At one time, Vijay Pande was better known in academic circles than in investor circles. That all changed a dozen years ago, when Marc Andreessen and Ben Horowitz — who had spent the first five years of their company conspicuously avoiding health care and life sciences — decided the category was worth betting on after all and handed the keys to Pande. At the time, he was a chemistry professor at Stanford University, best known for building Folding@home, a distributed computing project that turned millions of home PCs into supercomputers for disease research. Over the next decade, he grew a16z betting into a habit that managed nearly $4 billion.
So it was a bit unexpected when Pande stepped away from all work and started something smaller last June. In fact, his new company, VZVC, co-founded with longtime investor Zack Warner, is built around making a few focused bets a year rather than dozens, has no employees, and relies heavily on AI for day-to-day operations.
To learn more about Pande’s hard pivot, we spoke to him this week about why he’s making a few concentrated bets rather than thinly diversifying the current market, and about one of the most interesting challenges in AI-driven biotech. Unlike text, biological data cannot be collected from the Internet, so nearly every company will build its own walled dataset. What does it mean for the medical advances promised by AI, and who will actually have access to them?
This conversation has been edited for length and clarity. You can also listen to a more detailed conversation (below).
You’ve said that biology is moving from a “science of discovery” to something you can manipulate. What does that mean?
There was a highly serendipitous aspect to much of the process by which drugs were developed. What has changed is that AI and machine learning have enabled computers to understand very complex things. This means we can now figure out which targets to target with drugs for specific diseases, manufacture those drugs, and even support the most expensive part of the process, clinical trials.
They thought clinical trials were getting cheaper because drug developers were using more synthetic data and clinical trials didn’t require as many people.
I think that is a very big desire.
Although the cost and time of clinical trials are shrinking, especially with the use of AI, trials can still cost hundreds of millions of dollars to conduct, which is why drugs are so expensive. The drug has only a 20% chance of success from the first trial to the end of the third trial. When 8 out of 10 fail and these cost hundreds of millions of dollars, the amortized costs are very high. Usually, the reason they fail is not because the biologist did anything wrong. That’s because all the experiments in which these drugs were designed used animal models, such as mice, and after all, animal models aren’t very predictive of humans. AI models will not be perfect, but they will be far better than any animal model. Once you cross that line, that’s where it gets really exciting.
(The subsequent steps are): Is the drug the right drug for me?
It’s about personalized medicine. . .
The technical term here is so-called precision medicine. When you go to the doctor with a non-trivial problem, the doctor has to guess what’s going on. This is because there are limits to what a doctor can tell you. Then they give you a drug – and if it doesn’t work, they give you another drug, and then another. This happens in cancer, and it happens in many other areas as well. If the first drug was right, we would all be better off. Blood test values are usually compared to the population average. But really, you have to compare it to: ‘Is this (result) strange?’ What we’re starting to do on the medical side as well is understand what’s right for the individual.
Do you think the path to this moment has been slow and steady, or has there been a recent spike?
I think a lot of things are happening. For example, precision medicine has long been based on genomics. But the reality is that your genome is like a blueprint for your home on day one, but your home is quite different compared to the moment it was built. So there are many other things that we can currently measure, such as with proteomics, that are much more relevant to understanding disease and the current state of the body. There is also (a lot) automation of robotic measurements that are naturally tied to AI, and the two work very well together.
The past decade has seen such steady progress in both biological and chemical AI. The biology part is like how can we treat this disease? And the chemistry part is how can we come up with a drug that targets that particular protein. In fact, tremendous progress has been made in the past decade.
You mentioned that biology is one of the few places where AI can’t collect data from the internet. What does that mean for the development of this field?
Where there is no such data at all, people can just train the same thing and cannot extract data from one model to another. It’s a really interesting play just from the pure sense of AI.
But doesn’t this reflect a common problem in medicine: doctors operating in (territorial and often competitive) silos?
We’re onto something really big here. Let’s say (someone) has some kind of cancer and it’s both an oncology and an endocrinology problem. These two doctors don’t really work together very well. What’s really interesting about AI is that it can in principle become a specialist in any field, and it can actually start to see things that no single human being could see. That would be the equivalent of a team of the best doctors shouting in unison at that moment.
But is there enough data sharing to actually realize that vision? I can understand why founders and investors want to protect their (respective findings), but… .
I think one of the bigger trends is that we’re starting to see a shift towards building atlases of biological information, which from a technology perspective are usually fundamental models. And as they become more common, I think we’ll see the same thing happen with open source LLMs that work very well for corporate ones. In short, open source foundational models in biology have a very wide impact.
You’re involved with Genesis Therapeutics, which was born out of your lab at Stanford University, and Incitro, a drug discovery company started by your former Stanford colleague Daphne Kohler. You mentioned that you are starting a company with the founder, who you have known for 20 years. What do you look for in a founder and what areas are you looking for?
There are two areas where I spend the most time. One is AI for healthcare delivery (which I also did quite a bit of work on at a16z) and then AI for clinical trials.
One of the most important things for me[about founders]is that we can really trust each other. Founders have high integrity and do what they set out to do. Ideally, we expect this relationship to last 5, 10 or more years between companies. I want to work with people who are thinking about long-term things like that. Ideally, these are people who are not just trying to win and outdo others, but seriously think about the question: “How can we win together?”
What have you gotten right and what have you gotten wrong in your investment career so far?
Over 10 years ago, when I started talking about AI, machine learning, technology, medicine, and bio, there was a lot of resistance and a lot of people saying, “Oh, that’s never going to happen.” That will never help.” That resistance is almost gone and watching this arc is extremely fulfilling.
I think it took me a while to realize that as exciting as the coolest technology is, it’s always coming to market. I tell founders, especially those who come from the science or product side, to take all of their brilliance and creativity and really apply it to the market development side, which is at least as difficult, if not more difficult, than the technical side.
Understand how we are designing this new company differently compared to what we were running at a16z.
Now we’re doing something completely different…VZ is named after me, Vijay, and my co-founder Zach Werner. He is Z. We’re intentionally very small…On the investment side, it’s really just the two of us. I had originally intended to hire an associate, but thanks to the agency I had built, it turned out that was not necessary.
What level of concentration is “concentration”?
We (not) make 30 bets a year. Probably about 5 investments, not that many, but very concentrated. Adding a company to a general fund is like adding a Facebook friend and can be done very quickly. For Zach and I, it’s more like . . . I would like to have another child. This is a big deal for us.
In that structure, who are you competing with for deals?
The interesting thing about this model is that we’re usually not trying to compete for hot rounds. People make space for us. That’s very different from trying to get a coveted Series A or Series B. Primarily, people seek us out as investors because of what Zach and I can do and how hands-on we are. When I look at people who inspire me, I look to people like Antonio Gracias from Valor. He’s better known now thanks to his contract with SpaceX, but he’s been doing what he’s been doing for 20 years. What Thrive has done with a more focused portfolio is also a real inspiration. Of course, a16z is in my DNA, but I think the others are new additions to the way we think about things.
What’s all the hype about AI and biotech right now?
In fact, AI can find insights that humans alone cannot. What always gets tricky is when people say that AI is going to solve everything. The reason for hesitancy is not due to doubts about AI, but rather doubts about data. LLM works because there is a lot of data to learn from. If the data simply doesn’t exist, AI can’t magically solve the problem.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
