Meta but true: coding-os.dev, its docs, the prompt library, the blog, the CI/CD pipeline, and the forum you’re reading were built and deployed by an AI agent operating under Coding OS discipline, over a handful of sessions on a single VPS.
Highlights of what discipline bought:
- Verify-by-executing: every “it works” claim was executed — the deploy pipeline was watched end-to-end (push → CI → build → healthcheck → live), and user flows were clicked through in a real browser, not assumed from code.
- Docs-alignment: every infra change updated its runbook in the same session. The server’s ops repo has a bootstrap runbook plus addendums for every fix — the next operator (or agent) inherits decisions, not mysteries.
- Smallest-correct-change under pressure: debugging auth (CORS + email link paths + a proxy redirect bug in one night) stayed surgical because each fix was classified, scoped, and committed separately.
Stack, for the curious: Next.js standalone container, self-hosted Supabase (auth + REST only, behind Caddy), Discourse with single sign-on, one shared Postgres cluster, GitHub Actions → SSH deploy with a forced-command key. Total RAM at idle: ~2.5 GB.
Happy to answer questions about any piece of it — that’s what Q&A is for.