You've Never Lost $19 Billion. So Why Are You Playing It Safer Than Zuckerberg?

Mark Zuckerberg renamed his whole company for the metaverse, lost tens of billions of dollars, over $19 million in one year in 2025, by the way. Watched the whole world mock it and call it a flop. And he didn't respond by getting careful. He swung again, and now 10 million people have set up Muse in 6 weeks. Look, I've never lost $19 billion. You probably haven't either. And yet, we're still playing it safer than he is. You can think whatever you want about Mark the person. What I want to talk about is Mark the founder, and why AI rewards the people who swing big. If you strip out the money, there are three things behind that success that you and I can copy. Who they hired, how obsessive they got about the details nobody asked for, and the big one, the lesson that Mark Zuckerberg paid tens of billions of dollars to show he learned. After you miss, swing again. It doesn't matter how smart the models get at this point. Agency matters, and the story of Muse is the story of agency. When I covered Muse in September, I talked about the work it had done for me. I named mine Marvin after Douglas Adams, and it helped me cancel $1,285 in annual subscription spending, which, hey, thanks, Marvin. That was a real job that I had been putting off. I also talked about how approachable the product felt. Now we have a ton more people using it, and a much fuller account of how the team built Muse. And I think that account helps explain why this moment, this attempt, is reaching

people in a way the metaverse never did. To understand the size of the change here, please think back to what Zuckerberg has been trying to do for a long time. Meta reaches enormous numbers of people, billions, but its apps mostly live on devices and operating systems other companies control and manage. You can build one of the biggest businesses in the world, and you still can have Apple change something that affects how you reach your customers or make your money. Mark Zuckerberg wrote about that specific frustration explicitly in his 2024 letter on open source. He wants more control over the next generation of computing. And that ambition goes back at least to the Oculus acquisition in 2014, if not earlier. Back then, when Facebook bought Oculus as a virtual reality company, Mark was talking about the platforms that would follow mobile. Mobile was early then. And by 2021, he had renamed the company around the metaverse. Imagine how much conviction that takes. Your company's name, its investment priorities, billions and billions of dollars, a huge part of your public identity, all attached to a future you believe people are going to want. And all we remember is the meme of the digital Mark Zuckerberg that's heavily pixelated by, I think it was the Eiffel Tower. And most people don't ever follow you to the metaverse. They don't reorganize their social lives around

virtual worlds. They don't decide that this is how they want to spend their evenings. There are people who love virtual reality, sure, and there are some useful things we're going to get out of it, but the mass market platform Mark described just didn't arrive. Reality Labs, the division responsible for headsets, for glasses, for related research, lost 19.2 billion dollars in 2025 alone. And by the way, that is just one year's worth of losses. It gives you a sense of the cost of this search that Mark has been on, but it's not the whole bill. It's easy to say at this point that Meta should learn from its failures, that Mark made a big failure, and Mark should change directions and get more cautious and focus on Meta's core cash flow. At this point, changing direction, when you've invested this much, when you've named your company, it can feel like an admission that everybody was right. And you end up finding yourself defending the original vision long after you should have changed it. Or maybe you just try and hope everybody forgets about it, and you abandon it, and you leave it up. Mark Zuckerberg was not cautious. Mark Zuckerberg did not abandon his approach. Instead, Mark Zuckerberg took an already successful advertising business where he could have focused his attention, made money, and spent decades avoiding any other promise at scale, and been wildly successful by any measure.

No, he kept pursuing a platform. Meta's AI and hardware bets have overlapped, and the work on glasses is going to continue, but fundamentally, Muse is the bet that Mark is making that we will all enter the world he wants to build. Mark came back. Mark took another swing. Now, with Muse, you don't have to sign on to some grand vision that Mark has to get value out of the AI agent in your pocket, but you can show up to Muse as an ordinary consumer with a bill, with an account you need to sort out, with a question about your day, and you can use the phone you already own, and the relationship helps you get something done. Your relationship with your Muse. If the assistant helps, you're coming back, and that's what the team obsessed with. And that gives Muse a way to start connecting with you that draws you in to a deeper engagement over time. And I think that conversation, that point matters when we try to build things that are ambitious. We might see the whole future in our heads with all the pieces that work together, but the person we're trying to reach, they just see the first thing that we're asking them to do. And if that first thing doesn't make their life tangibly better, that larger vision you have never gets a chance to be heard, and it doesn't really matter, right? Muse matters because people can live their own lives with bigger ambitions with Muse. That's the only reason it matters. Mark's corporate strategy, nobody cares about other than, you know, nerds like me. And the fun thing about agency, which is one of the things we're talking about in

this video, is that when you exercise agency, you change, too. In his October interview with David Senra, Alexander Wang has described Mark approaching him after disappointment with Meta's AI progress, and asking what the company ought to do. Wang came over from Scale to lead the AI effort, and Nat Friedman became central to that product work. Alexander Wang says fewer than 200 people worked across the model and the product, and frankly, it's a small team that creates an incredible experience. Those people had Meta's enormous resources behind them, but they were small enough they could make sharp decisions, right? They kept insisting that the group be kept small by the standards of a big company like Meta, so that they didn't just add people to automatically resolve the hardest product decisions by sort of weight of consensus. They needed to stay sharp. Sometimes, if you have all the people you need, you don't make the focused decisions you need to make to make a product excellent, and instead you're doing awkward compromises.

Now, when you use an app, you and I can feel the compromises, right? Something in one tab behaves differently from the same thing in another tab. You have to learn a second set of rules to do a closely related task. Every individual feature might make sense, but using them all together takes way more effort than it ought to. That effort is now you, the user's problem, even though it came from consensus bureaucratic decisions inside a big company. Getting the right people matters, because someone has to make those decisions and take responsibility for the entire experience. And by the way, I'm talking about Mark by name a lot, but it's not just Mark that made this happen. Alexander Wang, a couple of hundred people in Menlo Park made this happen. And one of the things that we need to remember here is that Mark let them cook. Mark let them build an incredible product and got out of the way once he found the right people. And that's another piece of this, if you are a builder or a leader, is you got to find the right team, and then you got to let them cook. You got to let them exercise their agency to go run after this bet.

Mark's willingness to recruit a lot of different people, but not too many, give them important work, and spend behind that work is part of what I'm admiring here. He had to let the next attempt be shaped by people who could bring something his existing approach just didn't have. He admitted failure, and he took another swing. And that's a question I take with me now for disappointing projects. What would actually make my next attempt better? How do I wipe my brain of the failure and take another swing? The missing ingredient might be a deeper understanding of the customer or design judgment that I don't have, or maybe I need to stop splitting my work among different priorities. Because if I come back to the work and I'm not like Mark, and I just come back with the same set of assumptions and exactly the same way of working, and I'm willing to bet again, and I'm hoping enthusiasm fixes the problem, it's not going to fix the problem. Sometimes you have to change out who you're working with, sometimes you have to change out your approach, sometimes you have to change out how you think about the customer, but you got to make meaningful changes if you want that next bet to work. And then there's the bit of Muse that I find really, really interesting.

The care that went into the product. You might not think that care going into the product and polish is what matters in the age of MVPs and speed to market, but I think Muse is the most polished consumer AI app out there right now. And when I say polished, I'm actually talking about something that I think is very specific and tangible. There are places throughout that app where the team could have stopped at a perfectly functional product, but somebody clearly kept working. I love the way the avatar behaves. It's my favorite example of polish. But think about that more tangibly. Think about a hotel or a restaurant. You learn whether they care about detail by going to the bathroom. The entrance is supposed to impress you. Everybody knows the entrance matters. The fine dining room matters. But who takes care of the bathroom? Who takes care of the parts of the experience that don't feel as dramatic and public-facing? When you take care of that, you are showing your customer that everything is cared for. And that's the feeling I have when I look through Muse. If you get past the first conversation, if you get into other tabs, if you look at the parts of the product that did not have to be the opening shot of a launch video, there's a consistency there. It matters because a personal assistant needs to feel like they're a part of your day.

And you're going to encounter those dull edges repeatedly if you use that assistant. So, a detail that feels minor in a demo is going to become very irritating very quickly the longer you live with it. One example I described in September was how Marvin showed me that it was working. There was a little indication of what it was doing. Marvin would get out a little laptop and type or Marvin would scratch Marvin's head. I didn't have to interpret a technical status message. I just glanced at it and I understood. It's a tiny piece of design. It took a ton of work from the team and it answers a question that people have every time they delegate to AI, which is, "Hey, what's happening? I asked you to do something. Are you doing it?" So, if you're building an agent, that question follows you through the whole experience for the user. Did the user understand what's going on? Does the agent understand? Is the agent waiting on the user? Has the agent finished the task? Every time a person has to stop and think and investigate and put on their thinking hat, the magic disappears. They're supervising software instead of getting on with their day.

You know, Wayne described a list of more than a hundred specific model behaviors that the team tested and improved before launch. I find that detail a whole lot more revealing than a general claim that the model is a smart model. Frankly, I'm just going to be honest here, Muses not the smartest agent on the market. That's not the point. Getting back to the story, that detail, a hundred specific changes, says that the team was looking at what the product needed to do and working through the places where the underlying model was not doing good enough. A model can know a ton and still be a really annoying assistant if it doesn't understand when it needs you and when it does not. That same attention it reaches the character itself. The character has facial expressions, right? Meta actually worked on a model that generates the expressions and movements that feel like live conversational avatars and synchronizes them to speech. So, Meta's anticipating this live voice experience that they subsequently launched, and we don't have this in our hands yet, but where there's going to be like a Tamagotchi-like wearable where you can hold it up and talk to your Muse. They put that work in before the launch of their phone app so they would be ready to build a continuing relationship with the customer. But, a charming character that fails to do the job is still not good enough. Imagine you have a very useful assistant in theory, a smart model in theory with lots of

tool access, and it's smiling, but every interaction is confusing. Well, you still have a problem. People need help, and they need to be comfortable asking for it, and the experience of the character in the conversation and the work must fit together seamlessly so that they feel like they can approach this character in a way that feels natural. And I think that's especially relevant now because AI makes it so easy to produce something that looks really impressive at first pass, right? You can get an interface, a prototype, a presentation together super fast, and that can create this really weird temptation to keep making first passes. But, starting the next idea can sometimes be a temptation when you're really supposed to be fixing the awkward parts of the experience you're working on now. The speed is great. I love the speed. But, if you want people to keep using a thing, you eventually have to stick with one of those ideas through the less dramatic work and really make it your bet. Now, I don't want to tell you that Muse is already a platform.

Muse has a great early audience. 10 million people is great. It's not numbers for Mark, right? Mark is looking at billions. And so, Mark has to bring the cost to serve down at a much larger scale. Muse is expensive. Mark has to figure out how to turn Muse into something that is a natural network and works across assistants and businesses, which they've started to do. Mark needs to get to a hundred million and then a billion scale by scaling that utility geographically. Now, of course, Meta has advantages that most of us will never have, right? There's the huge existing business that can drive this, there's distribution, there's compute capacity, there's talent access. You and I both know that. I don't think that we should use that as an excuse to turn this into an uncomplicated story about inevitable destiny, as though everyone else was just never going to be able to compete and this was just destined to win. Instead, I think that we should examine what this kind of behavioral pattern can tell us about ourselves in this moment in AI. How timidly am I living my life?

How timidly are you living your life? And what would be different if we were a little bit bolder? And I think what makes this relevant to the rest of us is that our own circumstances are changing really fast. We all know that. That's why we're here. AI is reducing the cost of attempting a lot of the things that used to require a ton of money, that used to require a ton of specialized help, like life coaching. The question that we have to consider is how our ambitions change in the light of that and in the light of this moment, right? Imagine you've spent years working in an industry where customers struggle to compare complicated quotes and they can't tell what's included, what's missing. I've literally looked at these quotes. This is very personal to me. You want to build a tool that helps people work through that and you know which details matter because you've worked in that industry. Every time you thought about building it, though, you've pictured hiring developers, funding months of work. With AI, you may now be able to make that version yourself. You can just start with quotes and work through what a useful comparison would look like and get going. This is what I say so much on this channel, right?

You're not blocked. You can just get going. You can put together something a person can interact with and then check it against what you know about the work. And you might still need technical help to scale that system. In fact, that is a big block for a lot of live coders right now, but at the very least, you can show someone how it works, work through it a little bit, get some validation and traction etc. This changes how we think about ideas and ambition in the age of AI. Because instead of being stuck explaining an idea on a PowerPoint and asking whether it sounds good, we can just show someone. We can see if they like it. What that does is it lets us shorten the distance between ambition and something specific that we can bet on, which in turn allows us to scale our ambition and bets and keep learning. That's when the lessons from this whole Muse story, which was not an inevitable success, become useful at a scale you can afford. Just step back a minute. Before we go farther, I want you to replay what I just told you in the video for 10 seconds. You have been a failure in betting on a platform. You hire someone to lead your next bet that you're spending billions on who has never built a massive consumer product that is wildly successful at the billion scale, but you need this one to be that. You have a tiny team even though you could spend more. You don't have a winning model internally, and everyone thinks your models are terrible.

How does that add up to inevitable success? It doesn't. It's another bet. It was uncertain, but it was a bet that Mark needed to make to get where he needed to go. And it's a bet that he needed to make for his consumers. And this is something where we talk too much about corporate strategy. We obscure the fact that this only works because it works for consumers. It only works because it's genuinely useful for us. We obscure the passion for the customer that that team has, and I don't want to lose that because I think that's part of the reason the polish showed up. That's part of the secret sauce. And so, if we take the story of Muse as not an inevitable destiny by a billionaire, but a series of really bold bets, we start to reap lessons we can all learn. You might need a collaborator who understands the difficult bits better than you do. You might need your Alexander Wang, your Nat Friedman. You may need to dig deeper with customers. You may need a lesson around polish and how you take a bet and you don't just show an MVP, but you really invest in making that experience as polished as you can.

You may need another swing because you've tried something and it just didn't work. And you may be thinking I have to play cautiously because it's the age of AI. Don't think that. This is the opposite. The lesson of Muse is that you need feedback to win and you have to go out and build and put it out there to get there. Your time matters and getting to something that is really, really good. Getting the product market fit is so hard. Some ideas are going to need to be abandoned. In fact, let's be honest. Anyone who's built this stuff, most of them are going to need to be abandoned. But we ought to make those decisions with an understanding that we can go after another attempt a little more easily in the age of AI. And so if the reason you ruled something out was that the first version would take a team you couldn't afford or you can't scale it, come back. Have another look. See if you can be ambitious again. And what's interesting is that in this world, part of what holds us up is embarrassment. Embarrassment is not getting cheaper. It can still feel awful to show someone something that sucks. It can still feel awful to fail. It should.

And so the technical and financial barriers have come down, but the emotional barrier is exactly where it was. We may keep describing something as impractical when part of what we mean is that we really just don't want to take the risk because we don't want to fall on our face publicly. It's worth being honest about. I think a bunch of us have an old list on the desk of things we believe we can't do. A haunted graveyard, if you will. And some of those limits come from previous experience. Come from being embarrassed. Come from looking at the resources someone else had and deciding that their kind of work wasn't available to us. Look, I I have been there. I have founded before. I have run a business before. I have gone into a new product space before and it has failed. It's not just me talking theoretically. I literally have scars on my hands from welding a product that did not work. But getting it right, betting big, allows us to take advantage of this unique moment, this powerful technology that can make our existing lives dramatically different if we let it. And so, find something that you want to do that doesn't make your stomach drop the next time a new model comes out. Find something you want to do where you get more excited about building it when the next model comes out.

That's part of the aspiration behind Muse itself. Alexander Wang has talked about this idea of wanting to build in the direction of AI, wanting to build a personal AI that helps people pursue bigger and bigger ambitions throughout their lives over time. Ironically, it's a very meta point, right? Haha, pun intended. That remains a vision for what these products can enable. And so, when I think about this moment, and I think about the kinds of questions I tend to get from builders, from people early in their careers, from people senior in their careers, even from leaders, I come back to the question of agency a lot. Because I hear questions like, "It's too hard. It's too difficult. It's too expensive. It's too late to get started. I am behind. I have FOMO." We're too early to have FOMO, guys. I'm going to ask you how timidly you're living. I'm asking myself how timidly I'm living. I'm thinking about the next thing I'm willing to put real work into, even though I can't be sure it's going to work. I'm thinking about how I can take bigger swings, how I can go after something I care enough about to get other people to try it, tell me where it fails, help me make it better.

I don't want to look back on this moment and realize that I had the tools, I had an idea, I talked myself out of finding out what it could do. Whatever that is for you, let me know in the comments, and don't stop swinging.