I spent a chunk of last year building Walnut, an AI classroom platform, and at some point a demo turned into a pitch — including a room at Oraseya Capital in Dubai. I’d written plenty of code by then. What I hadn’t done was defend the idea to people whose job is to find the hole in it. That turned out to be the most useful product review I’ve ever had.
Here are the things that stuck.
A demo answers “does it work.” A pitch answers “should it exist.”
I came in ready to show features. Dashboards, grading, the tutoring engine — all the stuff I was proud of having built. The questions I got back were almost never about whether it worked. They were about why anyone would switch, who pays, and what happens when OpenAI ships the obvious version of this next quarter.
I didn’t have crisp answers, and the lack of them was more informative than any bug report. You can build something that works perfectly and still have no reason for it to exist.
”It uses AI” is not a wedge
Everyone in the room had already seen a dozen “AI for education” decks. The moment Walnut became interesting was when I stopped talking about the model and started talking about the constraint: a tutor grounded in the specific course, that refuses to just hand over answers. The technology was the boring part. The opinion about how it should behave was the product.
Talking to a skeptic is a compiler for your thinking
There’s a specific kind of clarity that only shows up when someone smart is mildly unconvinced and waiting. Half-formed assumptions I’d been carrying for months either snapped into a sentence or fell apart on contact. I rewrote a good portion of Walnut’s roadmap on the flight back — not because the code was wrong, but because the reasons finally were.
What I’d tell past me
Build the thing, yes. But find a skeptical, informed person and try to convince them long before you think you’re ready. The gaps they expose are the real backlog. Code is cheap to change; a wrong premise is expensive, and you usually can’t see your own.
More on the technical side of Walnut — the retrieval and guardrail layer — in a future post.