Vijay Pande was once more recognized in academic communities than in investment ones. This changed suddenly about twelve years ago when Marc Andreessen and Ben Horowitz — who had dedicated the first five years of their firm to steering clear of healthcare and life sciences — opted to invest in the sector, entrusting Pande with the initiative. At that time, he was a chemistry professor at Stanford, best known for creating Folding@home, the distributed computing project that enabled millions of personal computers to function as a supercomputer for disease research. Over the following decade, he expanded a16z’s investment into a venture managing nearly $4 billion.

Thus, it was somewhat surprising when Pande decided to leave it all behind in June of last year to embark on a much smaller venture. In fact, his new firm, VZVC, co-founded with long-time investor Zach Werner, centers around a limited number of concentrated bets each year rather than numerous ones, has no staff, and significantly relies on AI for its daily operations.

To dive deeper into Pande’s significant shift, we engaged him this week on why he is focusing on a few concentrated bets instead of spreading himself too thin in today’s market — as well as addressing one of the more intriguing dilemmas in AI-driven biotechnology: unlike textual data, biological information cannot be extracted from the internet, causing almost every company to develop its own isolated dataset. What implications does this hold for the advancements that AI in medicine has promised, and who actually gains access to these innovations?

This discussion has been edited for brevity and clarity. You can also listen to the full conversation (below).

You mentioned that biology is transitioning from a “science of discovery” to something that can be engineered. What does that entail?

Historically, drug development has involved a fair amount of luck. I believe the change now is that AI and machine learning enable computers to process and understand complex concepts… to determine which targets your drugs should aim for regarding specific diseases, to create those drugs, and now even to assist in clinical trials — the most costly phase of the process.

I thought clinical trials were becoming less expensive due to drug developers utilizing more synthetic data, thus requiring fewer participants.

That’s certainly a goal.

The expenses and duration for reaching clinical trials have been decreasing, especially with AI, yet executing a trial can still cost hundreds of millions of dollars, explaining why medications are so pricey. The odds of a drug progressing successfully from the initial trial to the conclusion of the third trial are merely 20%. If 8 out of 10 fail, and these processes cost hundreds of millions, the overall financial burden becomes extremely high. Typically, the cause of failure isn’t because of error from the biologist, but rather that all the tests these drugs were based on utilized animal models like mice, which are ultimately poor predictors for human outcomes. The AI model won’t be flawless, but it’ll surpass any animal model, and once it achieves that, it becomes really thrilling.

[The following phase is]: Is this drug appropriate for me?

You refer to personalized medicine. . .

The term used here is precision medicine. When individuals consult a doctor for something significant, the physician often has to make educated guesses about the situation since their capabilities are limited. Subsequently, they prescribe a medication — if that fails, another is provided, followed by yet another. This pattern occurs in cancer and various areas. We would all benefit significantly if the initial drug was the correct one. Typically, your blood test outcomes are compared to population norms. However, they should truly be compared to: is this [result] unusual for you? What we are also beginning to achieve on the medical front is [the capacity] to comprehend what would be appropriate for the individual.

Would you say the progression to this point has been gradual and consistent, or has there been a recent surge?

I believe it’s a collection of various factors [coming together]. For example, precision medicine has long relied on genomics. However, the truth is that your genome is somewhat like the original blueprint of your house; yet your home has altered considerably since its construction. There are numerous other elements now measurable in proteomics and more that are much more pertinent for comprehending diseases and your current bodily state. Moreover, there has been substantial automation in robotic measurements that integrates well with AI, complementing each other effectively.

Over the last decade, there’s been a steady progression in both AI for biology and AI for chemistry. The biology aspect focuses on how we can treat a disease, while the chemistry facet deals with how we can develop a drug targeting that specific protein. Notably, there have been substantial advancements in those ten years.

You indicated that biology is one of the few domains from which AI cannot simply extract data from the internet. How does this impact the evolution of the field?

This domain lacks the data that allows everyone to train comparable models, and your data cannot easily transfer between models. It’s a fascinating scenario from a purely AI perspective.

Doesn’t that reflect a familiar challenge in medicine — healthcare professionals working within [territorial, frequently competitive] silos?

You’re touching upon a significant issue here. Consider a patient diagnosed with a type of cancer that involves both oncology and endocrinology; these two specialists often don’t coordinate well. The intriguing aspect of AI is that it can, in theory, function as a specialist across disciplines, potentially observing patterns that any individual physician couldn’t perceive. It’s akin to assembling a team of top-notch doctors collaborating instantaneously.

Is there sufficient data sharing for that vision to become a reality? I understand the motivation for founders and investors to safeguard their [individual findings], but . . .

I believe a significant trend is emerging where we’re witnessing a shift towards compiling atlases of biological information — which, from a technological perspective, typically involve foundation models. As these become more prevalent, I anticipate we will observe a trend similar to the one observed with open-source LLMs, which outperform corporate counterparts: open-source foundation models in biology having a wide-reaching effect.

You’re affiliated with Genesis Therapeutics, which originated from your lab at Stanford, and Insitro, the drug discovery firm launched by Daphne Koller, a former Stanford colleague. You also mention incubating a company with a founder you have known for 20 years. What attributes do you seek in founders, and in what sectors?

I’ve primarily focused on two areas. One is AI for healthcare delivery, where I have considerable experience from a16z, and the other is AI for clinical trials.

One of the most crucial aspects for me [regarding founders] is establishing mutual trust — founders who demonstrate high integrity and follow through on their commitments… I envision this collaboration lasting well into the future, ideally, extending over the next 5 to 10 years and beyond, as they develop their next company. I want to work with individuals who prioritize long-term thinking, not just competitive success, but genuinely pondering the question: how can we succeed together?

What do you think you’ve achieved correctly and incorrectly in your investment journey so far?

When I first began discussing AI, machine learning, technology, and bio-medicine over a decade ago, I faced considerable resistance, with many asserting, ‘Oh, that will never materialize. It won’t be useful,’ and so forth. That skepticism has largely dissipated, and witnessing this progression has been immensely rewarding.

I suppose it took me time to fully grasp that while the allure of the most advanced technologies is undeniable, the core always comes back to go-to-market strategy. I remind my founders, particularly those from scientific or product backgrounds, to channel their brilliance and creativity towards the go-to-market strategy since that aspect is at least as challenging, if not more so, than the technical side.

Can you elucidate how you’re structuring this new firm differently compared to your previous experience at a16z?

Currently, we are operating in a distinctly different manner… VZ is named after myself, Vijay, and my co-founder, Zach Werner — the “Z.” We intentionally maintain a smaller size… from an investment standpoint, it’s just the two of us. Initially, we planned to hire associates, but we discovered that with our existing networks, that wasn’t necessary.

What does “concentrated” really mean?

We’re [not] aiming for 30 investments per year… we’re looking at approximately five, not many — very concentrated. Adding a company in a typical fund is akin to adding a Facebook friend — it’s a quick process. For Zach and me, it’s more like… wanting to have another child. It’s a significant decision for us.

Given this structure, who are you up against for deals?

Interestingly, with this model, we aren’t typically competing for top rounds — people often create opportunities for us. It’s a markedly different approach compared to pursuing desirable Series A or B investments. Generally, others are eager to bring us on as investors due to the unique contributions Zach and I can make and our hands-on involvement. When I consider figures who inspire me, I look at someone like Antonio Gracias at Valor — he’s become well-known through the SpaceX deal, but he has been committed to his work for 20 years. What Thrive has accomplished with a more concentrated portfolio also serves as a real source of inspiration. Obviously, a16z remains part of my essence, but I see those other entities as newer influences on our perspective.

What trends do you believe are currently overhyped in AI and biotech?

The truth is that AI can uncover insights concealed from human comprehension. The nuance becomes complicated when claims arise that AI will solve everything. The hesitation stems not from mistrust in AI — it’s about skepticism regarding data quality. LLMs thrive due to the abundance of data available for training. When the necessary data simply isn’t present, AI cannot magically provide a solution to that issue.