Fintech firm Mercury recently dropped some data that made me smile. It ranked the top five coffee shops powering founders in San Francisco based on actual transaction data: Sightglass, CoffeeShop, Equator, Saint Frank, Ritual. I’ve built Octolane with my cofounder, Rafi, from every single one of them.

But here’s what the data doesn’t show: the $500,000 investment term sheet I negotiated over a cortado at Cafe Réveille. The $800,000 deal I closed while sitting next to a grad student cramming for finals. The three customers who became friends, then advocates, then our biggest champions, all because we met first over coffee, not Zoom.

When I was in high school, I cleaned offices at night, empty offices with ergonomic chairs and standing desks and those motivational posters about “innovation.” Meanwhile, I’m building an AI company worth millions from a wobbly table at a coffee shop, and somehow this feels more real, more honest than any corner office ever could. 

The distributed office isn’t dead, it just moved to cafés.

I wake up at 5 a.m. here in San Francisco, because those three hours before the city stirs are mine. I review what our AI models learned overnight. I write. I think. Then I head to whichever coffee shop matches my energy that day.

Saint Frank when I need to focus, since it’s quieter, more intimate. Sightglass when I want that productive hum of energy around me. Equator when I’m meeting someone for the first time and want them to feel comfortable, not intimidated.

Rafi, my cofounder and CTO, moved internationally to build this with me. One of our engineers handles the front end from one continent, another tackles the back end from another. So why would I pay $8,000 a month for an office in SoMa (the neighborhood South of Market Street) when I can spend $200 a month on lattes and have the entire city of San Francisco as my workspace?

The serendipity factor is real. Last month, I was debugging a particularly nasty prompt engineering issue, trying to get our AI to detect deals in Gmail without false positives. I was muttering to myself (yes, I’m that guy) when someone at the next table leaned over: “Are you working on LLM classification?” Turns out he’s an AI researcher at Anthropic. Twenty minutes later, I had a completely new approach that cut our error rate in half.

You can’t engineer that........

© Fast Company