# A mobile product shipped to TestFlight in six weeks, by one engineer with agents

Product | 2026 | React Native, Expo, TypeScript, Supabase, PostgreSQL, Drizzle, grammY, GitHub Actions, EAS

A social app for a city's motorcycle community, with a Telegram bot for the chats riders already use: iOS and Android from one codebase, one verification gate shared by people and CI.

## Evidence

- From the first commit on 2026-08-10 to a production build on TestFlight on 2026-09-17; 1,235 commits by 2026-09-23
- One gate list in one script, called by the developer and by every CI workflow, so a check cannot pass locally and be missing in CI
- All database structure in one Drizzle package; CI replays every migration from an empty Postgres and checks for schema drift

## The problem

Riders in a city organise group rides by hand in chat groups: a message, a list of names, and nobody knows who will actually show up. The product bet is that the rider's profile (bikes, rides, service history) is the centre, and that every feature both feeds it and reads from it.

## What I built

**Two surfaces.** A React Native app (Expo) for iOS and Android, and a Telegram bot that works inside the chats riders already use, so the first contact needs no install.

**One backend contract.** Supabase with row-level security. The whole schema, policies and migrations live in one package; shared result and error types are used by the app and the bot alike.

**Gates, not good intentions.** One command runs every check. Four CI workflows call that same script by tag, and the database workflow replays all migrations from nothing on plain Postgres and checks for drift. A clickable design replica is generated from the app's own tokens and copy, so design is reviewed on the real thing, not on a mock-up.

**Built with agents.** I work as the product owner and reviewer; coding agents write most of the code under written repository rules and the same gates as a human.

## What is not solved

It is not in the App Store yet, and it has no field validation: the show-up mechanics for rides are designed, not shipped, and the product hypotheses are marked as hypotheses in its own documents.