CatchBuddy

Pickup sports are dying in cities, and the apps meant to fix that all assume you want a season. They want commitment, a schedule, a recurring team. Most people just want a game on Saturday.

The problem was never scheduling. It was getting two strangers to agree to meet at a park with both of them feeling fine about it.

A kid can't post a game until a parent is verified. The panic button is reachable from every screen you can be on during a game. And the meeting spots are a list I curated, not somewhere any user can drop a pin.

That last one gets argued with a lot. Letting people add their own locations is more flexible, and I still won’t do it.

AI wrote the RLS policies, the Supabase migrations, the Stripe integration and the OAuth flow. That is real work and it did it fast.

What it could not do was decide who gets in, who gets gated, and what a stranger sees about another stranger before they agree to meet. Every one of those I made by hand. AI's own security review also caught a recursive RLS policy that would have leaked data in production.

Testers kept reading "Matches" as a dating thing, which is not what anybody needed here. It is "Browse" and "Players" now.

I built a Quick Start wizard that nobody wanted. Testers skipped it every time, so I stopped making them skip it.

Apple, Outlook and ICS calendar support all got built, then all got cut. Barely anyone used them and I was going to be maintaining three integrations forever for that.

I made the palette warm. A trust product that looks like a fintech dashboard reads as a company rather than a neighbour. The safety states sit inside the same system from v1, instead of arriving later as status chips bolted on the side.

It shipped. Auth, RLS, Stripe, Google OAuth, realtime updates, the minor-approval flow and the curated meeting spots, designed and built by me.

The safety layer went in first, in v1. Every product I have seen add one later ended up with a settings screen nobody opens.