Three Years on the Frontier of AI and Biology

May 31, 2025

Intuition is the latent space of experience

Preamble

In 2022, I resigned from my position as the Chief Strategy Officer at the genome engineering company, Synthego, and stepped into the unknown. Many articles are written from curated standpoints about visible career trajectories, but fewer about the experience of exploration. This article is about that: it's about a return to principles and a few things learned along the way about science, technology, company building, people, focus, and mission.

Contents:

The Back Story

My entire adult life has been characterized by two things: a polymathic interest in the intersection of engineering and the life sciences and an almost dilettantish penchant for exploration. I received all of my technical education at UCSD mostly in biomedical engineering/biotech and a PhD in materials science with a focus on nanotechnology applications in cancer. San Diego was an interesting place at the time. My education was funded by a full scholarship given to the top 10 applicants to the engineering school personally by the founder of Qualcomm, who was previously a professor at UCSD. In the early 2000s, a nascent Illumina was growing right down the road, which would later usher in the genomics revolution with SBS technology. In school, I dabbled in far too many things. I wrote my first gradient descent neural network in 2004, spent time in the electrical engineering department learning quantum physics, and even applied to Google via a newspaper ad they were running called the GLAT (Google Labs Aptitude Test) which had an interesting algorithmic problem about vending machines that I solved. They were unfortunately not as focused on life sciences in 2004.

I decided to study mostly biology-related fields because biology was going to be the next great frontier. There is no doubt that the biological sciences have been radically important in the last quarter century, but the technology world has certainly not slacked off. It is hard to say what a better career choice might have been. There is a lot to be said about how to "do school" correctly, and I almost certainly didn't follow standard higher education practices. I was part of starting a new lab, my advisor was an electrical engineer moving into the cancer field, and I had too many outside interests, like competitive tango, and running a non-profit that educated and funded student entrepreneurs.

For two years, one each after completing my undergraduate and graduate degrees, I went traveling solo around the world, a good chunk of it hitchhiking. It was the time in life to do it. I wrote a 300-page book about it which I might one day clean up enough to publish. My first professional interaction with the corporate world came after I returned the second time.

Finding a job post graduate school was probably easier then than now, but I nevertheless struggled. I was not interested in a research job but didn't have the "credentials" for a business job. It took over a year to find a place and it came by happenstance to be almost perfect. It was technical, commercial, and international. I joined a genomics company, Natera, on a small international business development team. I was given 1/3 of the world as my territory and told to bring back business. In the following few years, I traveled even more of the globe, my territory grew to 8+ figures in annual revenue, I led the establishment of public health reimbursement in Canada, and built relationships with leading laboratories and clinicians around the world. Natera ushered in the first major wave of population-scale clinical genomics in liquid biopsies and went through an IPO. I got to participate in many aspects of this. Before this, I had been turned down for an entry-level inside sales job at a competing company. I suppose it was serendipity.

Natera was particularly interesting because it was a clinical genomics company founded by an electrical engineer. This intersection of engineering and biology made the product categorically unique. It also caused blind spots. Early on, there was an obsession over technical performance but much less on clinical evidence. Competitors were all clinically focused. Our customers were doctors, and the pitch was tough. It took the company a while to pivot more toward clinical evidence and publish in that field, but it made the transition and now is the market leader in the field.

In 2016, CRISPR was starting to take hold. It had been 4 years since the seminal publications and the research and clinical worlds were just starting to figure out how to use it. Synthego was at the right place at the right time. CRISPR is a genetic engineering tool. Synthego was founded by engineers from SpaceX who wanted to bring engineering principles to biology. It was an excellent match, and it is fair to say that Synthego was seminal in introducing CRISPR to the mass life science market. The company was at the forefront of CRISPR therapeutics, high throughput biology, cellular and genetic therapies, and a broad transformation in drug discovery R&D. It is also fair to say that Synthego was one of the high-flying platform companies of the techbio ZIRP period (zero interest rate phenomenon). I won't go into the dynamics of the techbio ZIRP phase, because it is covered in many places, but suffice it to say hyper growth was the goal and fundraising was at a peak. The period saw many technology investors accustomed to SaaS dynamics enter the life science and tremendous amounts of money fueling valuations and expectations. This lasted through COVID with a stark reset on the backend.

I joined to lead business development (which unbeknownst to me meant sales) as the sole lead pre-revenue, built the revenue to 8+ figures in 18 months for the second time, and ended up in the C-suite having done at least part of nearly every major function at the company, formally as the Chief Strategy Officer. I can say I would not recommend this title to most. It has sometimes been characterized as "accountable for everything, responsible for nothing", a shadow CEO of sorts in my case. That said, the scope of experience concentration was unbeatable. I cut my IP teeth on the CRISPR field, which is notoriously complex. As an aside, the original CRISPR/Cas9 patent battle between UCB and Broad is probably the last major battle of an era, being not just very technically demanding and complex but also straddling directly the massive transformation in the US patent system from a first-to-invent to a first-to-file system.

I was core and/or led/built companywide restructuring efforts, fundraising, sales, product, strategy, bd, legal, and touched almost every aspect of the company at some point as the company grew to ~500 people from the ~20 when I joined. I became a deep generalist in the mechanics of company building. It is one of the hardest things anyone can do. I have deep respect for founders and entrepreneurs. It also holds up a mirror to what one might really want to do, and for me, that was getting back to my exploratory ways. Alphafold, GPT3, and a host of other AI technologies were just starting to become available, and I wanted to know what that space was all about, so I went to the frontier.

Stepping into Empty Space

Stepping away from an executive position without a specific plan was not easy. It required decompression which took longer than expected, likely on account of the added stress of still being under COVID working conditions. It was also somewhat sad. I believe deeply in the mission of connecting the engineering and biological disciplines and the potential benefits that it can bring to the world and attach a large part of my personal identity to my professional work, and that was now gone. A recurring theme in this article will be how challenging (and also powerful) it can be to be without a specific professional label.

After a short decompression, the first task was structuring my days. This period for the next 6 months was deeply satisfying. I read incessantly and talked to everyone. At the outset, this was very open-ended; I followed my curiosity. I got quite interested in the DeSci space and went to hacker houses in Boston with some of the early developers in this space. I attended the first North American DeSci conference in late 2022 at MIT and was an early advocate of platforms like ResearchHub. However, I found my interest continually pulled more toward the hard sciences, specifically the idea of LLMs, deep learning, and the idea of a "language of life". As I read more of the technical literature about this space, it seemed natural to me that the idea of training a transformer on sentences should be conceptually similar to genomics, which had been a large part of my technical background.

The Enformer paper had been published the year before that described a combination convolutional/attention mechanism and it seemed natural to extend that. I even wrote a philanthropic proposal to fund the creation of a genomic foundation model using this approach. It turns out that the Nucleotide Transformer and the Geneformer models were already in the publication process, but not yet out. It would become a recurring theme that as I was becoming acquainted with the field, I would re-invent the intuitive approaches that were pending publications but not yet out. The experience of reinventing my own ideas that were just emerging without having prior knowledge led me to believe I was on the right track.

Meeting the People

There is an unfortunate siloing that happens when working at a single company in that the majority of people that you interact with are relatively bounded within your industry. It continues to surprise me how relatively siloed many ex-colleagues are to anything outside of the specific work vertical they occupy. The interesting stuff is at the intersections.

In late 2022 I embarked on a rather ambitious campaign to meet people. It was a rather unabashed cold outreach to the authors of every paper that I read to talk about their work. And there were many, many papers. I also added the hopscotch asking everyone who else I should talk to and forwarded introductions. All in all, I added several hundred new researchers, investors, and operators to my network, all at the top of their fields. One of the most satisfying aspects was just being able to genuinely engage with their work without a specific agenda other than curiosity. I found that many people genuinely appreciated this. And it was also deeply interesting. This also led to some fascinating intersections: new companies being started (some of which I had the opportunity to invest in, several of which I have advised), increasing invitations to different technical groups and private events, speaking and writing engagements (including being an invited panelist across from the Deputy Director of the FDA discussing AI in drug discovery, among others), and the growing ability to be an effective integrating resource across disparate disciplines.

I make a lot of introductions.

I now intersect multiple different technical disciplines with networks in each ranging as broadly as high-performance computing (down to the technical details of writing GPU kernels), hardware architectures, foundation model development and training at major AI labs, data engineering, drug discovery, chemistry, tech news/media, AI-Safety, AI-agents, software development, startup founders, venture capital, intellectual property, business development, clinical trials, health-tech, therapeutics manufacturing, cell therapies, and more or less the whole gamut of disciplines in and around the intersection of engineering and the life sciences. And I can speak pretty fluently across all of them.

I started attending NeurIPS out of personal curiosity and met a whole new cohort of people who are now regular colleagues on the conference circuit. I started hosting casual dinners with interesting people. I "attend" JPM at the periphery events with a full schedule each year to connect, reconnect, and enjoy the serendipity of the occasion. In 2024, an encounter with Jensen Huang led to a (at least from my perspective) productive bonding over our mutual penchant for black leather jackets and a subsequent email exchange about his assertion that "biologists were angry" because we used words like “inhibit”, and “target”, and had a “war on cancer” — a claim he made I suspect a bit in jest at an event. I don't know if I was convincing in explaining that we're just angry that biology is hard and disease is bad, but I was certainly surprised to hear a response from him personally in my inbox.

This nexus certainly has its positives. But there is one major downside: it is difficult to label.

In this exploratory quest to "meet the people", there has been no shortage of recruiters along the way who have been looking to create a label for me that fits the roles they are looking to fill. For various reasons, these have either not been interesting to me or successful for them. Part of it is my mission-driven personality and finding/adopting a third-party mission is harder, and part of it is my desire to be able to use the scope of the skills that I uniquely have, which is admittedly very difficult to find a role within existing companies to do. This has, however, led to a number of informal advising relationships for founders, CEOs, VCs, and other business leaders who are looking for insight into the implications of AI in the life sciences. However, like my experience post graduate school, it's often a square peg/round hole situation. I'll discuss this in more detail in the Lessons in Venture section.

Cancer Research and Kilimanjaro

One of the more serendipitous aspects of meeting the people was being introduced to Luke Timmerman, the eponymous editor of the legendary Timmerman Report and adventurer extraordinaire. Over the last several years Luke has created a community of ~150 executives from across all dimensions of the biopharma industry to give back by raising money for a variety of nonprofits and create experiential relationships through an adventure that combines his love for the outdoors. Collectively this has raised over $13M and has ventured to the top of Kilimanjaro, Everest Base Camp, and a host of other beautiful and challenging destinations.

I was invited to join the 2024 Kilimanjaro team. We raised over $1M for the Damon Runyon Cancer Research Foundation and summited the 19,341 ft peak as a team in February 2024. It was epic. It was also challenging.

Raising money for companies is hard; raising money for foundations is harder. We each committed personally to raise $50k for the DRCRF, a wonderful foundation that funds early-career scientists doing leading-edge research. The philanthropic community of individuals that fund foundations was an interesting one to be plugged into and the support of those in my network for the effort was wonderful.

An adventure like climbing Kilimanjaro is a bucket list item and a personal one. It challenges people in different ways and creates a unique shared memory. The Timmerman Traverse community will continue to grow and continue to bring value to the world. It is an honor to be part of it.

Lessons In Venture

New ventures are unique. They are fragile at the outset, come together through circumstance, serendipity, and collaboration, and are hard. Most ventures that end up being successful have some kind of origin story which is often redrafted in retrospect. LinkedIn is full of posts about venture advice, fundraising, co-founder selection, etc. that all seem rather pithy, but some of it is true. In this exploration, I initiated, was part of, or assisted with no fewer than 5 new ventures at the earliest stages of inception and development. I've also received dozens of pitch decks for feedback, advice, or investment. There are several things I have learned along the way.

It had been evident to me for a long time that data in the experimental life sciences was a limiting factor in the utility of deep learning models in practice. It was almost cliche to say that "we need more data" in biology, but it was also true. By and large, existing data was not collected with AI in mind (e.g., provenance, metadata, etc.) and/or siloed in private companies. There were interesting proof-of-concept publications on genomic foundation models, but in practice, they did not have great utility.

In 2023, I started to pitch a venture that would aim to address this: a product-led AI software platform that used revenues to fund data generation to improve the models. This was different from other proposals out at the time which were largely either pure software or platforms for therapeutic data generation. The model was designed to balance the cost of datasets, with the value of the tasks they enabled in a very customer-centric and product-driven manner. In short, it was a model that would iteratively fund proprietary data generation in a sustainable way on the way to building better foundational models. The economic viability would depend on the balance of the value-add for model task performance and the incremental cost of data generation. The vision was big: I tend to think in ecosystem scale ideas.

The idea became quite developed and I went through the rounds of pitching early-stage VCs, potential co-founders, customers...the whole gamut. This was part of the "meeting the people" campaign. In the end, however, ideas only go so far and there were some practical economic issues to gaining traction. I still believe that this general model is viable with the right approach, however, with large-scale efforts like CZI funding data on the scale of 1B cells, the scope changes. However, during this process, I met many new and interesting people which led to the second phase.

Lesson 1: Vision alone doesn't sell: necessary but not sufficient

Silicon Valley is great for private dinners. Collections of unique people, researchers, investors, and operators that gather for connecting and relationship building. Being part of this circuit is a great experience. It was at one of these dinners that I met the founder of a company doing something very similar to what I had been proposing and we started to chat. He was deeply technical and had built a very impressive platform but had done so largely separate from tactical customer feedback, as many technologists are inclined to do.

A strong shared vision and complementary skills (mine having focused much more on business and economic implementation) led to a natural pairing and we began discussing working together. There was a hitch, however. His company had already been a few years in the making, had received some early funding, but needed to raise again, and investors were looking for commercial traction which hadn’t yet been solidified. The confluence of circumstances made it challenging to find an arrangement that could work on all ends. The timing wasn't working out. It was a situation that a few years earlier could have been an excellent match.

Lesson 2: Timing, for the right relationships, life stage, and risk appetite is very difficult to get right

This led to the third stage. Around the end of 2023, I was introduced to and developed a relationship with another founder who was leaving his first venture and looking to start a new one in the AI x healthtech space. He was a software executive in a prior life and had learned the hard way that a software product alone doesn't provide for a successful business if you neglect sales and business development, an area that I had spent a great deal of time focusing on. We spent the next few months discussing these ideas, day-long working sessions on decks, business strategies, and fundraising. I had found myself frequently in the position of providing feedback to early-stage founders like this.

There is generally an open community in the Bay Area of this type of support and I enjoyed it. I thought over time that the vision was compelling and ultimately was offered to join as a co-founder – with a single-digit equity stake. The company at the time was pre-funding. Admittedly, this was not an inspiring offer. It also led to a recalibration of my estimation of the perceived balance of technology vs business expertise in early-stage ventures. Silicon Valley is a technology-centric investing culture. It is not uncommon for a technical team to get funded and say that they will hire "business" talent later but is uncommon for a commercial founder to get funded and say they will hire "technical" talent later.

Lesson 3: Business and technical talent are often seen as mutually exclusive as a default. Technical talent is often more highly valued, but companies more often fail for lack of business talent.

This dichotomy between technical acumen and business acumen was one I have encountered regularly. In a single day, I might hear from someone "I had no idea you had commercial experience" and "I had no idea you were so technical" depending on who the audience was or which article of mine they had recently read. There can exist technically deep business leaders.

This led to the fourth stage: strengthening the tactical technical chops myself.

In early 2024, I decided to start getting into the technical weeds myself. I had always had a penchant for software and algorithms; I used to program graphic games on my Ti-82 calculator in high school but had not been a career engineer. That said, LLMs were making both the practice of writing code and the process of learning about it, very accessible. I indulged my intellectual curiosity across the whole stack, from learning about operating systems, programming languages, processor architectures, software stacks, distributed computing, databases, deep neural network training, and a whole host of other topics. I started writing code to prototype a diverse range of interesting use cases, train neural networks, and experiment with new architectures. Much of this process has been documented in various articles on this Substack.

One of the primary focuses I had was to understand how effective LLMs could be at analyzing scientific literature, ultimately on the path toward the automation of science (which has been a consistent thread in my career). A natural testing ground was to evaluate drug development pipelines and to get real-world feedback, which was developed into a system for assessing public biotech equities. During this time, I met a person who said they could help raise capital to potentially start a hedge fund. This seemed at first like a good fit, but it became apparent early on that there was a mismatch in our communication and operating styles, and the relationship quickly ended.

Lesson 4: Maturity and communication style are essential in early ventures but it is often rare to find a match.

Starting a new venture is about an idea, but it is dependent on the people, the timing, and the complementary skills. In my view, there is also a high value placed on the maturity of experience of what has worked, and what hasn't, and how to navigate the messiness and complexity of building a new venture. One of the best lessons I have learned in life is that I do not have a monopoly on the right way to do things or the ability to predict the future. I have, however, built a solid foundational set of frameworks to make decisions that are likely to be right and use them as alignment filters and rubrics to choose where to invest energy and with whom. Turns out you can learn a lot about company building by hitchhiking in foreign lands.

Writing is Thinking: On the Future of AI, in Bio and Beyond

I highly value writing cultures. Language is a special construct. It forms a lot of how we build our interpretations of the world. It is not the sole form of intelligence, but it is a significant one. And the written word, when constructed well, is a powerful form of thinking. I have traced my thinking on the topics at the intersection of AI, Biology, and the role of AI in the world throughout the posts on this Substack and through proposals written to leading AI labs, philanthropic funders, and government research agencies. The audience has grown to include leading researchers, VCs, Fortune 10 founders and executives, techbio founders and CEOs, leaders in technology media, and a diverse range of others.

The field of AI is split between the doomers and the accelerationists. The middle ground seems more likely to be true. There is a maxim that the best prediction for how long something will last is how long it has already existed. To this point, we should consider the duration of the lowest frequency features of our society and expect that those are not going to disappear overnight. The introduction of AI into the world will be disruptive, but it does not solve all problems, nor is it likely to. This is particularly true in complex systems like biology.

I expect that AI will create a form of "pseudo-knowledge" in biology that is deceptively difficult to tell from actual knowledge and this will propagate extensively. There is a real risk of losing critical thinking skills. There is also the practical reality of the intersection with the physical world and the feedback latencies that it introduces. In drug discovery, I describe this as the "impedance mismatch" between, for example, software platforms for nominating targets and the physical prosecution of those targets to validate them. Impedance mismatching between AI systems and the real world will create more disparate impact rates across different domains. Anything that can be conducted in a computer system will likely see rapid developments, and cyber warfare will be a significant concern. AI safety will continue to be a very important topic both technologically and socially.

On the Horizon

The questions on the horizon are complex – they intersect science, technology, policy, product, safety, and a diverse scope of implications. The ecosystem is equally as diverse. Arguably there has not been such an intersection of the technology and biology worlds as there is today. Vast sums of money from the tech world are pouring into biology as the next great frontier and there is continual talk of "curing all diseases". This intersection is new. It is beyond the computational biology wave that has been widely adopted in the life sciences and borders more closely on the philosophical one about what "intelligent machines" can learn about complex systems. There are few translators that bridge this gap, but many that promote it aggressively. Andrew Dunn wrote a good article on this and how it parallels the CRISPR investment life cycle. Having a front-row seat to many startups at this intersection and lived through a boom and bust cycle, it remains striking to me how few have thoughtful business models. In practice, any evolution of such technology will still be done by humans, and human dynamics will be the dominant factor in their success, at least until we have a single-person, full-stack drug development company.

For my part, the real-world impact of AI on the evolution and development of life science companies will remain my focus.

I’ve written more on this topic in the Related Reading section, where I explore many of the themes I continue to work on today.

Intuition is the Latent Space of Experience

I believe I coined this phrase in a LinkedIn post a few years back, at least to the best of my knowledge. This pairs nicely with another life maxim:

Good judgment is the result of experience which is often the product of bad judgment

When I spent two years traveling and hitchhiking around the world, one of the most valuable aspects was the intensity of the experiential diversity. The rate of learning is very high. You get the same experiences in start-ups. The latent space of this experience begins to generalize better in what we call judgment. The role of judgment is not to be right in all cases but to bias to the upside. That, at least in part, is what exploring the frontier has been about.

for more: jasonsteiner.xyz

jason@jasonsteiner.xyz