More or Less · Topics · Episode #171
Sam's Decade-Old AI Assistant Startup Already Learned Every Lesson Instinct Is Learning Now
Clip from the full recording, 19:08–23:07 · download mp4 · watch at this point on YouTube
Ten years before Instinct booked the wrong city, Sam's startup Fin screwed up Mark Andreessen's lunch reservation the exact same way — and Sam lays out, in detail, why human-assistant startups can never make the economics work.
Sam can’t resist jumping in on Instinct’s reservation disaster: a decade ago he and Andrew Cortina (Venmo co-founder) built Fin, a human-plus-AI concierge service — “artificial artificial intelligence” — that took requests, categorized them, and used a credentials vault and personal graph to execute anything from dinner reservations to standing weekly flower orders. It worked, people loved it — until, hilariously, Fin booked Mark Andreessen’s work lunch at the wrong place, the same exact mistake Instinct just made to Brit.
From that scar tissue, Sam distills two structural problems that still apply to every agent company today. First, you become liable for every downstream service provider the moment you promise to handle something end to end. Second, the economics of mistakes don’t work at consumer margins: a wrong restaurant is annoying, but a wrong plane ticket is financially ruinous, and unlike DoorDash, these services don’t have the margin to eat that cost.
The deeper problem, Sam argues, is psychological: a human assistant who screws up gets forgiven because of the relationship and sunk cost; a digital service gets fired on a hairline trigger, with zero emotional bond. That’s the real reason, he says, that people who keep asking him to “restart Fin” are missing the point — the service and the unit economics both worked fine; the unsolvable problem was taking on that level of responsibility as a business.
Key points
- Ten years ago, Sam and Andrew Cortina (Venmo co-founder) built Fin, an "artificial artificial intelligence" concierge combining a credentials vault, personal graph, and human-in-the-loop execution.
- Fin famously booked Mark Andreessen's work lunch at the wrong location — the identical mistake Instinct just made to Brit.
- Structural problem one: once you promise to handle a task end to end, you become liable for every downstream service provider in the chain.
- Structural problem two: the economics of mistakes don't scale — a wrong plane ticket or important reservation is too expensive to simply eat the cost, unlike DoorDash's cheap, high-margin food orders.
- The psychological problem: a human assistant who errs gets forgiven due to relationship and sunk cost; a digital assistant gets fired on a hairline trigger with zero emotional bond.
- Sam's answer to people who keep asking why he doesn't "restart Fin": the product and unit economics worked fine — the unsolvable issue was the level of liability the business had to absorb.
Where they landed
Quotes
“Mark Andreessen was a user at one point and got livid because we screwed him, Finn did, the same exact way, booking effectively a work lunch at the wrong place for him.”— Sam
“If you book the someone the wrong plane ticket or send them to the wrong restaurant, that's like not a small deal... you don't have the money, you don't have the margin to pay for the mistakes.”— Sam
“The tolerance for mistakes goes to zero... the problem was what does it mean to take on that level of responsibility as a service, and can you make the economics of mistakes work?”— Sam
Suggested tweets
A decade before Instinct booked the wrong city, Sam Lessin's startup Fin booked Mark Andreessen's work lunch at the wrong restaurant. Same mistake, same category, ten years apart.
Tweet this →Sam Lessin's answer to "why don't you restart Fin": the product and the economics were fine. The problem was taking on liability for every service provider down the chain.
Tweet this →Sam Lessin on why people fire AI assistants instantly but forgive human ones: "There's no emotional bond... the tolerance for mistakes goes to zero."
Tweet this →All topics · Full episode #171 + transcript · Subscribe on YouTube · Spotify