I’ve seen a lot of hype in Silicon Valley. I’ve seen founders promise the moon and deliver a rock. So when I first heard about an AI platform that could discover new drugs 10x faster and cheaper, I rolled my eyes. "Here we go again," I thought. Another black box making bold claims it can't back up. I’ve been in the trenches for over a decade, with two exits under my belt and over 200 angel investments in companies like Anthropic, OpenAI, and Scale AI. I’ve learned to separate the signal from the noise. And my initial reaction to most AI in healthcare pitches? Pure noise.
Most founders in this space are brilliant data scientists. They can build elegant models that look impressive on a slide deck. But they have no idea what it takes to bring a product to market in the brutal, regulated world of healthcare. They think the algorithm is the business. They’re wrong. And I’ve seen the wreckage of too many startups that learned this lesson the hard way. I was convinced that for all the talk, we were still years away from AI making a real dent in drug discovery. I was wrong.
The Graveyard of Healthcare AI Startups
Let me be blunt. The reason most healthcare AI companies fail is not because of the tech. It’s because of a fundamental misunderstanding of the market. I’ve seen this pattern repeat itself endlessly. A team of brilliant PhDs from Stanford or MIT will spend two years building a model with 99.5% accuracy on some curated dataset. They raise a seed round, get some press, and then… nothing. They hit a brick wall.
Why? Because they never stopped to ask the right questions. Who is the user? Is it the doctor? The hospital administrator? The insurance company? How does this tool fit into the existing clinical workflow? A doctor isn’t going to log into a separate, clunky interface to get a prediction, no matter how accurate it is. They live in the EHR. If you’re not integrated, you’re invisible.
And then there’s the data problem. Everyone talks about "big data," but in healthcare, the data is a mess. It’s fragmented, siloed, and full of errors. The pristine datasets used to train models in academia bear zero resemblance to the chaotic reality of a hospital’s IT system. I once invested in a promising AI diagnostics company. They had a fantastic algorithm for detecting early-stage cancers from radiology scans. But they spent 18 months and nearly all of their seed funding just trying to get clean data from a single hospital system. By the time they were ready to go to market, a competitor had already locked in a distribution deal with the largest EHR provider. Game over.
This is the uncomfortable truth. In healthcare, distribution and integration are everything. A mediocre model with a seamless workflow will beat a brilliant model with a clunky one every single time. Most founders learn this too late.
I remember a pitch from a company that had a model to predict sepsis in ICUs. The accuracy was off the charts. But their 'product' was a standalone dashboard. A nurse in the ICU is already juggling a dozen different screens and alarms. The idea that they would voluntarily open another one was absurd. The company burned through $5 million in two years and never signed a single hospital. They were a solution in search of a problem.
The Platform That Changed My Mind
So, with all that scar tissue, you can understand my skepticism when I was introduced to the team at "Catalyst Bio." The founder wasn't a fresh-faced coder, but a 20-year veteran of Pfizer who had lived the pain of drug discovery firsthand. He didn't start by talking about the model's architecture. He started by showing me a demo of their platform integrated directly into the research workflows of a major pharmaceutical company. It wasn't a standalone tool; it was a layer of intelligence that sat on top of the systems scientists were already using.
They didn't try to boil the ocean. They focused on one specific, painful part of the process: target identification. This is the very first step in drug discovery, where scientists have to sift through mountains of genomic and proteomic data to find a single protein or gene that a new drug could target. It’s a process that can take years and cost millions, with a staggering failure rate. Catalyst Bio's platform didn't just predict potential targets; it explained why they were promising, linking them to specific biological pathways and existing research. It was a glass box, not a black one.
What really got me was their data strategy. Instead of trying to clean up messy real-world data, they built a system that could learn from it. They used a clever combination of transfer learning and reinforcement learning to build models that could handle the noise and inconsistencies inherent in biological data. They had a real-time feedback loop from the scientists using the platform, which meant the models were constantly getting smarter. It was the first time I’d seen a company truly crack the data problem in a way that scaled.
I grilled the founder for two hours. I threw every objection I had at him. What about regulatory hurdles? They had a former FDA official on their advisory board. What about the sales cycle? They were already running paid pilots with two of the top five pharma companies. They had done their homework. They weren't just building an algorithm; they were building a business.
The 10x Difference: From Years to Weeks
Talk is cheap. Results are what matter. And the results I saw from Catalyst Bio were unlike anything I had ever encountered. One of their pharma partners had been stuck on a particularly nasty type of cancer for three years. They had a team of 50 scientists and had already sunk over $100 million into the project with zero viable drug targets to show for it. They were about to kill the program.
As a last-ditch effort, they gave Catalyst Bio access to their data. Within six weeks, the platform had identified three novel targets that had never been considered before. Not only that, it also predicted which existing, FDA-approved drugs could be repurposed to hit those targets. This wasn't just a theoretical exercise. The pharma company immediately took the top candidate into preclinical testing. Nine months later, that drug is now in Phase 1 clinical trials. A process that normally takes 3-5 years was compressed into less than one.
This is not a one-off success story. I spoke to another one of their clients, a mid-sized biotech firm. They used the platform to screen a library of 10,000 compounds for a rare neurological disorder. The process, which would have taken them over a year and cost a fortune in lab resources, was completed in two weeks. The platform identified a lead compound that is now showing incredible promise in animal models. The CEO told me, "This didn't just speed up our research; it fundamentally changed the economics of our business."
This is what 10x looks like. It’s not a 10% improvement. It’s a complete paradigm shift. Think about the downstream effects. That pharma company didn't just save time; they saved hundreds of millions in development costs that would have been wasted on dead-end targets. They also gained a massive competitive advantage by getting a promising drug into trials years ahead of schedule. For the biotech firm, the calculus was even more stark. As a smaller player, a single failed program can be an extinction-level event. By de-risking their pipeline so dramatically, they could suddenly take on more ambitious projects and attract partnership offers that were previously out of reach. It’s the difference between a company surviving and a company thriving. It’s the difference between a patient getting a life-saving drug in five years versus ten. This is the real, tangible impact of AI in healthcare, and it’s happening right now.
My New Thesis: Invest in the Shovels
My experience with Catalyst Bio forced me to completely re-evaluate my investment thesis for AI in healthcare. I’m no longer investing in companies that are just building a better mousetrap—a slightly more accurate algorithm. I’m investing in the companies that are building the factories. The platforms. The shovels for the gold rush.
The future of AI in drug discovery isn’t about a single, magical algorithm that will cure all diseases. It’s about creating systems that empower scientists, augment their intelligence, and slash the time and cost of R&D. It’s about building businesses that understand the messy reality of the healthcare industry and have a clear strategy for navigating it.
I put my money where my mouth is. I led Catalyst Bio’s Series A round. It was one of the easiest investment decisions I’ve made in years. The way I see it, we are at an inflection point. For the past decade, we’ve been talking about the potential of AI in healthcare. Now, we are finally seeing that potential translate into reality. The companies that will win in this new era are not the ones with the fanciest algorithms, but the ones with the deepest understanding of the problems that keep pharma executives up at night. They are the ones who are not just building tech, but building the future of medicine itself. And I’m betting big on them.
Frequently Asked Questions
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.