Five years. That’s how long I spent trying to force an idea to work. Five years of my life, millions in funding, and countless sleepless nights, all culminating in a spectacular failure. We were building an EdTech platform, let’s call it ‘LearnSphere,’ and I was convinced that artificial intelligence was the key to unlocking personalized education for every student on the planet. I was wrong. Dead wrong.
Everyone talks about AI in education as the future, but nobody talks about the brutal realities of building it. My first startup was a testament to that failure. Here’s what I wish I knew before I started.
Lesson 1: AI Can't Fix a Broken Pedagogy
Our grand vision for LearnSphere was an AI tutor that adapted to every student's learning style in real-time. We spent the first two years and over a million dollars building a recommendation engine from scratch. It was a technical marvel. It could track every click, every answer, every moment of hesitation, and serve up the ‘perfect’ next piece of content. We had data scientists, PhDs in machine learning, the whole nine yards.
But here was the problem: the content itself was based on a flawed teaching methodology. We were essentially using a super-intelligent system to more efficiently deliver a bad lesson plan. We digitized standard, boring textbook content and expected the AI to magically make it engaging. It didn’t work.
Students would log in, get a few AI-powered recommendations, and then zone out. Engagement was abysmal. We had a 90% drop-off rate after the first week. We tried tweaking the algorithm, adding more data points, and refining the models. Nothing helped. We were so focused on the ‘AI’ part of ‘AI in Education’ that we forgot the ‘Education’ part is what actually matters. A state-of-the-art engine in a car with no wheels goes nowhere. Our pedagogy had no wheels.
Lesson 2: Data Is Useless Without the Right Context
As an engineer and data-driven founder, I have a bias: more data is always better. At LearnSphere, we collected everything. I mean everything. We had terabytes of data on student performance. We could tell you that 73% of students in Ohio struggled with a specific type of algebra problem on Tuesday afternoons. Impressive, right?
It was also completely useless. We were drowning in data but starved for wisdom. The numbers told us what was happening, but they gave us zero insight into why. Why were students struggling? Were they tired? Was the content confusing? Was the user interface clunky? Our dashboards were a sea of charts and graphs, but they were just vanity metrics.
I remember a pivotal board meeting where I proudly presented a deck showing a 15% improvement in ‘session duration’ after an algorithm update. One of our investors, a former teacher, asked me a simple question: “Are the kids actually learning more, or are they just taking longer to get confused?” I had no answer. That question hit me like a ton of bricks. We had optimized for engagement metrics, not learning outcomes. We were tracking the wrong things.
Real context doesn’t come from a server log. It comes from talking to your users. It comes from sitting in a classroom and watching a teacher struggle with your software. It comes from seeing a student’s face light up when they finally understand a concept, not when your AI serves them another video.
Lesson 3: Teachers Are Your Customers, Not an Obstacle
This was our most arrogant and fatal mistake. We saw teachers as a hurdle to overcome. Our pitch was, “Let our AI do the teaching, freeing up teachers to be ‘facilitators’.” We fundamentally misunderstood their role. We were trying to replace them, and they knew it.
We built this beautiful, complex product for students, assuming schools would just buy it and force teachers to use it. The reality was the opposite. Teachers are the gatekeepers, the champions, and the ultimate users of any tool in the classroom. If they don’t like it, it’s dead on arrival. And they hated LearnSphere.
It was too complicated. It didn’t fit into their existing workflows. It generated reports they didn’t have time to read. We asked them to trust an algorithm they didn’t understand to make decisions about their students. We were asking them to give up control, and we offered them nothing but more work in return.
I spent months trying to sell to school districts, getting top-down mandates. We even got a few pilot programs. But in every single one, teacher adoption was near zero. They would log in once, get overwhelmed, and go right back to the tools they knew and trusted. The pilots all failed.
The Turnaround That Came From Failure
Shutting down LearnSphere was one of the hardest things I’ve ever done. It felt like a personal failure on every level. But those lessons were the most valuable education I ever received. They became the foundation for my next venture.
Instead of starting with AI, we started with a problem. We spent 6 months just talking to teachers. We didn’t write a single line of code. We found a simple, painful, and universal problem: grading. Teachers were spending 10-15 hours a week on it.
Our new product didn’t have a fancy AI engine. It was a simple tool that helped teachers create, distribute, and grade assignments faster. We built it with them, getting feedback every single day. Once we solved that core problem and earned their trust, we began to layer in intelligent features. Not a black-box AI, but simple, explainable tools that gave them superpowers—like automatically grouping similar mistakes so they could address them with the whole class at once.
This time, it worked. Teachers loved it because we built it for them. We solved their problem first. The AI was in service of the user, not the other way around.
My journey in EdTech taught me that AI is not a product. It’s a tool. And like any tool, it’s only as good as the person using it and the problem it’s meant to solve. Stop chasing the AI hype. Go find a real, human problem and solve that instead. That’s where you’ll find success.
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.
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.