And I think about this with you and your fund, because you have this mantra that to invest in software, you need to know how software works. And you are the biggest tinkerer I know. Every time we catch up, you’re working on something wild or whimsical, or you’re building software for your own fund. Can you speak a little to some of the things you’ve built and are using?

Matt Hartman

Sure. Becca, who works with me at Factorial, is also a builder. So she and I often build the mediocre version, and she makes it a really good version. There are things we use to manage the fund properly. But I’m also building to test and learn. A product I built that got a little bit of attention earlier this year was called Ghost Pepper.

There are many of these apps where you hold down a hotkey and speak. A lot of people use them to speak their prompts. I was using one, and I was thinking, I’m seeing all these local AI models for voice-to-text. Why do you need to pay for inference for this thing? Couldn’t you just do it on your own machine? So I went to Claude Code, I think, and said, “Look at the Hugging Face repos. Here are a couple that I know about, but find whatever you can. I want to build something like this.” I used this skill, which I love, called Superpowers. I’d recommend Superpowers to any product developer. I think it’s available for both Codex and Claude. I was using it in Claude. It’s by this guy Jesse—sorry, Jesse Vincent—who is a genius prompt engineer. The difference between building things with it and without it is night and day, in my opinion.

So I used Superpowers and had it do a bunch of research with me on voice models. I showed it what kind of computer I had, and then I built this voice-to-text tool that I called Ghost Pepper and put it on Hacker News. It ended up going to number one on Hacker News. I had never posted on Hacker News before, so I was really excited about that. I open-sourced it. You don’t have any inference costs, so why not let anybody try this and run it for free?

It was an interesting lesson. I learned that you could build a very high-quality product. A lot of people shared that it was actually faster than a bunch of state-of-the-art products, probably because it’s constantly looking for the newest, fastest models, and local voice-to-text works really well. But the other thing I learned was that when you vibe code something for yourself, you have to maintain it. There’s all this talk about the SaaS apocalypse. These paid apps aren’t that expensive, and in a way, what you’re paying for is them making sure it stays a great product. The job to be done for SaaS apps, in a way, is to always continue to iterate and be at the edge. I now have a different appreciation for the SaaS products that really do release all the time. So it’s definitely a fun project, and I think there are thousands of people using it.

Natalia Quintero

Yeah, I think I saw it has over 3,000 GitHub stars. That’s so cool. I ran this by some of our engineers, and the maintenance piece is so real. One of my earliest learnings of this was when I vibe coded our CRM tool and used it for months. Then at some point I realized I was spending too much time maintaining the tool I had built, when instead I could be spending time with our clients, working on AI implementation and the work we do on the consulting side of the business. I realized it made more sense for us to migrate the data to Attio, a product that just does CRM and data management, instead of me trying to play engineer on the side and maintain this tool.

So maintenance is a really, really big part of it. I know this is something our engineers think a lot about. Monologue is one of these voice-to-text tools that we built in-house, by one of our great engineers, Naveen, and he mentioned that this is certainly one of the competitive edges. With Monologue, for example, latency is something they think a lot about. I think it’s around 800 milliseconds of latency for it to load. So it’s certainly a trend that we’re following. So cool that you built Ghost Pepper. We’ll also include Jesse’s skill as a follow-up to the pod, as a resource for anyone listening.

The other thing we were talking about is the thesis you have now. You had a thesis and some themes you were exploring in 2015 that led to investing in Hugging Face. What are the themes or theses you’re exploring today?

(00:40:00)

Matt Hartman

I think so much is still infrastructure. It’s amazing to me—developer tools, for example. You just gave us a perfect one with maintenance. I think there are going to be products that help you maintain all of this code. I’ve seen a few. So there’s the continuation of developer tools. Different people are becoming developers, which probably means there are different kinds of tools.

I continue to think about frontier non-language models. As I said earlier, transformers are really good at next-token prediction, and it turns out that LLMs have gotten much, much better with more compute. If that’s the case, what are the other areas where you could predict time series and create genuinely new value? An obvious example is robots. What is a robotic movement if not time-series information with the locations of a bunch of servos? People are already doing this. We invested in a couple of companies with the general thesis that the way we train robots could change. People talk about whether there will be a ChatGPT moment. The data is a bit different, so we’ll see how it works. But I think that’s a category where transformers, and AI more generally, seem to be really applicable.

I’m fascinated by the science models. Hugging Face has a hackathon they’re doing called Rare Disease, Real Kid, where they’ve open-sourced the genome and, I think, the doctors’ notes of a kid with a very rare disease. I think only 50 people have it. They have a track for diagnosing it, but also a track for drug repurposing. People talk a lot about drug development, which is interesting, but drug repurposing is another one that’s fascinating. And it’s interesting to me that the people building this stuff are really technologists. As I said before, it’s becoming programmable.

And then there’s the application layer, where everyone talks about vertical applications and consumer applications. You said something earlier about latency. I believe that even if applications look like consumer applications or verticals, the people who build them are going to be technical enough to go fully down the stack. The example I keep thinking about is the Instagram era. There were a lot of photo-sharing apps, and Instagram did a bunch of things right. If you looked at the phone, it was very obvious that you had a camera and a cell connection. What was not obvious is that load time mattered. If you’re standing, waiting for something, and it takes 30 seconds for the app to load, you’re done. So Instagram spent time really making load time a priority.

I have read—and I assume this is true—that a side effect of the pictures being square, which I don’t think was on purpose, is that you don’t have to ask the server whether to show the image in portrait or landscape. And that shaves off a millisecond of load time. A team that has the ability to do great design with that nostalgic feel, as early Instagram did, and also to go all the way down the stack to shave seconds off server time, leveraging a decision they made on the design side—I think there are a lot of parallels to what AI-native companies are going to look like in the future. They’ll look at a paper about an interesting new way to train reasoning for language and ask, “What can we pull from that and apply to images, or to what we’re doing?”

They’ll have both the ability to build their own models and the humility to know that these big-company models are getting great. If we can just use them for the stuff we need them for, we can spend all of our time on the things that are more unique to us. Being able to balance that and go full stack looks different in an AI-native world, but being able to do it is going to be a real challenge.

Natalia Quintero

As we think about the future, along the lines of the things you’re excited about and the companies you’re tracking—as an investor in so many compelling companies, what advice would you have for founders trying to navigate this really dynamic environment where the technology is changing so quickly? What are you seeing really great founders doing right now?

Matt Hartman

The best people I have seen building things are playing. We are in an era that reminds me of the early internet. Everyone is playing. There are so many fun tools. I tried this text-to-CAD thing. I don’t have a 3D printer, but I was like, I bet you could vibe—how good of a job could I do making a robot? We’re not there yet. I don’t think you can 3D print a robot. You can make some 3D structures, but we probably will be able to. I think the way you get the best ideas is by playing and tinkering with this stuff, seeing where the problems are, and then solving them.

Pre-2016, before the AI era, it was really the mobile era, and there were a lot of fun things to play with. Then there was this question-mark period where the phone was at scale, lots of people knew how to make iPhone apps, distribution was getting harder, and AI hadn’t quite become what it is today. I thought AI was going to be at least as big as mobile. I think it will turn out to be bigger than the internet—and to me, mobile is not quite as big as the internet.

So again, the kinds of companies I’m most excited about are playing with this stuff, finding what’s newly possible, seeing where the gaps are, and then solving problems, whether it’s a problem in your life or a friend’s company or whatever it is. That’s how I would be spending my time. There are maybe the same number of problems as always, probably more, but there are definitely more solutions to existing problems than there have ever been. And the solution space is so untouched relative to all the companies being built, and that I think will be built.

I used to play—I think it was World of Warcraft or something—and you would move around and the whole world was dark. As you moved around, you got to see the world. I think we’re in that right now. Even the people building these AI models are still exploring what they can do, and we’re all exploring what’s possible. Our focus certainly is: What are the solutions that are newly possible? Maybe they’re for existing problems, maybe they’re for brand-new problems. I think there are numerous ones. So it’s an amazing time to be exploring that and building new products.

Natalia Quintero

It really is so fun. Talking about games reminds me of one of the companies that spun out of Every, led by Alex Duffy, who literally built a gaming company because he was testing the capabilities of models playing different games against each other. It really does come from this spirit of exploring what’s possible and then seeing how far you can take it.

Matt Hartman

The best way to explore what’s possible is to listen to Every so that you know what’s new. There’s never been a time when there’s more newly possible stuff every week. So that’s my personal recommendation.

Natalia Quintero

Great. You’re on the team now, Matt. Welcome. All right, maybe just to close, Matt, any resources you’re really excited about? People you’re following? You already shared Jesse Vincent’s Superpowers skill, and we’ll link that below. Anything else you’re looking at day to day that’s really helping you stay up to speed with everything that’s going on?

Matt Hartman

There’s so much that’s shared on X. There’s a lot of noise, but a lot of it is really good. I look at what’s trending on Hugging Face. I’m fascinated by open source, so I try to find any resources on open source. There are newsletters I like. I think Ben’s Bites is great. I think Every’s newsletter is great. One of the reasons I’ve been reading X more and more again is that people are writing these longer-form pieces. It seems like Medium content moved over to X. And you can take this stuff and put it into Claude or ChatGPT and ask, “How can I apply this to my life, or to my investment thinking, or to my company?” How can I take this insight about model memory and incorporate it? It’s fascinating.

So I probably have a barbell approach. I have what’s new, new, new. And then—I forget what it’s called, the rule where the longer something has been around, the longer it’ll be around for. So I’m reading things that are older. I’ve been reading a book about the fall of the Roman Empire, and things that are hopefully timeless. It feels like we’re in a moment where certain things are repeating themselves, whether it’s technology or culture or human behavior.

(00:50:00)

I find it useful to look at some of the older stuff, too. I know we’ve talked about some of this as well. Not only spending time looking at the most recent technology can often end up being informative.

Natalia Quintero

Thank you, Matt. It’s so fun to have you. You’re one of my favorite thinkers and investors. Where can people find you?

Matt Hartman

I’m Matt Hartman on X, and you can email me at matt@factorial.com. Thank you so much for having me. It’s been fun.

Natalia Quintero

Such a treat. All right. Thanks, Matt.

Natalia Quintero is the head of consulting at Every. To read more essays like this, subscribe to Every, and follow us on X at @every and on LinkedIn.

Everyone’s a builder now. Every All Access gets you the full membership plus the Builder Pack—$9,000+ in credits for the tools we build with.