''' We were about to go bankrupt.
That’s not a metaphor. We had three weeks of cash left in the bank. My co-founder and I were staring at a term sheet for an acqui-hire that felt more like a surrender. All our work, all the late nights, all the belief we had in our "revolutionary" AI model—it was all about to be sold for parts.
Most founders in Silicon Valley don’t like to talk about these moments. They’ll post the funding announcements, the celebratory exit photos, but they won’t tell you about the gut-wrenching terror of watching your dream die. I’m telling you because the story of how we pulled back from that brink is more valuable than any success story you’ll read on TechCrunch.
It’s the story of how we abandoned our original idea and stumbled into a $10 million niche in drug discovery. It’s the uncomfortable truth about what it really takes to build a successful AI company in the brutal, heavily regulated world of healthcare.
The Seductive Lie of the "Better Model"
Our first idea was brilliant. On paper, at least. We were building a diagnostic tool for radiologists. We had a team of incredible engineers, access to a massive dataset, and an algorithm that was, by every academic metric, state-of-the-art. It could detect certain anomalies in medical images with a 2% higher accuracy than the leading solutions. Two percent! In machine learning, that’s a huge win.
We thought that was enough. We believed that if we just built the best model, the market would beat a path to our door. Hospitals would be lining up. Doctors would hail us as heroes.
We were dead wrong.
We spent a year and a half trying to sell it. The feedback was always the same. "It's interesting," they'd say. "But integrating a new AI tool into our workflow is a nightmare." Or, "Our current system is good enough. A 2% improvement doesn’t justify the cost and risk of switching." Or the killer: "Your model isn't FDA-cleared, and we have no idea how to even start that process with you."
Every "no" felt like a punch to the gut. We had the better technology, but we couldn’t sell it. The market we thought was desperate for innovation was actually resistant to it. The problem wasn’t the tech; it was everything around the tech. Workflow integration. Sales cycles that lasted for years. The terrifying, opaque beast of government regulation.
Building a great AI model isn’t the hard part. The hard part is getting someone to actually use it in a high-stakes environment like healthcare.
The Pivot That Saved Us
The turning point didn’t come in a flash of genius. It came out of desperation. We were running out of money and options. We started taking any meeting we could get, talking to anyone in the healthcare space who would listen.
One of those meetings was with a mid-level research scientist at a pharmaceutical company. It was a total Hail Mary. We weren’t even trying to sell him our product anymore; we were just trying to understand the industry, looking for a lifeline.
We showed him our tech, and he was polite but unimpressed. Then, on a whim, my co-founder mentioned a small, internal tool we had built to manage and analyze our own training data. It was a simple interface for sorting and filtering different types of cellular images we used for our radiology model.
He stopped us. "Wait. Can you go back to that?"
He leaned in, his eyes lit up. "You can sort by cell morphology? And cross-reference that with the staining protocol we used?"
We said, "Yeah, it was just a weekend project to help us clean up our data. It’s nothing fancy."
He looked at us like we had just discovered fire. "Nothing fancy? We have a team of ten PhDs who spend six months doing just that, and they do it in Excel. It’s slow, it’s full of errors, and it’s the single biggest bottleneck in our preclinical drug discovery process. Can we buy this? Like, right now?"
That was the moment. We had spent two years building a complex, "sexy" AI for diagnosing patients, and no one wanted it. We had spent two days building a simple internal tool, and a multi-billion dollar company wanted to buy it on the spot.
The $10 Million Niche We Almost Missed
We had stumbled backward into a massive, unglamorous, and incredibly lucrative problem. The world of preclinical drug discovery is a mess of manual processes, disconnected data, and overworked scientists. They aren’t trying to get a 2% improvement in diagnostic accuracy; they’re trying to figure out if a billion-dollar drug candidate is a dud before they spend a decade and a fortune on human trials.
Our "simple" tool solved a massive pain point for them. It automated a tedious, error-prone part of their workflow that was costing them millions.
So we pivoted. Hard.
We threw out our entire roadmap. We stopped talking about our fancy radiology AI. We took that internal tool and transformed it into a robust platform for preclinical research. We focused on one thing and one thing only: helping pharma scientists make better decisions, faster.
Within six months, we had our first paying customer—that same pharmaceutical company. Within a year, we had three more. We weren’t just selling a model; we were selling a solution to a painful, expensive business problem. We didn’t need to go through the years-long FDA approval process because we were operating in the preclinical space, a completely different regulatory world.
Our niche wasn’t radiology or diagnostics. It was workflow automation for drug discovery labs. It wasn’t as glamorous, but it was a real business. A business that quickly grew to over $10 million in annual recurring revenue.
The Hard-Won Lessons
I’ve invested in over 200 companies, including AI leaders like Anthropic, OpenAI, and Scale AI. I’ve seen this pattern play out again and again. The most successful companies aren’t always the ones with the most groundbreaking algorithm. They’re the ones that do the dirty work of solving a real-world problem.
Here’s what I wish I had known from the start:
Your Model is a Commodity. This is a tough pill to swallow for technical founders, but it’s true. Unless you are OpenAI, your model is not your moat. Someone else will always be building a model that’s slightly faster or more accurate. Your real defensibility comes from your data, your distribution, and your deep understanding of your customer’s workflow.
Fall in Love with the Problem, Not Your Solution. We were so in love with our "elegant" AI that we couldn’t see the market was telling us it didn’t want it. We almost went bankrupt because of our own egos. The moment we started listening to the customer’s actual problem, we found our business.
Regulation is Not an Obstacle; It’s a Moat. We were terrified of the FDA. We saw it as a giant wall we couldn’t climb. But in the right context, regulation is a competitive advantage. The reason our niche was so underserved was precisely because the compliance and validation requirements scared other startups away. By embracing the complexity of the preclinical world, we built a business that was very difficult for others to copy.
Sell Shovels in a Gold Rush. Everyone wants to find the cure for cancer. That’s the "gold." It’s a noble goal, but it’s also incredibly crowded and high-risk. We found our success by selling the "shovels"—the tools that enabled the researchers who were looking for the gold. It’s a less glorious but far more predictable business model.
We didn’t set out to build a workflow tool for drug discovery. We set out to revolutionize radiology. But the market has a way of humbling you. It forces you to confront the difference between a cool piece of technology and a valuable company.
Our near-death experience was the best thing that ever happened to us. It stripped away our illusions and forced us to find a real problem worth solving. And for founders in the AI space today, that’s the only lesson that matters. '''
Frequently Asked Questions
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
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.
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.