Skip to content

All writing

Fourteen people signed up. Nobody came.

In July, 14 people signed up for our first running group and nobody showed up. That morning is why Ponyo lives inside the AI people already talk to.

ponyocommunityfounders

Thursday, July 23rd, nine thirty in the morning. I was standing outside a coffee shop with Guada, waiting for the running group. Fourteen people had signed up on Luma. We waited thirty five, forty minutes. One person showed up, a friend I had messaged on Instagram. We ran anyway, the three of us.

I thought being specific was enough. Builders and creators who run, a tight group, a real movement. It was well designed. Signing up on Luma is easy. Showing up is hard. Showing up again is harder.

The running event was not the plan. The plan was the intake interview. We spent a couple of weeks designing it. Psychology, lifestyle, age group, life situation. We wanted to know people well enough to put them in the right group of nine or ten, not just fill a room. Nobody completed it. Nobody messaged our WhatsApp. Nobody filled out the form. So we thought, fine, let’s meet people in person first, then the interview will feel natural. That’s what the Luma event was for.

Then nobody came to that either.

Guada and I talked while we ran. I was disappointed. I said let’s figure out why people don’t show up, let’s figure out how we actually build a curated group. We kept asking the same question for days after. How do we get people to fill out the onboarding flow so we can put them with the right people?

Then the question changed. Instead of asking how we get people to tell us about themselves, we asked where people already do that. And the answer was sitting there. People spend hours talking to their AI. What they want, their relationship problems, the kind of people they want to meet, the kind of community they want to join. Nobody makes them fill out a form for that. They just talk.

So why build another form? Why not go to where the information already lives, with the user’s permission, and curate from there.

I wrote the product requirement document for version 0.4. Fourth round of iteration. Then I wanted to prove it was possible architecturally, so I built a demo MCP worker on Cloudflare and connected it inside Claude. Four or five days, maybe seven. When it worked I could talk to my own server through my own Claude.

I typed: hey, use Ponyo to find a cofounder for me.

It came back and asked me to allow Ponyo to create a profile. Then it asked the first question on the intake script. What’s your name, and what are you looking for.

I said out loud, this is so cool.

What I would tell myself: stop talking to the AI about strategy, stop building the website, go get user feedback as fast as you can.