The Worst Problem: When Users Love Your Product Too Much

Most founders fear silence from users. But Amit Jain, CEO of Luma AI, has discovered something far worse: universal satisfaction. "The second worst scenario is you put something out and everybody is very happy with it, because that means there's nothing else left to do," he explains, describing the counterintuitive problem of building AI products that work too well.

In just two years, Jain has built Luma AI into a $200 million company that's redefining how we think about artificial intelligence. While others chase perfect language models, Luma is training AI that learns like the human brain—through video, audio, and text simultaneously. Their Dream Machine video generator became a household name overnight, appearing on CNN and Good Morning America after users discovered they could create Hollywood-quality scenes in 30 minutes.

From a physics PhD dropout who built iOS apps to an Apple Vision Pro engineer who saw the future in two breakthrough papers, Jain's journey reveals the art of building revolutionary AI products. In this interview, he shares the unorthodox strategies behind Luma's rapid rise: why they deliberately shipped "shit" products, how they used extreme pricing to discover their best customers, and why the future of AI isn't about models—it's about building worlds.

Amit Jain: Hey, my name is Ahmed. I'm one of the co-founders and CEO of Luma AI. At Luma, we are building multi-modal general intelligence, and that starts with building the world's best video models. We have so far raised $200 million from Andreessen Horowitz, Amplify Partners, Matrix Partners, Nvidia, AMD as well. Amazon, Luma is training models that are able to learn and generate video, audio, text all together. Our premise is that by building models that train like human brain, we will be able to not only generate worlds but actually solve the limitations of LLMs.

I grew up in India. As a child growing up, I was mostly very deeply interested in physics from very early on actually. I spent the majority of my time learning about advanced physics and trying to understand how the world works, honestly. I studied math and physics in college. I was about to go for a graduate school PhD in physics. Around the same time, I also started building iOS apps. My first one was actually for solving differential equations got actually relatively popular in the App Store. That was very interesting. Once you make something that other people find useful, it doesn't matter a large number or small number, that really kind of changes you.

Amit Jain: When a couple of my friends started a company, I left to join them instead. And as time has gone on, a few of my friends, the company was acquired by Apple. At Apple, I started working on this agent system called Shortcuts or workflows. When we shipped that, I learned that there was this part of the world where in Apple, where this new thing was being built. Some of my friends had gone there already and they got completely silent about what they were doing. It was this insane secret. If you don't want me to know anything, don't tell me it's a secret. I really wanted to figure out what the hell was going on. And it turns out they were building this new thing called the Vision Pro.

So I joined the team that was working on this really ambitious project for 3D capturing the world with Vision Pro. The idea was, if you're wearing this thing, we can take you anywhere because we have full control over what your eyes see. This was really close to my heart because anything that has to do anything with simulating reality, that really draws me very close. I started to work on that project and I worked on that for about 3 years.

Amit Jain: This was also the time when huge things were happening in language model land. In 2020, it really hit me in the head because two things came out. It was the Dali paper from OpenAI and this paper for 3D instruction called NERF, Neural Radiance Fields. And honestly, that was really interesting to me because all the things we were doing procedurally by writing handwritten code and for rendering and for capturing and reconstruction, if these two things work, you can learn to represent the world. You don't need to do any rendering, you don't need to write any graphics algorithms.

And what Dali proved was that from nothing, you can generate images. That was a huge thing for me. Nerf gave you static 3D worlds, and Dali gave you images. But our world is extremely dynamic. There's a lot of stuff happening, right? We walk around, cars move around, clouds move, all these kind of things. The question was, how can we actually simulate that world?

So, I started experimenting with these things. All I did was really experiment with these methods, write all kinds of networks, train them in about 3 months' time. I was extremely convinced that this is how most videos, images, things humans see will be made.

Amit Jain: At my work, I was doing all these traditional techniques, right? So my choices were stay at Apple, right? Try to convince this big giant organization that like, hey, this is the future, and that we should do this, and then I need $100 million to actually do this, or go find 15-20 of the most brilliant people on the planet who can actually do this for us. I tried this at Apple first, like, yeah, this is not gonna happen here. I left and then we started Luma in 2022.

Amit Jain: In 2023, we built a 3D generative model actually called Gini. At that time, large scale infrastructure for training and coders and things like that didn't exist. So we had to build and invent all these pieces, basically. So we started that work, we built a lot of large scale data collection systems, large scale training systems, all those kind of things. It took us about a year, a year and a half actually. We started the work on the first video model that Luma had released in 2024 called Dream Machine.

At that time, our chief scientist Xiao Ming had joined our team, right? So Xiao Ming was a leading image and video generation at Nvidia. At the time, these new graphics cards were coming about or these new training chips were coming about from Nvidia H100s, when we looked at that for the first time, we were like, OK, the cards are capable enough, the chips are capable enough. I think we can gather enough data, and the algorithms are there from our previous work on 3D generative models and things like that that we can actually do this.

Amit Jain: It was also at the heels of OpenAI Sora. So OpenAI had announced Sora in February, and that was actually really interesting because, before Sora, our video efforts were a little bit smaller because we're a very small company at the time. We were barely 20 people. We had a good amount of compute, but we didn't have very OpenAI level of compute, so we could not have scaled it without knowing that, oh, scaling could work. Once that evidence was in front of us, we just scaled our efforts significantly at that point.

Three months later, we had the first dream machine model, which was really funny. It was like the very first early model. Today, you wouldn't consider that to be very good, right? But at that time, that was huge and got so popular. It was on Good Morning America. It was on CNN and then people were absolutely astonished and mesmerized by it, like, oh, you can generate video. So, that made Luma into a very well-known household name at that time, and gave us all the resources we need to continue our research, to continue our work.

Amit Jain: Whenever you're developing a new capability into the world, when you're developing a new technology, oh man, it never worked the first try, it never worked the 10th try. I would really say, if you have found a good market and you want to find what fits into that market, iterate like hell. Anything that slows down your iteration velocity, avoid.

If you're an engineering mindset like myself, you wanna build a very stable system, right? Extensible system with all good technologies and all these kind of things, but sometimes those systems really slow you down because you build this monolith and it's good, it's very scalable, it's all the things, right? But the problem is to change one thing, you need to now change 5 modules. Compared to that, just a bare bones thing you built in Python, and you just put it up there, you have no allegiance to it. It's already shit, so you don't care, you just iterate, you change it all day, all night, right? And I think that's really, really important.

Amit Jain: Honestly, we thought not many people will try it because it's a really new thing. We know eventually the demand will be huge, but initially we thought not many people would try it, but man, the number of people that tried it was insane, so we put some pricing on it. It was only because I was like, OK, well, let's see how much people are willing to pay for it. So I had this extremely unscientific way, let's make it really expensive. And again, we'll give a small number of videos for free, and then it will be $30 and then $100 and then $500.

And the reason was solely to discover customer bases, discover who is willing to pay for it, because if someone is willing to pay $30 they are getting something out of it. If someone is willing to pay $100 they're clearly making money from it. And if someone is willing to pay $500 right, I need to go talk to them because I need to... why are you paying so much per month to use this product?

Amit Jain: So, we took the approach of putting something out there in very unpolished state. If someone found it valuable, we would know a lot, we would learn a lot, and we did. So this was one of the people who were paying like 100 or 500 bucks. I was curious, what are they doing with this thing. I got on the call, and this person was very happy, grinning ear to ear at the time, and they were like, oh, I'm really great to meet you. And I was like, is that you? They're well known, especially if you know the art directors and people in the world who make movies.

What I learned was that they were trying to create this scene in one of the more famous movies, right? They weren't able to do that with their traditional techniques. They tried Dream Machine and they got a really great scene out in 30 minutes. And now this person was talking to me because they were asking me to release the rights so that they could use it in the movie, and I was just blown away. First version of the model, it was not good enough to be used in movies generally, like really it wasn't. But here was someone who actually found even that useful.

Amit Jain: Early days, put it out, talk to users, see what they're doing, what they're not doing. Talk to them like, hey, how are you using our stuff? If you're not using our stuff, why are you not using our stuff, right? Do you know about our stuff? How did you find out? How did you not find out? We have a group of about 2000, 3000 of our most engaged users in Discord. Myself, but also our research team, our product team, our engineers, they're all in that chat all day.

When these are your most dedicated users, so I think paying users, people who are spending a lot of time on your platform. They have 1000 things to complain about, and that's very good. The worst thing that can happen to someone who's making anything in the world is apathy. You put something out and nobody cares. That's the worst scenario. The second worst scenario is you put something out and everybody is very happy with it. Because that means there's nothing else left to do. Generally, everybody's just really telling you all good things about it. That probably means they just wanna interview you or they're lying to you.

Amit Jain: Unlike traditional products, where you know, OK, this has feature A, B, and C, and this is how feature A interacts with feature B, like you know the whole state very well. With large models, people can do anything with it. People can generate anime, people can generate videos of a tomato rolling down a hill, with large models. You don't know what your users are gonna do, and there's no physical scenario in which you could test all the capabilities of the model. So you really have to see what people are doing with it, where it's succeeding, where it's failing.

Amit Jain: We are not a video model company. We are not building video models or image models, we are goal is very simply to solve multimodal general intelligence. What does that mean? Our belief is that people don't want image generation models or video generation models. What people want are world builders.

Every video, every movie is a world, a universe that someone created. Whether you're talking about high fantasy like Lord of the Rings, for instance, that's a whole universe with different laws of physics, with different characters, their personalities, all these things. Or if you think about a YouTube video or TikTok, they create a personality, a persona, all these kind of things. We need models that let people create worlds and then hit play. For that you need to build a very different kind of intelligence.

Amit Jain: To build that kind of intelligence, text alone is just not enough. Think about the way humans learn a concept. We see it with our eyes and video, we hear about it with our ears, and we reason about it logically in text. Everything humans learn from day in and day out doesn't just happen in text, it happens in all these modalities. So if you want to build intelligence that can collaborate with humans, digitally and physically, that can understand us, that can entertain us, you need to build intelligence that is trained on all the data a human brain is trained on, right? So video is a big part of that. So we believe that multi-modal intelligence or multimodal data actually is on the critical path to AGI.

Amit Jain: When you're working on a problem that is worth solving, that really motivates you. I find anything interesting, but finding something interesting and finding something you are just so mad about, that you wanna do it, right? It's very different. Passion could be a moment in time and it can come and go. Oh, this problem, it would be so great if you could do that and you can imagine that, oh I'm gonna spend all my life doing it. Generally, that's not the case.

My suggestion would be really try a lot of things. And try to go deep into them. And what you wanna find is not that you were interested in the depth of the problem, but whether that depth gets you more excited or less excited. When you go deep into it, you have to put effort. Sometimes when you're putting effort into it like, oh, this is as boring as it gets, right? Don't do that. If you found something, then it gets harder, you get more excited about that, right? That's a unique thing, honestly. I can guarantee you that most people around you will just quit.

Try a lot of things, try to go deep into them, and try to see, can you stay excited, not artificially, even after you're done thinking about that problem, right? But you can't stop thinking about that problem. Like it comes to you at night, it comes to you like, oh, but how do I do that, right? If you can find that, and somehow magically or luckily that happens to also be an opportunity in which a company can be built, that's it.