How to Deploy Without Breaking Production (Checklist for Small Teams)
Deploy safely by checking before, watching during and testing after. Write the rollback plan before you start, and roll back the moment a trigger trips. With that checklist, you can ship on any day. The free AI starter kit we build our own products with runs most of it for you.
Before you deploy
- CI is green on exactly the commit or tag you're shipping.
- Every environment variable and secret the new code reads exists in production.
- Migrations are reviewed and tried on staging, and any that can't be undone are named out loud.
- The rollback plan is written down: what to revert, how, and how long it takes. If a migration can't be undone, the plan says what happens to the data.
- Feature flags are set for this release.
- Anyone who uses an API, event or schema you changed has been told.
- Traffic is low and someone's around for the next hour.
If anything fails, stop there. Never deploy around a failed check.
While it deploys
Watch the deploy logs live. Unexpected errors at startup mean stop and roll back, so don't wait to see if they settle.
After it's live
- The health check returns 200.
- You've tried the flows that make money and the ones users hit first, for real.
- Error rate, latency and slow queries look like the hour before.
- No new alerts.
- Release notes are out and tickets are closed.
Roll back now if
| Trigger | Why it matters |
|---|---|
| Errors above 1% of requests | Users are hitting failures you didn't have before |
| p95 latency more than doubles | The app feels broken even when it works |
| Any sign of wrong or lost data | It gets harder to fix every minute |
| A main flow returns 5xx | People can't sign up, log in or pay |
Finish with one line for the team: what you deployed, when, whether it's healthy or rolled back, and anything left to watch.
Get the deploy check we use
The starter kit checks CI, config and endpoints before you ship, watches after, and tells you the moment to roll back. It works with Claude Code and Codex. Get the starter kit free on GitHub.
Questions
What should I check before deploying to production?
CI is green on the exact commit, every environment variable exists in production, migrations are reviewed and tried on staging, and a rollback plan is written down.
When should I roll back a deploy?
Roll back straight away if errors rise above 1% of requests, slow-request latency more than doubles, any data looks wrong or lost, or a main flow returns server errors.
Is it OK to deploy on a Friday?
Avoid Friday evenings unless it's a fix. Deploy when traffic is low and someone can watch for the next hour.