Piggy Picks
Our friend group's had an NFL picks pool for over a decade. Now it's a full app with live scores, because I can do that.
A decade of spreadsheets for 6 friends
Every year we’d use a Windows-95-esque spreadsheet, then the functional-but-annoying Yahoo Fantasy app.
Tiebreaker logic that handles every edge case
Claude Code kept refactoring it into something that technically worked but handled edge cases wrong.
Two AI agents, same codebase, different problems
Brought in Codex as reviewer while Claude moved on to other features. Multi-agent workflows stopped being theoretical.
Live scores, standings, picks. Every game
Real-time ESPN scores, betting lines, season-long stats, automated winner tracking. The whole nine.
My friends and I have had an NFL picks pool for over a decade. Every year we'd use a very Windows-95-esque spreadsheet then eventually the (functional but annoying) Yahoo Fantasy app. In 2025, I built the app we'd always wished for.
Real-time scores, betting lines, and a full season tracker
Real-time ESPN scores, betting lines, season-long stats, automated winner tracking. The whole nine. Full-stack with Next.js, Supabase, and the ESPN API pulling live game data. Six of us use it every Sunday during the season and it's become part of the tradition now.
Live product
Sunday flow from week context to picks and stats.
The app carries the game-day loop across the screens our pool checks most.
The tiebreaker almost killed the project
Claude Code kept refactoring the tiebreaker into something that technically worked but handled edge cases wrong. So I brought in Codex as the reviewer agent and had it sort out the tiebreaker while Claude moved on to other features. Two AI agents, same codebase, different problems. That was the first time I realized multi-agent workflows weren't just theoretical. They were practical.
Detail proof
Picking the tiebreaker, then settling the week.
Predicting the SNF and MNF totals on the left; on the right, the resolution the logic had to survive.
I'd never used a database or an API before, so there were a lot of hours in this one. If I built it again, I'd fight the urge to solve every edge case at once. The tiebreaker nearly killed the project because I didn't know when to stop refining and start shipping. But in the end, we have our own app that everyone loves, because I can just do that now.