For the first five years of my journey into AI, I was a model student. I read all the books, I followed the gurus on Twitter, and I built products exactly the way they told me to. We chased state-of-the-art models, obsessed over accuracy benchmarks, and presented demos that made VCs’ eyes light up. And for five years, I was basically just lighting money on fire.
I was building beautiful, powerful, and utterly useless machines.
One of the first products we ever built was a "semantic-aware news aggregator." The pitch was slick. We used cutting-edge NLP to understand the nuance of your interests and deliver a perfectly curated news feed. We spent a year and a half on it. The tech was incredible. We could map concepts and entities in ways that made my computer science-loving heart sing. We launched, and the world responded with a deafening silence. A few tech-savvy friends said it was "cool," but nobody actually used it. Why? Because nobody wakes up in the morning thinking, "My biggest problem today is that my news feed lacks semantic awareness." People just want to know what’s going on, and their existing tools were good enough.
I honestly had no idea what I was doing. I was following the playbook, but the playbook was wrong. It was a playbook written by researchers and academics, not by people who have to make payroll.
The Obsession That Led to the Truth
I couldn't accept that this was just the cost of doing business in AI. So I went looking for answers, not in another blog post, but in the graveyard of dead AI startups. I became obsessed. I spent months digging through old press releases, post-mortem blog posts, and financial filings. I compiled a list of over 1,000 AI products that had raised significant funding and then vanished. I wanted to find the pattern.
And I found it. It was a shocking, uncomfortable truth that went against everything I’d been taught.
The pattern was this: The products that failed all started with the AI. The products that succeeded all started with a boring, expensive, or painful human problem.
That’s it. That’s the secret. The successful founders didn’t sell "AI." They sold a solution. The AI was just an implementation detail. It was the engine under the hood, but they were selling the car. We, on the other hand, had been trying to sell people a fancy engine, and then wondering why they wouldn’t buy it when they just needed to get to work.
From Selling AI to Solving Problems
This realization changed everything for me. It was the guiding principle behind my next company, RemoteTeam. When we started it, the goal wasn't to "build an AI for HR." The goal was to solve the massive, tangled, and expensive nightmare of managing a global workforce. We talked to hundreds of companies. We didn’t ask them, "What AI tools do you wish you had?" We asked them, "What’s the most annoying part of your day? What keeps you up at night?"
They told us about the agony of calculating payroll across different currencies and tax laws. They complained about the soul-crushing task of approving expense reports. They talked about the difficulty of tracking time off for a team spread across a dozen time zones.
Pain. Boring, expensive, human pain.
Only then did we look at AI. We didn’t try to build one master "HR AI." We built tiny, focused, "boring" AI models that attacked these specific pain points. We used a simple OCR model to scan receipts and auto-fill expense reports. We used a forecasting model to help companies predict their cash flow needs for global payroll. We sold "one-click global payroll," not "a significant, AI-powered human capital management platform." The AI was our secret weapon, not the product itself. And it worked. RemoteTeam was eventually acquired by Gusto because we solved a real, costly problem.
My investment strategy in companies like Scale AI and Hugging Face follows the same logic. They aren’t just building cool tech; they are building the essential tools that solve the painful, foundational problems for thousands of other companies trying to build their own AI solutions.
A New Playbook for Building in AI
So, if you're a founder or a product manager in the AI space, I beg you, throw out the old playbook. Here’s a new one, based on what I learned from my own failures and the failures of a thousand others.
1. Find the Human Pain First
Forget about AI for a month. Seriously. Lock the term in a box. Instead, go find a problem that people complain about constantly. Look for a process inside a business that is slow, manual, and expensive. Look for the overflowing inbox, the spreadsheet with 100 tabs, the daily task that everyone on the team dreads. That’s where you’ll find gold.
2. Quantify the Pain
Don’t just say it’s "annoying." Put a number on it. How many hours per week does this task waste for the average employee? How much money does that translate to in salary? What is the cost of the errors that the manual process introduces? A problem with a clear dollar sign attached to it is a problem a business will pay to solve.
3. Map the Manual Solution
Before you write a single line of code, could you solve this problem for one customer, manually? What are the exact steps? Be the "human in the loop" yourself. This forces you to understand the problem at a granular level that you’d otherwise miss. You’ll discover edge cases and complexities that you can’t see from a 30,000-foot view.
4. Identify the Automation Target
Now, look at your manual workflow. Where is the biggest bottleneck? Where does the process slow to a crawl? Is it a human making a repetitive decision? Is it tedious data entry? That specific, narrow point is your target for AI. Not the whole process, just the most painful step.
5. Build the Simplest "Boring" AI
Your goal is not to build an artificial general intelligence. Your goal is to build a tool that is just slightly better, faster, or cheaper than the manual process it’s replacing. If a human is 80% accurate at a task, and your model is 85% accurate, that’s a win. It doesn’t need to be perfect. It just needs to be better. This is the core lesson I try to impart when I talk about my book, "Becoming Top 1%".
Look, I get it. This approach is less glamorous. It doesn’t sound as good in a headline as "We’re Building a Sentient AI to Revolutionize Everything." But it’s how you build products that people actually buy. Stop trying to sell the engine. Find someone who is walking, and sell them a car.
Frequently Asked Questions
How long did it take to see results?
Most meaningful business results take 3-6 months to materialize. Anyone promising overnight success is selling something. The companies in my portfolio that grew fastest were the ones that stayed patient and consistent.
What was the biggest challenge in this case?
Almost always, the biggest challenge is people and alignment, not technology or strategy. Getting the right team focused on the right problem is harder than any technical challenge I've encountered.
Can these results be replicated?
The specific numbers will vary, but the underlying patterns and principles are transferable. The key is understanding the context behind the results, not just copying the tactics. Every company has unique constraints that shape what works.