They tell you an MBA is the ticket to the big leagues in business. I have one. It taught me about discounted cash flow and Porter's Five Forces. But scaling an AI learning platform to 50,000 students? That taught me things no classroom ever could. That was my real MBA.
We started with a simple, almost naive, idea: use AI to create a truly personalized learning experience. We called it Cognify. We thought the tech would be the hard part, a classic engineer’s mistake. We were wrong. The hardest part, the part that kept me up at night, was dealing with the messy, unpredictable, and wonderful humans on the other side of the screen.
We hit 50,000 active users before we were acquired by a major EdTech player. Along the way, we made a mountain of mistakes, burned through cash, and pivoted so many times I got dizzy. But we learned. Here are the seven biggest lessons I took away from that journey.
1. Personalization is a Buzzword, Not a Strategy
Everyone in EdTech loves to talk about "personalization." It’s the holy grail, the magic wand that will fix education. But what does it actually mean? For most companies, it just means adaptive quizzing. The questions get harder if you get them right, and easier if you get them wrong. That’s not personalization; that’s a glorified flowchart from the 1980s.
Real personalization is about understanding a student's motivation, their learning style, and their ultimate goals. It’s about knowing why they’re learning, not just what they’re learning. We spent our first year and a million dollars in venture capital building the most complex adaptive algorithm on the market. It was a technical marvel, a symphony of Bayesian logic and neural networks. And engagement was completely flat. Our users were bored.
I remember a specific user, a 19-year-old named Maria from Brazil. She was trying to learn Python. Our algorithm kept feeding her abstract coding puzzles, and her progress stalled for weeks. We reached out to her directly. It turned out she didn't care about computer science theory; she wanted to build a simple web scraper to track prices on an e-commerce site for her family's small business. That was her why. We junked the algorithm-first approach and started over, focusing on a simple onboarding questionnaire that asked students what they wanted to build. Maria’s feedback was the catalyst. A student who wants to learn coding to build a video game needs a different path than one who wants to pass a certification exam. The tech should serve the goal, not the other way around. It sounds obvious now, but it’s a lesson most EdTech companies, drunk on the promise of AI, never learn.
2. The Algorithm is Not the Product
I’ve been fortunate to invest in some incredible AI companies, like Anthropic, Scale AI, and Hugging Face. I’ve seen what truly groundbreaking models look like from the inside. And I can tell you this: for most products, the algorithm is just a commodity. It’s a feature, not the product itself. It's the engine, but you still need to build the car.
At Cognify, we obsessed over our recommendation engine. We A/B tested a dozen different approaches, from collaborative filtering to deep learning models that cost a fortune to train and run. The impact on our core metrics was marginal, a few percentage points here and there. What actually moved the needle? We redesigned our code editor. We added a dark mode, improved the syntax highlighting, and cut the page load time by 500 milliseconds. Engagement shot up by 15% the next week. The lesson was brutal and clear.
Students don’t care if you’re using a multi-headed attention mechanism. They care if they’re learning. They care if the platform is easy to use. They care if it’s not wasting their time. We spent more time on UI/UX design and front-end performance in our third year than we did on machine learning, and our growth doubled. Don’t let your data scientists and their fancy algorithms run the company. Let the user experience lead.
3. Engagement is an Outcome, Not a Feature
"How do we make it more engaging?" I heard this question in every single product meeting. The team, fresh from reading some blog post, would inevitably suggest adding points, badges, and leaderboards. Gamification, they called it. It’s a sugar high. It creates a short-term bump in activity, but it doesn’t create long-term value. It can even be counterproductive, making learning feel like a chore you have to grind through.
Real engagement comes from intrinsic motivation. It comes from a student feeling a sense of progress and mastery. It comes from them seeing a clear connection between what they’re doing on the platform and their real-world goals. It's the feeling of 'I built that!'.
We discovered that the most powerful driver of engagement was project-based learning. Instead of just doing abstract exercises, students would work on a capstone project. For example, in our data science track, we had students analyze a real-world dataset of San Francisco crime reports to build a predictive model. They had to clean the data, visualize it, and present their findings. The projects were challenging, sometimes frustrating, but they were meaningful. The completion rates for our project-based courses were three times higher than our standard courses. Stop trying to bribe your users with points. Give them something real and meaningful to work on.
4. Teachers are Your Most Important Customers
We thought we were building a direct-to-consumer product. We were going to bypass the slow, bureaucratic school system and go straight to the students. It was a classic Silicon Valley move, full of arrogance and a complete misunderstanding of the market. And it was a huge mistake.
Students don’t exist in a vacuum. They have teachers, parents, and schools. And teachers are the gatekeepers. They’re also the ones who see firsthand what’s working and what’s not. They are your best source of feedback and your most powerful distribution channel. We initially viewed them as an obstacle.
We resisted working with schools for two years, convinced it would slow us down. The turning point came when a high school teacher in Ohio, Mrs. Davison, emailed us. She had been using Cognify with her students on her own, without any official support. She loved the platform but said it was impossible to manage 30 students without a central dashboard. It was a wake-up call. We flew out to meet her. We spent a day in her classroom, watching how she worked. We finally started building tools for teachers—dashboards to track student progress, resources for lesson planning, tools to create custom assignments. Our growth exploded. We signed our first district-wide deal six months later, and that opened the floodgates. Don’t try to replace teachers. Empower them. They are not your competition; they are your partners.
5. Data is Not Learning
We collected a staggering amount of data. Every click, every answer, every second spent on a video. We had dashboards that could track hundreds of different metrics. We thought that if we just collected enough data, we could 'solve' learning. We were drowning in data but starved for wisdom.
But data is just a proxy. It can tell you what a student is doing, but it can’t tell you what they’re thinking. It can show you correlation, but it can’t prove causation. I saw students who had 100% completion rates on all the exercises but couldn’t apply the concepts in a slightly different context. They had learned how to game the system, not the material itself. They were pattern-matching to get the green checkmark.
This is the ethical tightrope of AI in education. You have to be careful that you’re optimizing for real learning, not just for the metrics that are easy to track. We started incorporating more open-ended assessments. We built a system for peer-reviewed projects where students would grade each other's work based on a rubric. We even experimented with having students record short video explanations of their projects. It was messy, subjective, and hard to scale, but it gave us a much richer picture of student understanding. It was the only way to know if we were actually making a difference.
6. Don’t Ship Your Org Chart
Conway’s law is real and it will bite you. The law states that an organization will produce a design whose structure is a copy of the organization's communication structure. In our early days, we had a separate team for content, a separate team for the core platform, and a separate team for AI. The result? A disjointed, Frankenstein-like user experience.
The content didn’t feel integrated with the platform. The AI features felt tacked on, like they were designed by a different company. It was a product built by committees. For instance, the content team would create a video, and the platform team would just embed it on a page. The AI team would then add a quiz below it. There was no flow, no narrative. It took a painful re-org to fix it. We moved to a pod-based structure, with each pod owning a specific part of the student journey (e.g., 'First Week Experience' or 'Project Building'). Each pod had a mix of engineers, designers, and content creators. It forced people to talk to each other. It forced them to think about the entire experience.
Your user doesn’t know or care about your internal team structure. They just know if the product works. If it feels like a collection of different products glued together, you’ve failed. The re-org was the best thing we ever did for our product quality.
7. The Exit is Not the Goal
Every founder dreams of the exit. The acquisition. The big payday. I’ve been through it twice, with RemoteTeam acquired by Gusto and MovieLaLa acquired by Gfycat. It’s a validation of your hard work, a moment of intense relief and celebration. But it’s not the end of the story.
When we sold Cognify, I felt a strange mix of pride and sadness. I was proud of what we had built, but I also knew that the product would change. The acquiring company had its own priorities, its own roadmap, its own culture. A year after the acquisition, I logged into the platform. The color scheme was different, the logo was gone, and some of our core project-based features were buried three clicks deep. It was still a good product, but it wasn't our product anymore. Some of our original vision was lost.
That’s the reality of the startup lifecycle. But the real goal isn’t the exit. It’s the impact you have along the way. The 50,000 students who used Cognify to learn a new skill, to change their career, to build something new—that’s the real legacy. I still get emails from former students. One of them, a guy from India, used our platform to learn enough to get a remote job that paid him three times his previous salary. That’s the thing that stays with you long after the money is in the bank.
Building a company is a marathon of learning, not a sprint to an exit. It’s a series of hard-won lessons, paid for with stress and sleep deprivation. These are just seven of mine. Now go out and learn your own.
Frequently Asked Questions
Are these recommendations still relevant in 2026?
Absolutely. While specific tools and tactics change, the underlying principles remain consistent. I update my thinking regularly based on what I'm seeing in the market and across my portfolio companies.
Can I implement all of these at once?
I'd strongly recommend against it. Pick the 2-3 items that resonate most with your current situation and focus there. Trying to do everything simultaneously is a recipe for doing nothing well.
How do I know which items apply to my situation?
Start by honestly assessing where your biggest bottleneck is right now. The items that address that specific constraint will give you the highest return on your time and energy.