I’ve seen it a thousand times. A founder comes to me with a brilliant idea for an AI product. They’ve raised a bunch of money. They have a team of PhDs from Stanford and MIT. They’ve spent the last two years building the most sophisticated AI model the world has ever seen. And then they launch. And nothing happens.
It’s a story that plays out over and over again in Silicon Valley. And it’s a story that I’m tired of hearing. Because the truth is, building a successful AI product has very little to do with the AI itself. It’s about everything else.
I’m not a researcher. I’m a builder. I’ve built and sold two companies, RemoteTeam and MovieLaLa. I’ve invested in over 200 startups, including some of the biggest names in AI like Anthropic, OpenAI, and Scale AI. And I’ve learned a few things along the way. In this post, I’m going to share with you the playbook that I use to build and evaluate AI products. It’s not the generic advice you’ll find in blog posts. This is the real stuff. The stuff that actually works.
The Data is the Moat
What’s the most valuable company in the AI space? Is it OpenAI? Is it Google? No. It’s a company you might not have even heard of: Scale AI. I’m an investor in Scale, and I can tell you that they are one of the most important companies in the world right now. Why? Because they have the best data.
Scale AI is a data labeling company. They have a massive network of human labelers who are experts at annotating data for machine learning models. They can label images, text, audio, and video. And they do it better than anyone else in the world. This data is the fuel that powers the AI revolution. Without high-quality labeled data, even the most advanced models are useless.
This is the secret that the big AI companies don’t want you to know. They want you to think that it’s all about the models. But it’s not. It’s about the data. If you have a unique, high-quality dataset, you have a defensible moat that no one can cross. Your competitors can copy your code, they can even copy your model architecture. But they can’t copy your data.
So, before you write a single line of code, before you even think about what model to use, you need to figure out your data strategy. What unique dataset can you create or acquire? How will you label it? How will you keep it updated? These are the hard questions that will determine the success of your AI product.
I remember when I was starting RemoteTeam, which was later acquired by Gusto. We were building a platform to help companies manage their remote teams. We had a lot of ideas for AI features we could build. We could build a tool to automatically schedule meetings across time zones. We could build a tool to analyze team communication and identify potential problems. We could build a tool to recommend the best remote work setup for each employee.
But we didn’t start with any of that. We started with a simple, non-AI product: a directory of remote-friendly companies. We spent months manually curating this directory. We researched thousands of companies. We talked to hundreds of remote workers. We built a unique dataset of remote-friendly companies that no one else had. This dataset was the foundation of our business. It’s what allowed us to build a community and attract our first users. Only after we had this unique dataset did we start to build AI features on top of it.
Start with a Slice, Not a Layer
Another mistake I see founders making is trying to build a general-purpose AI that can do everything. They want to build a platform that can be used by anyone for any purpose. This is a recipe for failure. You need to start with a very specific problem and a very specific user.
At MovieLaLa, which was acquired by Gfycat, we didn’t try to build a recommendation engine for every movie ever made. We started with a very specific niche: indie movies. We built a community around that niche and then expanded from there. We knew that indie movie fans were underserved by the big streaming platforms. They had a hard time finding new movies to watch. So we built a product just for them.
Your MVP should be a "slice" of the problem, not a "layer" on top of it. What I mean by that is you should solve a real problem for a real user, from end to end. Don’t just build a cool piece of tech and then try to figure out how to sell it. Find a user who is desperate for a solution and build it for them. Even if it’s a small number of users at first, that’s okay. You can expand from there.
I once invested in a company that was building an AI-powered writing assistant. They had a team of brilliant AI researchers. They had built a model that could generate incredibly human-like text. But they had no idea who their user was. They were trying to sell their product to everyone: students, marketers, journalists, lawyers. As a result, they weren’t selling to anyone. They were a layer on top of the problem, not a slice of it.
I advised them to pick a niche. I told them to focus on one type of user and one type of writing. They decided to focus on marketers and to help them write better ad copy. This was a much more specific problem. They were able to build a product that was tailored to the needs of marketers. They were able to build a community around their product. And they were finally able to get traction.
The "Wizard of Oz" MVP
One of my favorite techniques for building an MVP is the "Wizard of Oz" method. This is where you manually perform the tasks that your AI will eventually automate. This is a great way to validate your idea and get feedback from users without having to build a complex AI system.
I did this with one of my early startups. We were building a "chatbot" that was supposed to help people find the perfect gift. We had a fancy landing page with a chatbot interface. But behind the scenes, it was just me and a few interns. When a user typed a question into the chatbot, we would get a notification. We would quickly research the answer and then type it back to the user. It was a lot of work, but it was worth it.
This "Wizard of Oz" MVP allowed us to learn a ton about our users. We learned what kind of gifts they were looking for. We learned what questions they had. We learned what their pain points were. We used this information to build a real AI-powered chatbot that was much better than anything we could have built on our own.
This approach has a few advantages:
- It’s fast. You can get a product in front of users in days, not months.
- It’s cheap. You don’t need to hire a team of expensive AI engineers.
- You learn a ton. You get to see firsthand what your users are trying to do and what their pain points are.
Of course, you can’t do this forever. At some point, you’ll need to build the real AI. But the Wizard of Oz MVP is a great way to get started.
The Team is Everything
I’ve talked a lot about data and users and MVPs. But there’s one more thing that’s even more important than all of that: the team. At the end of the day, a startup is just a group of people working together to solve a problem. And if you don’t have the right people, you’re not going to succeed. It’s as simple as that.
When I’m evaluating a startup, the first thing I look at is the team. I want to see a team of founders who are obsessed with the problem they’re trying to solve. I want to see a team that has a deep understanding of their user. And I want to see a team that has a track record of execution. I don’t care if you went to Stanford or Harvard. I don’t care if you worked at Google or Facebook. I care if you can get stuff done.
One of the best teams I ever invested in was a company called Hugging Face. I invested in their seed round. At the time, they were just a small team of three founders. But they were obsessed with natural language processing. They were building a library of pre-trained models that made it easy for developers to build NLP applications. They had a deep understanding of their user. And they were executing like crazy. Today, Hugging Face is one of the most important companies in the AI space. They’re the GitHub of machine learning. And it’s all because of the team.
So, if you’re a founder, my advice to you is to focus on building a great team. Find people who are smarter than you. Find people who challenge you. Find people who are as passionate about your vision as you are. If you can do that, you’ll be well on your way to building a successful company.
My Final Take
Building an AI product is hard. There’s a lot of hype and a lot of noise. But if you focus on the fundamentals – the data, the user, the problem, and the team – you can build something that’s truly valuable. Don’t get distracted by the latest and greatest models. Don’t try to boil the ocean. Start small, be scrappy, and focus on creating real value for your users. If you do that, you’ll be well on your way to building a successful AI product.
And who knows, maybe I’ll even invest in it. I’m always looking for the next big thing. If you’re building something interesting, feel free to reach out to me on Twitter. My DMs are open.
Frequently Asked Questions
Do I need technical skills to build a minimum viable ai product that isn't a joke?
Not necessarily. While technical understanding helps, the most important skills are clear thinking and the ability to break problems into smaller pieces. Many successful founders I've invested in started with zero technical background and either learned enough to be dangerous or found the right technical partner.
How do I measure success with this approach?
Pick one or two metrics that directly tie to your goal and track them weekly. Vanity metrics like page views or follower counts rarely matter. Focus on metrics that reflect real engagement or revenue impact.
What are the most common mistakes when building a minimum viable ai product that isn't a joke?
The biggest mistake I see is overcomplicating things early on. Start with the simplest version that works, get real feedback, and iterate from there. Another common trap is copying what worked for someone else without understanding the context behind their decisions.
How long does it take to build a minimum viable ai product that isn't a joke?
The timeline varies depending on your starting point and resources. For most founders, expect 2-4 weeks for initial setup and 2-3 months to see meaningful results. I've seen teams move faster when they focus on one thing at a time rather than trying to do everything at once.