There were two overarching lessons I took from my experience.
The first is that a successful educational product needs to deeply and intimately understand pedagogy and materials -- teaching and learning techniques and how they vary between students, and what kinds of materials are suitable -- deeply, I mean like at a postdoc level.
I've seen so many products fail because the engineers and designers didn't have the slightest idea how students actually learn, think that algorithms could make up for crappy materials, or have zero understanding of what teachers actually do.
And creating good teaching materials with good pedagogy is very expensive. You have to hire educational+domain experts. They don't come cheap. There just aren't that many people who can clearly explain complex concepts gradually, relatably, and accurately in a simple manner.
For some reason, because we all went to school, so many of us think we understand teaching and learning. It's pretty humbling to discover you don't at all.
The second lesson is that education software isn't sold to the kind of market you're probably familiar with. It's not bought by consumers, or even the kinds of enterprises you might be familiar with. It's bought by school districts (or universities), each of whom have incredibly extensive needs for customization -- for interoperability, for the curriculum, for whatever -- and each of whom expect to be given lots of individual attention for a long time before they sign.
Your company will be primarily salespeople, not engineers.
But because there is a ton of money in EDU, you actually only need a handful of school districts or universities to sustain your company. Which then turns into a problem, because it means everyone is using different products, and so much is custom-built. So standards and interoperability don't emerge -- or when they do, they're just, well, really bad.
It's a complicated space. The only other space that feels similarly insanely complicated to me is healthcare.