I once burned through $500,000 and six months of my team’s life building an AI feature that nobody wanted.
It was a painful, expensive lesson in what not to do. We were building my second company, MovieLaLa, and thought we had a world-changing idea. We would use a sophisticated neural network to predict what movies a user would want to watch next. It was technically brilliant. It was also a complete failure. The experience taught me a lesson that now informs every investment I make and every piece of advice I give: most founders are thinking about AI feature prioritization completely backwards.
They get seduced by the technology. They read about the latest models from OpenAI or Google and immediately start brainstorming ways to cram that tech into their product. This is the single biggest mistake I see, and it’s a recipe for disaster. It leads to bloated products, wasted resources, and solutions looking for a problem.
The Tech-First Fallacy
So many founders fall into what I call the "Tech-First Fallacy." They start with the AI, not the user. Their thought process goes something like this: "Wow, this new language model is incredible! What can we build with it?" They get excited about the possibility of building something "groundbreaking" and lose sight of the only thing that actually matters: solving a real, painful problem for a customer.
This approach is a trap. You end up building features that are technically impressive but practically useless. You create a sophisticated AI-powered analytics dashboard, but your users just want a simple CSV export. You design a complex AI chatbot to handle support, but your customers just want to talk to a human.
I’ve seen this happen more times than I can count. A portfolio company of mine—I won’t name them—spent almost a year trying to build an AI that could write entire marketing campaigns from a single prompt. The tech was amazing, but the results were always just a little off. The tone wasn't quite right, the ideas were generic. Their customers, who were professional marketers, didn't want a machine to do their job. They wanted a tool to make their job easier. They wanted an AI that could generate 50 headline variations in 10 seconds, not one that tried to replace their strategic thinking. The company eventually pivoted, but only after wasting a huge amount of time and money chasing a technical fantasy.
My Big Failure at MovieLaLa
My own failure was even more direct. At MovieLaLa, we were competing with giants. We thought our edge would be superior AI-driven recommendations. We hired PhDs and built a recommendation engine that took into account hundreds of signals. It was a masterpiece of engineering. We celebrated the day it went live.
And then… nothing.
Usage metrics didn't budge. We talked to our users. It turned out they didn't want a machine to tell them what to watch. They wanted to see what their friends were watching. The social proof from their own network was infinitely more powerful than our black-box algorithm. Our fancy AI was solving a problem that didn't exist for our users. We had built a solution in search of a problem.
That product eventually failed, and Gfycat acquired us for the tech and team, not the user base. It was a tough pill to swallow. We had the right technology but the wrong philosophy. We were so focused on building impressive AI that we forgot to ask our users what they actually wanted. We could have saved ourselves half a million dollars and months of work with a few simple conversations. Instead of starting with the user's problem, we started with our own technical ambition.
A Better Way: The Problem-Backwards Approach
After that experience, I completely changed my approach. I now use a simple, counterintuitive framework for AI feature prioritization. I call it the "Problem-Backwards Approach." It has four simple steps.
Step 1: Forget AI Exists
Seriously. For a moment, pretend AI hasn't been invented. Look at your user. What is the most painful, tedious, or expensive problem they have that your product is trying to solve? What is the core "job-to-be-done"? Write it down in plain English. No tech jargon. For example: "Our user, a freelance designer, spends 10 hours a month creating invoices and chasing payments."
Step 2: Design the 'Manual-First' Solution
How would you solve this problem for one single user, manually? If you had to be their personal assistant, what would you do? For our designer, you might create a Google Sheet template for them, and then on the first of each month, you'd personally email them a reminder to send it out. Then you'd track the payments and send follow-up emails. This process of thinking through the manual solution forces you to understand the real-world workflow and all its messy details.
Step 3: Find the 'Dumb' Heuristic
Now, can you automate that manual workflow with simple rules and heuristics? No machine learning yet. Just simple if-then logic. For our designer, you could build a simple system that automatically generates an invoice from a template when they mark a project as complete. It could then send an automated email reminder 7 days after the due date if the invoice isn't paid. This isn't AI, it's just basic automation. But for the user, it solves 80% of the problem.
Step 4: Use AI as the Final 20%
Only now, after you've solved 80% of the problem with a simple, deterministic solution, should you look for opportunities to use AI. Where is the 'dumb' heuristic failing? Where is there still manual work? For our invoicing example, maybe the problem is that designers forget to categorize their expenses correctly. The manual solution is for you to look at their receipts and categorize them. The heuristic is a simple rule-based system based on keywords. The AI solution could be a simple NLP model that reads the receipt description and automatically suggests a category.
This is where AI shines—not as the first-line solution, but as a powerful tool to handle the complex, ambiguous tasks that a simple heuristic can't. You're not building a massive, end-to-end AI system. You're surgically applying AI to the part of the problem that offers the most value.
A Success Story: RemoteTeam
I applied this exact philosophy at my next company, RemoteTeam. We were building an all-in-one platform for managing remote employees. One of the biggest pains for our customers was international payroll and compliance.
Instead of trying to build an AI to predict global hiring trends or automate employment law, we started with the manual solution. We literally had a team of people who would manually research compliance rules for each new country a customer wanted to hire in. It was slow and expensive, but it taught us exactly what the real problems were.
Then, we built the 'dumb' heuristic. We created a database of checklists and document templates for the most common countries. This wasn't AI, it was just a well-organized knowledge base. This solved the problem for the majority of our customers.
Finally, we added the AI layer. We built a tool that could scan an employment contract and flag non-standard clauses that might conflict with local laws. It didn't write the contract, it just assisted the human expert. It was a simple, focused AI feature that provided immense value. It saved our customers thousands in legal fees and gave them the confidence to hire anywhere. RemoteTeam was eventually acquired by Gusto. That focused, problem-backwards approach to AI was a key part of our success.
Stop Chasing the Hype
So my advice to founders is this: stop chasing the AI hype. Stop getting excited about the latest models and start getting obsessed with your users' problems. The biggest opportunities in AI are not in building the most complex systems. They are in finding simple, elegant ways to solve real-world problems.
Your goal isn't to build impressive technology. Your goal is to build a successful business. And the best way to do that is to start with the customer, not the code. Forget about AI, and focus on the problem. The right technology will follow. The most successful AI companies I've invested in, like Scale AI and Hugging Face, aren't just building cool tech. They are deeply, almost obsessively, focused on solving a specific, painful problem for their users. That's the real secret. It's not about the AI at all.
Frequently Asked Questions
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.
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.
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.