Why I open sourced my models

I really enjoy teaching and sharing. That's a lot of why I do this. But I've watched a real shift happen this year: people ask ChatGPT and Claude instead of asking other people.

I do it too. If I need to understand a term, work through a calculation, or get started on something unfamiliar, asking an AI is often just the fastest way to get into it. I used to search, or dig through a doc, or ask someone who'd know. Now I open a chat window first, most of the time.

It's made me reconsider what I create and share. If someone can get an explanation in the tool they're already using, what would make them come to something I've written? What would they actually want to use, given that the explaining part is basically solved for them already?

This is one reason I decided to open source my spreadsheet models. It wasn't an easy call: I'd sold more than a million dollars of them, and that number was sitting in my head the whole time I was deciding whether to give the structure away. Fewer people seem to want a template as their starting point. They want to ask an AI to build a model for their particular situation: their round, their revenue mix, their specific cap table and SAFE stack. I understand the appeal, though I think starting with a proven structure is more valuable than ever. It's easy to generate a spreadsheet. It's still hard to make good decisions using it.

We're seeing a similar shift with software. People want tools they can adapt to their own work, and increasingly they want to use them alongside the AI tools they already have open. Nobody wants to leave their chat window to go fill in someone else's rigid form.

What I keep coming back to, though, is that giving away the structure only solves half the problem. Anyone can generate a spreadsheet now, AI or no AI. The harder part, the part I don't think a chat window replaces, is knowing whether the number in front of you is one you should actually act on: whether that runway figure means you raise now or in three months, whether that waterfall means the SAFE holders are fine with the terms or quietly underwater. That's not really a modeling question, it's a decision question. AI is still best when we use it to aid our judgement, not replace it.

So I'm trying to share more of the underlying work: the models, the code, the decisions behind them, and what I've learned from using them. People want to build their own things. I want to help them make great decisions using them.