The AI Notetaker That Was Just Two Guys Taking Notes
Fireflies hit a billion-dollar valuation on the premise of AI — and spent most of their early life with humans doing the work.
The AI notetaker that was just two guys taking notes. Not a scandal — the mechanism worked, customers got what they paid for. The observation about AI products being humans behind the curtain became a standard due diligence question.
Fireflies.ai is an AI notetaker. It joins your meetings, records them, transcribes them, summarizes the action items. It crossed a billion-dollar valuation. It is, on paper, a triumph of applied machine learning.
For most of the time it existed before that milestone, two guys were sitting in the meetings taking notes.
This is not a scandal. I want to be clear about that. The thing they were selling — accurate meeting notes, delivered reliably — worked. Customers got what they paid for. The mechanism behind the curtain was just, occasionally, a person with a laptop and a good typing speed.
The technique has a name in research circles: Wizard of Oz prototyping. You simulate the machine with a human until you can build the machine. It is considered a legitimate and even elegant methodology. The part that gets left out of the methodology papers is what happens when the simulation is the product and you've priced it like a product and you have paying customers who believe the product is the thing you said it was.
Fireflies apparently did not linger there. They built the actual thing. The AI caught up to the humans, or surpassed them, or at least got cheap enough that the unit economics stopped requiring someone to open a laptop at 9am and type while two consultants discussed quarterly pipeline.
What I keep thinking about is the valuation. A billion dollars for a notetaker. The notetaker, for a while, was people.
There is something almost philosophical about this — the AI economy running on an invisible substrate of humans performing AI, all the way down, until at some point the substrate switches and nobody marks the transition because the output looks the same. Which is maybe the most honest thing you can say about AI in 2025: the output looks the same. The mechanism is the variable. We have decided, collectively, not to care about the mechanism.
Fireflies cared enough to eventually build it. That's more than most.
The billion-dollar question — the one nobody asked at the valuation meeting — is how many other "AI" products are still in the two-guys-in-the-meeting phase and have not gotten around to mentioning it.
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.