Networking, BioSecurity and Generative AI

One thing I started doing a couple of years ago, that I've been meaning to share, is this. When I meet people for the first time, I think there are two kinds of meetings you can have in life: the meeting you're really looking forward to, and the meeting you're really dreading. It's unrealistic to imagine a world where there aren't meetings you're dreading — although maybe at some point you become so successful that you only have meetings you enjoy. But for the most part, when I meet new people, we all do a little pre-work, a little homework to understand someone's background. And I ask a second question: what is this person uniquely positioned to teach me?

You could flip that the other way, too — what should I want to learn uniquely from this person? I'll come in with a hypothesis, and as we're talking, I'll try to tease it out. And the converse is also true: based on this person's interests, or who I perceive them to be, what can I share that's unique? What can I contribute to their line of thinking that they might value? That's gone a long way in terms of finding meaning in every conversation, whether it's a five-minute catch-up about a deal or a sit-down breakfast with a founder sharing something new about a space.

As an investor, or as a founder?

This is going to sound cliché, but I think everyone has something really unique to contribute. Experience and achievement are one thing, but there's no substitute for passion, curiosity, or grit. And frankly, the people I learn the most from might be on that side of the equation. We've all studied the experienced, successful people in a very didactic way — but some of the tidbits that pique my interest throughout the day come from people who are just passionate, or hustling, or curious, or trying to break into somewhere new.

So I always come to conversations with a certain humbleness: I will learn something from this person, and it's my job to figure out what that thing is. If I leave and I haven't learned it, then we didn't have the right conversation. And if you feel there's a power dynamic in a conversation — especially if you're looking for mentors — the most important thing you can do is put yourself out there and have strong beliefs. People enjoy speaking with others who share a perspective you can then tease apart over time. If you only ask questions, only wanting to know what the other person thinks, that's not really a bi-directional conversation.

Some of my favorite mentors really instilled in me that it's a two-way street. It doesn't matter if someone's the CEO of a Fortune 50 company and you're an intern — there's a discourse to be had, and it's the responsibility of both parties to figure out what that's going to look like.

I think it all comes down to consumption. You're a physician, so you know that when you're steeped in this stuff day to day, it's really easy to become myopic, and to become disgruntled — oh, it's never going to work. So there's a balance between being a domain expert who understands what's theoretically possible, and mirroring that with a voracious appetite for information, whether it's healthcare-specific or coming from another industry.

One thing I'm really privileged in at Bessemer is that we're a generalist fund. I have colleagues who are experts in space tech, crypto, fintech, enterprise software — all these areas where incredible generational companies are being built. So I can sit there and say: okay, if generative AI is transforming consumer social because we now have synthetic voice and synthetic avatars, the logical next question is easy — what happens when that technology is available in healthcare and life sciences? So that's one: information diet is key.

I took a design class at RISD in undergrad, which was one of the more formative courses I took. Even though I was studying science, that class really shaped how I view the world. The one thing the professor told us was: it is your responsibility to consume as much information as you physically and mentally can. That is what will set you free. This professor, at the time, was also spending about 70% of his lived experience in VR — a really interesting person. But you look at someone like Warren Buffett, who reads for 80% of his day. So that's become important to me: how can I train my priors, if you're a Bayesian about it, and figure out where there might be green shoots, where there might be signal in all the noise?

The last thing, structurally, in healthcare and life sciences — you can cheat a little, at least in the US, because regulation has spawned so much of the innovation in the ecosystem, particularly in health tech but also life sciences. So if there are interesting bills or proposals happening politically, those are often helpful threads to run down. We could look at HITECH, we could look at the CMS interoperability ruling, and many others in this light.

It's important to do that. Whenever there's a cultural zeitgeist like we're having now with generative AI, it's important to ask yourself: why is there so much excitement about this thing? I'm an optimist, so when people are excited, I'm usually pretty excitable too. I also bring a skeptical lens — but especially in technology, it's usually a matter of when, not if. So approaching with optimism, plus that primary-source discipline, is key.

I'd take it a step further and say lived experience. If you work in care delivery and your only knowledge of care is your own experience as a patient, or headlines you've read on Modern Healthcare or TechCrunch or Twitter, that's insufficient to do groundbreaking work in care delivery. For me, going back to medical school and actually seeing what it's like on the ground to care for patients was a really important primary source. Granted, I'm early in that journey, and it's only a fraction of what lifelong committed physicians have — but it's enough to have empathy for their experience.

The other thing, which I've been thinking about recently: in healthcare we become complacent about our access to primary sources, because we rely so much on our own personal experiences. Well, I can go get X, Y, and Z at CVS, so that problem's solved. But for many people working in tech, we're not representative of the population. So if you had the opportunity to go shadow at a clinic, or volunteer at a nursing home — just to get an ounce of lived experience — it'll go a long way. Judy Faulkner at Epic has done this brilliantly over the years: they have rotational programs where their software engineers go into the hospital and see how physicians interact with the product. We could talk about Epic all day, but that was a brilliant human-capital strategy they've maintained over decades.

I'd honestly say you just have to care. You're talking to another human — you just have to care about them. Taking a step back, networking can be really anxiety-provoking. In venture and startups there are a lot of ways we collide: at conferences, over dinner. Some are big forums that feel very high-stakes; others are more intimate and casual. So there are different vibes to navigate.

In general, repositioning it as "I get to network," not "I have to network," is a really important mind shift. I'm in a position where I get to go talk to five people in this room who know something I don't. It's that underlying desire for learning and curiosity that fuels a lot of the networking I do. But I don't even consider it networking — the beauty of the venture job is that I really like the people I get to work with. I like my colleagues at Bessemer, I love the founders I get to work with, and even founders we haven't invested in, I'm texting them, chatting with them, FaceTiming them. Allowing yourself to build that personal relationship, and genuinely caring about someone as another person, is helpful.

And I'll be transparent: I don't look like a lot of people at a lot of the events I go to in healthcare, tech, and venture. I'm African-American and white, and my ability to share my personal story — how my personal identities inform the way I see the world — has allowed other people to let their guard down and share more about themselves. I'm not a proponent of "bring your whole self to work" — that's totally up to the individual — but bringing the authentic things about you that shape your worldview to conversations can go a really long way in building genuine relationships.

You would know best — your content is prolific. I was very averse to putting my ideas online for a long time, and I have to give credit to one of my best friends, Nikhil Krishnan, the thinkboi of healthcare — I think he'd be okay with me mentioning that. We'd been having super interesting conversations, hanging out in Brooklyn, getting to know each other as friends as well as colleagues in this space. And one day he said: look, I think it's more of a liability for you not to put your ideas out there than it is a risk.

I was really scared. What if I say the wrong thing? What if I get canceled? I had all these fears about the bad things that were going to happen if I shared what I thought online — and they were largely unfounded. He walked me through the talk track: what's the worst thing that could happen? I put an idea out, people disagree, they say I'm stupid, I get canceled. And what's the best thing? People are interested in the same topics, they reach out, I get to know them and learn something. Either way — if someone disagrees and I'm wrong, I learn something; if someone likes what I wrote and introduces me to a person or an article, I also learn something. So when I realized that no matter what, I'm going to learn something, I was like, okay, I'm in — let me figure out how this content thing works.

Then it was small things: do you want to sound formal or casual? As a person I'm built in a very formal way — I have to constantly tell myself to be less formal. My strategy overall is very scattered; I welcome feedback. But I generally just like to share things I find interesting or thought-provoking — shower thoughts I want feedback on — with the hope of meeting like-minded people thinking about the same things. If we had this conversation in three or five years I might say something different, but that's where I am today.

I think they're scared. There's a lot of fear. I know a lot of people far smarter than I am whose thoughts I'd love to hear. It's also a big time commitment — you actually have to carve out time to write content and make it coherent. And everyone says this, and it's true: writing is such a brilliant forcing function for congealing ideas and making theses coherent. It's amazing what the human mind will do as you go down that rabbit hole and run the exercise back.

But I agree there's opportunity in healthcare and life sciences. In some ways the industry is plagued by the history and historiography of peer review — this notion that if something's not peer-reviewed, it can't be put out or discussed. There's merit there; peer review is an important part of the scientific process, and on a separate day we could talk about its shortcomings. But the interstitial space between hypothesis generation and a published paper — there's immense opportunity there. Something conventional media doesn't do well is follow the exciting work being done. Every week there are new preprints hitting servers, new science coming out — some of it actually executable. I could download the model you built for your paper and run it on my machine hours after you put it on a preprint server. That's a remarkable velocity of information sharing, but traditional media isn't covering it well, and a lot of the discussion is happening on Twitter.

So there's probably an opportunity for someone to build that content platform or discourse that brings these papers to life. An area I'm really passionate about is the access and readability of science. One of the things I shared recently on LinkedIn was a study looking at how the readability of science is declining over time, particularly in molecular biology and genomics — areas really important to society. So the confluence of all these things — scientific writing, healthcare-specific writing, domain expertise, and the massive amount of information now produced online — creates immense opportunity to make it more accessible to people who want to get involved and fight the good fight with us.

Honestly, I don't, but I hear your point. There's actually a convention in venture: if you go to an early-stage startup's website and it's way too polished, you think, oh gosh, they spent so much time on the website and no time on the product — immediate pass. So the analogy would be, oh, you're spending so much time on all this external stuff, what are you actually working on?

But there are ways to be very efficient with content. I'm not saying use ghostwriters, and I'm not saying ChatGPT all your articles, because that's not interesting. But you can say: every week I'm going to sit down on a Sunday and write all my tweets for the week. If you're naturally a consumer of information, you're going to have thoughts — so keep a notebook next to your computer, or use Notion, and track it over time. It doesn't have to be a huge time sink. I couldn't maintain a content empire while studying for Step 1, being productive in the clinic, sourcing investments, and supporting portfolio companies. So it's got to be part of a strategy, but not a full-time job by any means.

As I alluded to earlier, I see two kinds of technological transformation in healthcare and life sciences. One is linear, incremental progress — what we've had for a long time, getting a bit better each year, posting better quality ratings. And then there's nonlinear: the implementation of software for medical records, various approaches to programming biology. Venture as an asset class is fundamentally designed for nonlinear outcomes — it's not designed for technologies that scale linearly. That's been called into question recently as we've looked at a lot of what's happening in healthcare services and asked: is this stuff actually well-suited for the venture asset class? And, derivatively, does venture do this category a disservice by putting unnatural pressures on organizations to grow in ways they weren't designed to grow? You could list a number of examples where that's manifested.

So right now, if you asked me — if I had a hundred dollars, I'd invest 75 of it in science and probably 25 in software or high-margin, tech-enabled services. On the software side, I'm spending a lot of time in generative AI. There's a lot of hype, but it's a technology that's been around a long time. At Bessemer we made our first generative-AI investment in 2018, into a company called Subtle Medical that was using generative AI for image denoising. Then in 2019 we invested in Abridge, which at the time was using generative methods to transcribe conversations between physicians and patients, and physicians and other physicians.

We're now at a point with GPT — 3, 3.5, soon to be 4 — where the performance of these foundation models is sufficiently interesting to produce results that might be valuable at scale. So I'm spending a lot of time thinking about how generative AI first transforms the healthcare back office — all the gutsy stuff you didn't get out of bed or go through training to do, particularly revenue cycle management. That's the 25%. The other 75%, in life sciences, I'm spending in computational biology — thinking about how these same generative methods, married with efficiencies in scientific methodology and sequencing, can help us build larger, more efficient, more successful drug-development organizations, with the end goal of making medicines that work for patients.

The wonkiest thesis I have right now — and I don't even know if it's that wonky — is biosecurity. Some quick context on why I'm interested. My dad worked at the EPA for 30 years; in his last role he was director of emergency response, which is one of the worst jobs in the world, because you get a phone call every time there's a wildfire, a hurricane, flooding, any toxin in the water. But in the last few years of his career, notably with COVID, the kinds of things he was responding to increasingly became biochemical in origin. Marry that with what we saw during the pandemic, and then with the fact that in an afternoon, using off-the-shelf computational methods, researchers have designed highly toxic molecules in silico that could in theory be easily shipped and manufactured. I think this is going to be a very important area over the next few years.

The questions that loom are: what are the business models for the technology companies — software as well as molecular technologies — contributing to the biosecurity discourse? And what kinds of teams and skill sets are required to build these companies? I'd imagine it requires folks with deep computational experience, deep biochemical and scientific experience, security experience, and an understanding of how contracting works in government — specifically defense contracting, and being able to tap into those RFPs, budgets, and deployment cycles. So that's a little tour of what's occupying my mind space right now.

For biosecurity. If you're a synthetic biologist today using computational methods to develop new molecules, how do you know that what you're developing isn't potentially toxic? Do you have access to every screening library produced — by the federal government, state government, international governing bodies — to say, hey, the sequences or nucleic acids you're thinking of putting together could actually cause harm to society? Do we have appropriate guardrails in the synthetic biology space today to prevent catastrophic biochemical events from happening at scale?

My gut and my research suggest that no, we don't. I don't mean to be alarmist, but my sense is that — again, as we saw with software — it's not a big mental leap. As things became digital and software proliferated across industries, we needed to secure that information, and the cybersecurity sector blossomed as those pieces of information came online and became interconnected. We have a very similar trend happening now, as biology and chemistry become increasingly programmable and digital. We're going to need infrastructure to make sure that information is both kept safe and not used for downstream harm. These technologies are very early — this is a very early thesis — but I'd imagine the category itself will produce tens of billions, if not hundreds of billions, of dollars of value to society over the next ten years.

I think there are going to be a couple of interesting billion-dollar platforms. I still think payviders are an interesting concept — this notion of more tightly coupling the reimbursement of care with the delivery of care. As some of the larger payers, whether United or others, pick up acquisition activity over the next year or so, as assets become more cost-effective to buy, we're going to see more experimentation with the payvider model. There are great examples already — Kaiser, or other IDNs in the US — but I think there are going to be massive businesses built that align providers and payers this way.

The other area of massive enterprise value is the application of generative technology in healthcare. I talked about the back office; if we think about the front office, there's huge opportunity in prognostics — actually predicting disease progression using digital biomarkers. You could use images to predict images, or clinical data plus images to predict some later disease phenotype. That's an area I'm extremely passionate about. And you might appreciate this as a physician: one of the biggest realizations I've had in medical school is how poorly we're able to characterize most diseases today. We've done a good job in a number of areas over the last several centuries, but there are still so many diseases where I leave the lecture and the takeaway is: these are the symptoms, but we don't actually understand the etiology and pathophysiology of why this disease is happening. There's going to be massive value provided as we use computation, and new omics or clinical data that's come online, to better characterize these diseases. Put that in the computational diagnostics realm, tightly coupled with therapeutic opportunities. And then, within the next decade, I certainly believe there's going to be a billion-dollar, or multibillion-dollar, biosecurity company — frankly, I hope there are several.

Being practical, the revenue cycle is probably one of the most interesting from a business perspective. Think about the role of billers and coders today — having to go through clinical corpuses, medical records within the EMR, to pull out relevant information and code it in a way that can be submitted to a health plan for reimbursement. This is a task generative models can be applied to. Is GPT 3.5, the model under ChatGPT, well-suited to do this today? No — I haven't seen published literature suggesting it can. But can you give ChatGPT a clinical vignette today and have it spit out an ICD-10 or CPT code? Yes. I've done it, with clinical vignettes from medical school and from the hospital, and it can do this.

So the opportunity is to take a foundation model like GPT and apply a domain-specific model from the healthcare industry — one that better characterizes the universe of appropriate codes — and combine the two. In theory, that amalgamation should be able to automate the work of billers and coders. With any technology like this, I'm a big believer in crawl, walk, run. I'm not that interested in people losing jobs — that will happen as these technologies are adopted, and our job as a society is to figure out how to reskill and redeploy people. But we have to start with human in the loop: first empower workers to be more efficient, then identify ways to peel off tasks and keep improving performance and accuracy. So the revenue cycle space is going to be massive. I'll be honest with you — I'm not not excited about revenue cycle, but I don't get up every day dreaming about it. Still, there are going to be a lot of really big businesses built there.

I'm a big believer that no single resource or platform has a monopoly on good ideas. Right now, honestly, I just got out of a hematology exam, so I've been reading a lot of lecture slides and First Aid on hematology. But in all seriousness, I read a lot of content from Twitter — preprints, Substacks — anywhere there seems to be a smoke signal, there's fire, so I run that down. I listen to a lot of podcasts and audiobooks, even movies. I'm really interested right now in history and historiography — historiography being the study of the way history about something was written. I'm interested in that as it relates to healthcare and life sciences: who's writing about the early days in the lab with some discovery? What was that like, and how did they write about that experience? I wish there were more people writing about them — that's actually a gap in the content sphere I haven't been able to solve.

One book I revisited recently is The Book of Why by Judea Pearl. Judea Pearl is a well-known computer scientist who's written a lot about causal inference — how we know what we know, and how we think about evidence generation. It's a thought-provoking book about the foundations of knowledge, and one I'm ruminating on at the moment.

I appreciate that. Thank you.

We've touched on this in a few ways, but to make it explicit — I've been trying to describe my content-consumption strategy, my focus on curiosity, and that's essentially Bayesian learning. For the computer science geeks on the podcast, Bayesian statistics explores how your belief that something will happen is based on prior distributions representing the information you've had access to. So when I have to make a decision, I make it based on the context around me plus everything else I've trained myself to understand — in Bayesian statistics, those are your priors. Without getting too into the weeds on the mathematics, the common theme is curiosity: it's information consumption, and appreciating what actually works for you. Do you learn from reading books? Do you learn from talking to people? A lot of what I learn comes from conversations with friends, and really prioritizing that — carving out the time. I'm going to spend the next hour or two — in Warren Buffett's case, 80% of his day — reading, synthesizing, and learning. There's no greater joy for me than a free weekend where I can just go on the internet, peruse, learn, watch stuff, and consume. That's a pastime. So that's the main thing: that Bayesian learning framework.

The other two things, particularly salient now as I balance medical school and investing. One: keep it simple. It's really easy to let your life get super complicated, and the likely outcome is unnecessary stress and burnout — you might lose desire or passion for what you're doing. I've definitely had moments where I've burned out and felt exhausted. So a big theme this year is keeping things simple and saying no to stuff that isn't giving me energy.

And the last one — funny because it's from Charles Darwin — is adapt or die. As new technologies or new information come out, it's easy to have a visceral reaction, especially if it's troubling. Like GPT: why is it so good at what it's doing, and what does it mean for me? Am I going to be replaced? It's very easy to resist change and innovation. And let me be very clear: I think all of this technology needs to be explored in collaboration with ethics researchers, sociologists, and folks who can help us characterize the impact it'll have. But that said, a lot of this stuff is a matter of when, not if. So the derivative is: don't sit on the sidelines. If you have an opportunity to get in early and contribute — not just to talk about it — lean in. Just keep staying curious, and that loop will feed itself.