Inspiration

I was walking my granddog Hank through San Francisco doing what every SF dog walker does: scanning the sidewalk like a minesweeper. The city has world-class AI labs and a well-documented sidewalk poo problem. It seemed rude not to introduce them.

What it does

A full-screen map of SF where neighbors flag human-excrement sightings so dog walkers and parents of small kids can steer around them. It's super easy to use. Two taps to report. Markers fade over 48 hours (Fresh: avoid → Fading: probably handled → Gone: the city works, sometimes). "Still there 😤" re-confirms a sighting, "Gone! 🎉" clears it with confetti, because nothing is better than a clean street. Other not so poopy features: a live stats bar, locate-me, keyboard-accessible reporting, and a 311 pointer for real hazards.

How we built it

The idea came from a dog walk. I was reading my email (as one does) and saw the OpenAI hackathon email and Sam Altman's tweet about 5.6 being half the price and more token efficient than fable. At just that moment, Hank lunged in an attempt to sink his teeth into a tasty street poo. Nooooo!!! I yelled and then thought okay what if the best SF models work together to fight this street crime!?! And Watch Your Step SF was born.

I know this isn't allowed, but Fable, acting as product lead, wrote the implementation brief, reviewed the code, and browser-tested every flow. GPT-5.6, driven through a handy piece of code, codex exec, built the entire app from that brief. One hardening fix later (an animation-timing bug), it shipped. Huzzah! Poo criminals be warned. Cross-lab collaboration: Claude Fable managed, GPT-5.6 did the typing, Codex held the session with love. The city's sidewalks: marginally safer?

Challenges we ran into

One: a requestAnimationFrame timing bug found in browser testing, fixed with setTimeout. That was genuinely the only one. Single file, no build step, no backend.

Accomplishments that we're proud of

An entire working civic app, from a written brief, in one-ishCodex session. Also the phrase "doo-diligence," which is maybe the best part of this app.

What we learned

I learned that the models can work together beautifully and look forward to working with them to solve other impactful global issues very soon.

What's next for Watch Your Step SF

Swap localStorage for a shared Supabase table and the map becomes truly communal. If SF Public Works calls, we're taking the meeting.

Built With

• fable
• gpt-5.6