Most AI tutors are a joke. There, I said it.
They’re glorified flashcard apps, digital drills that feel about as personal as a checkout kiosk. They test memory, not understanding. They can’t tell if a student is bored, frustrated, or on the verge of a breakthrough. They have no soul.
People keep asking me how our new AI tutor feels so… different. So human. The secret isn’t just a clever algorithm or a massive dataset. It’s a completely different philosophy. We didn’t set out to build a better testing machine. We set out to build a better teacher.
This is the story of how we did it.
The Ghost in the Machine
My journey here started with a simple observation from my time building RemoteTeam, which was later acquired by Gusto. We were creating tools to help companies manage distributed teams. The biggest challenge wasn’t payroll or time-off tracking; it was connection. It was making people feel seen and understood from thousands of miles away. A manager who can’t read the room—even a virtual one—is a bad manager.
I saw the same exact problem in education. We have all this technology, but we’ve been using it to build walls, not bridges. We track clicks and completion rates, but we ignore the human signals: the slight hesitation before an answer, the quickening pace that signals excitement, the long pause that screams, “I’m lost.”
I’ve been fortunate to invest in over 200 companies, including some of the foundational players in AI like Anthropic and OpenAI. I’ve seen what this technology is capable of. So the question that kept me up at night was simple: why was nobody applying it to build a truly empathetic learning tool? One that adapts not just to what a student knows, but to how they feel.
We decided to build that tool. The goal was ambitious: create an AI tutor that could sense a student's emotional and cognitive state and adapt its teaching style in real-time. It was a crazy idea. And it was much, much harder than we thought.
More Art Than Science
Building a truly adaptive AI is not just a coding problem. Your first instinct is to throw data at it. We started there. We gathered terabytes of anonymized interaction data: keystrokes, mouse movements, video feeds of facial expressions, even the tone of voice in spoken answers.
Our first models were a disaster. They were technically brilliant but emotionally tone-deaf. The system could tell a student got an answer wrong, but it couldn’t distinguish between a careless mistake and a deep conceptual misunderstanding. It would jump to a simpler problem when what the student really needed was a different explanation.
I remember one specific moment of failure. We were testing a module for learning algebra. The AI detected a student was struggling with factoring polynomials and immediately served up a basic multiplication problem. On the video feed, we saw the student—a bright 15-year-old—slam her laptop shut in frustration. The AI’s response felt condescending. It had insulted her intelligence.
That’s when we realized the data wasn’t enough. We needed a framework. We needed to teach the AI not just what to look for, but why.
We brought in educational psychologists and veteran teachers. They didn’t talk about code; they talked about kids. They told us stories. They explained that a student rubbing their eyes might be tired, but they could also be concentrating intensely. A student leaning back in their chair might be disengaging, or they might be pondering a complex idea. Context is everything.
This was our breakthrough. We started building a multi-layered system:
- Cognitive Layer: This is the easy part. It tracks right/wrong answers, speed, and consistency. Standard stuff.
- Behavioral Layer: This layer analyzes interaction patterns. Is the student jumping between questions? Are they re-reading the same paragraph over and over? This is where we started to see patterns the old teachers knew by heart.
- Emotional Layer: This was the moonshot. Using inputs from the camera and microphone, we trained models to recognize subtle cues for frustration, confusion, boredom, and excitement. This wasn't about being invasive; it was about being attentive, the way a great teacher is.
The real magic happened when we fused these layers together. The AI learned that a wrong answer plus a furrowed brow and a long pause meant “I’m deeply confused.” The prescription? Not a simpler problem, but a new way of explaining the concept, maybe with a visual analogy. A right answer plus a quick, confident click meant “I’ve got this.” The prescription? Move on to a more challenging topic to keep the student engaged.
From Algorithm to Aha!
Let me give you a concrete example. We call it the “Socratic Bridge.”
One of our core principles is to never just give the answer. When the system detects a student is stuck, it doesn’t retreat. It leans in. Instead of saying, “The answer is X,” it asks a leading question. “That’s close. Have you considered the relationship between the hypotenuse and the other two sides?”
This is computationally expensive. It requires the AI to understand the student’s mistake, generate a relevant question in natural language, and then parse the student’s response to see if they’ve grasped the hint. But the payoff is immense. You can literally see the “aha!” moment on a student’s face when they make the connection themselves. That’s not just learning; it’s confidence. It’s empowerment.
We had to build our own custom architecture for this. It’s a hybrid model that combines the pattern-matching power of deep learning with a rules-based engine grounded in pedagogical theory. Think of it as an expert teacher’s intuition encoded into software. It’s not perfect, but it’s a world away from the brute-force approach of most ed-tech.
After my first company, MovieLaLa, was acquired, I learned that connecting with an audience was about creating a real experience. The same is true here. This isn’t about a transaction of information. It’s about creating an experience of discovery.
The Future is Not a Test
We’re just getting started. The platform is constantly learning, not just about individual students, but about learning itself. Every interaction, every moment of frustration and every breakthrough, makes the entire system smarter.
Building this has reinforced a belief I’ve had for a long time, one that I wrote about in my book, Becoming Top 1%. Success isn’t about having all the answers. It’s about asking the right questions and having the resilience to work through the messy process of finding the solution. It’s about adapting.
Our AI tutor doesn’t just teach subjects; it models that process. It shows students that it’s okay to be stuck, that confusion is part of learning, and that the right kind of support can turn a wall into a doorway.
This is the future of personalized learning. It’s not a dystopian world of robot teachers. It’s a world where technology finally disappears into the background, becoming a seamless, supportive partner in the deeply human journey of education. We’re building the tool we all wish we had in school—a guide who is patient, perceptive, and always knows how to help you take the next step.
Frequently Asked Questions
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.
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.