Joshua Xu, CEO and Co-founder of HeyGen

MRSD, entrepreneurship, and building a foundation that spans technology and business

Interviewer: Thank you for taking the time to chat with us. We are working on the MRSD Newsletter, and this is going to be the third edition. We want to highlight alumni who have been successful in their fields, especially alumni who have chosen entrepreneurship paths. We wanted to talk to you specifically about HeyGen. To kick things off, you graduated from MRSD about 10 years ago and you are well placed to answer this question: how do you think MRSD helped you in your career after you left? You first worked at Snap, and then you went on to found HeyGen. What role do you think MRSD played, specifically in helping you think about entrepreneurship topics like problem market fit, building an early team, and finding co-founders or early team members?

Joshua Xu: One thing that is very unique about the MRSD program is that it does not only focus on technology, it also focuses on the business. The program builds a solid foundation. During the almost two years at CMU, I learned a ton. There were a lot of challenging projects where you spend days and nights working on them, and that was very fun. Those are some of my favorite memories when I look back over the past ten years.

What was also very important is that you get exposed to business concepts like building an MVP, building a prototype, and getting feedback from people. It is not only feedback from technical people, it is also feedback from people who have been running businesses and people who have been in the industry. It is hard to grasp how to run a business in a short amount of time, but you start to appreciate that there are different aspects to how people think about problems. Just hearing how Hagen thinks about problems and the questions he asks, I learned a lot from that. In summary, it is the mix of technology and business that becomes a strong foundation after school.

The early days of HeyGen, and why the idea started as “replacing the camera”

Interviewer: I would love to talk about your early days at HeyGen, because you started the company in 2020, which was early even in the broader domain of video generation. I also want to understand your early thought process, because at that time you were coming from a very strong technical background and you were operating in a space where the tools and the market were both still forming. What was the process like in terms of getting the company started, talking to early team members, thinking about a co-founder, and defining what the company would actually do?

Joshua Xu: The founding story goes back to 2018 and 2019, when GANs, Generative Adversarial Networks, started to emerge. We saw things like CycleGAN, and you could transform images in ways that felt mind-blowing at the time, like turning a face into a baby face or a Disney-style face. I stayed very close to that technology and I felt very excited about it.

What is funny is that I studied robotics and specialized in computer vision at CMU, but for my first four or five years I did not work on computer vision. I was working more on recommendation systems. When I came back to computer vision, I was really passionate about it because it reminded me of school.

I am personally an introvert. When I was working on camera-related work, it was challenging and awkward to test my own software because you need to take selfies or record videos. That gave me the idea: what if AI can generate video and generate content? That would be a huge unlock. I have seen many users struggle performing in front of a camera. People have fear and anxiety when they are on camera.

That was the core idea: use AI to generate video. This was early, before the big generative AI moment. I believed the technology would continue to evolve. I think our generation grew up in the mobile internet, and we live in what I would call a camera-first world. If you want attention and audience, you have to be on camera. That is not fair to introverts. The world rewards people who perform well on camera, not necessarily the best ideas. Many people have great ideas but do not have a way to share them. We wanted to replace the camera.

We founded HeyGen in December 2020. My co-founder and I went to school together. He was in HCI at CMU. We brainstormed for around six to nine months. It felt natural: I worked on tech and he worked on design, and we built from there.

Building in a fast-moving field, and why quality stays the top priority as technology evolves

Interviewer: You started in the GAN era, and since then the technology has moved rapidly. In 2022, diffusion models and tools like Stable Diffusion became mainstream, and the broader generative ecosystem began accelerating quickly. When you are operating in a space where the underlying capabilities change so quickly, the company can easily get pulled in many directions. How do you think about staying grounded internally? How has the direction of the company and the mindset changed as the technology evolved, and how has that affected your priorities and criteria for getting new team members?

Joshua Xu: Over the past five years, technology has only accelerated and evolved. Our north star has been consistent: replacing the camera for people who are not comfortable on camera. Quality is always number one. You need to produce something at the same quality as a camera studio, and actually it needs to be better, because the user is not comfortable performing on camera.

Technology evolution creates opportunities to improve quality and innovate. One key piece is that we built the entire pipeline in-house: training, data collection, and evaluation. That allows us to iterate faster and improve quality faster.

In terms of company building, I think of it in two phases. Pre product market fit, we were about 10 to 20 people, and among those, 7 or 8 were researchers because we put a lot of effort into R and D. We got to around 20 million in revenue with fewer than 20 people. That was an engineering, product, and research-driven way of building. Now we are in phase two at around 130 people. We have a bigger engineering organization and we also have go to market and business functions like marketing, sales, growth, operations, and finance. The organization changes meaningfully across those phases.

Leveraging the CMU network, hiring, and the desire to give back

Interviewer: For many current MRSD students and recent alumni, the CMU network is something they hear about constantly, but they often do not have a clear picture of how to leverage it in a concrete way. Beyond meeting your co-founder at CMU, are there stories or examples of how the CMU network helped you as you built HeyGen? I am asking this not only in the sense of hiring, but also in terms of finding early leaders, building credibility, forming research relationships, and learning from people who are further along in their careers.

Joshua Xu: Without CMU, HeyGen would not exist as a company. I met early members, including the co-founder and early founding team members through CMU, and that had a critical impact.

We have had multiple people from CMU. For example, our early head of engineering, Ray, was from CMU and was one of the first employees. Our CTO was also from CMU. We have hired interns as well, especially aligned with the computer vision side of what we do. We have also established research relationships with people at CMU as we explore open problems.

I want to stay connected with CMU talent. The culture at CMU, my heart is in the work, reflects a place with strong talent and strong work ethic, and we appreciate that. When I look back over my career, I got a lot of help from people who were more senior. I would like to help younger versions of ourselves at school, whether that is career development or technical development, and give back to the community.

What is next for HeyGen, and the transition from generation to agent-driven editing

Interviewer: Looking at what HeyGen has achieved, it is clear you have already passed through multiple stages of company building. You reached around 20 million in revenue early on, and you have also crossed major product milestones, including millions of videos generated. If you look ahead specifically to the coming year, what do you think is next for HeyGen? How should readers think about where the product and the mission are going, and what does the roadmap look like at a conceptual level?

Joshua Xu: Our mission is to make visual storytelling accessible to all. Step one is replacing the camera using AI to generate camera footage. Step two is editing. Traditional editing is complicated and the learning curve is high. We believe an AI agent can help: you describe what you want, and the agent produces the edited video for you. We are still at stage one and actively building toward stage two: intelligent AI that takes your footage and does editing for you. That is the biggest step on the product roadmap.

The founder lesson after five years: clear thinking, first principles, and the courage to revise assumptions

Interviewer: Last question. You founded the company more than five years ago, and you have built through very different phases, from an early research-heavy team to a larger organization with engineering and go to market functions. If you reflect across those years, what has been the most important or challenging moment, or the most important or challenging ongoing responsibility, for you personally as a founder and leader?

Joshua Xu: The most challenging part is making sure you have clear thinking. It sounds simple and basic, but it is the most important thing for a founder and leader. Especially early on, coming from a technology background, you have wishful thinking. You think it will work. You think it is what people want. You think people will use it a certain way. Very quickly you find out that most of what you assumed is not true.

Correcting wishful thinking is hard. Nobody forces you to do it. You have to figure it out on your own. If you do not do it, nothing happens. If you do it, it is painful because you have to admit you were wrong and align the team.

Clear thinking is also hard when you face pressure, many opinions, and conventional wisdom. You have to start from first principles: what matters most and what contributes to that, and think from the ground up. That takes courage, because conventional wisdom often pushes you the other way. Sometimes conventional wisdom is right, sometimes it is wrong. As a founder, you have to identify those gaps and break the problem down into essential components.

Interviewer: Thank you, Joshua. That was very insightful and extremely thoughtful. We really appreciate you taking the time to speak with the MRSD community, and we are excited to share your story and perspective with our readers.