Agents Wrote 230K Lines of His AI Tool With 160K Installs — Here's His Production Checklist

PawelHuryn · x · 2026-09-29

Pawel Huryn's AI coding tool hit 160,000+ installs with 230,000+ lines of code and 7,000+ tests, all written by agents with zero hand-written code and word-of-mouth growth. His six-step checklist for going from "works on my machine" to production:

Separate dev from prod: at minimum dev/prod branches on GitHub; a separate Supabase prod instance with migration scripts; a Clerk prod instance wired to real Stripe; Netlify deploying from the prod branch
Put a number on "unlimited": plan feature limits (e.g. 10 active rooms in Clerk), rate limits (60 POSTs/min per IP on Netlify, 100 req/10s per IP on Cloudflare) — test until you hit HTTP 429
Write the intent first: AGENTS.md covering users, roles, out-of-scope; docs on value proposition, permissions, flows, business rules — then have agents write tests against them. Otherwise agents test what the code does, not what it should do
Use it in a real browser: a test account per role, Playwright for key journeys, plus 15-20 minutes of manual clicking — agents still miss what flashes on screen briefly
Review before every release: one pm-skills prompt yielding three reviews — security (permissions, user input) and performance (queries under load) — findings ranked High/Medium/Low
Monitor post-ship: UptimeRobot on key pages, a footer status page, Supabase logs then Sentry, Clarity for user drop-off

Different stack? Swap the tools. The checks stay the same.

Original post →

Agents Wrote 230K Lines of His AI Tool With 160K Installs — Here's His Production Checklist

More from coding & agent