I've worked on enough startups to know that "deployment pipelines" can go from "helpful" to "why are random people asking me about YAML" pretty fast. I've watched teams death-flail over Kubernetes configs, bespoke CI servers, and one brave soul's "temporary" Bash script that somehow became critical infrastructure.
So I landed on a workflow that's boring, repeatable, and doesn't require a cloud priesthood.
The Workflow
1. Build and Break Things in Replit
Replit is my sandbox. I ideate, build, and iterate fast. Instant feedback matters. I'm not waiting for containers to build or fighting local environment drift. The messy experimentation part happens here.
2. Database Goes in AWS, Period
This is where people get sentimental and keep the staging database in Replit. Don't. Replit containers are ephemeral, and you'll lose data eventually. Put your staging DB in AWS (or GCP if you must) so staging mirrors production. Use Replit Secrets for connection strings and move on.
3. Link Everything to GitHub
Every Replit app gets connected to a GitHub repo. This is the bridge between "it runs" and "it ships." Commit often. Push regularly. Write decent commit messages so you remember what you did.
4. Deploy via GitHub Tags
When it's time to ship, I create a GitHub tag using date-based versioning: YYYYMMDDHHII (like 202501201430). That tag triggers deployment to AWS Lightsail via GitHub Actions.
Tags beat "just push to main" because they create an audit trail and make rollbacks simple. More importantly, they let you answer the question "what version is in production?" without needing a séance.
You can even create the tag in the Replit shell! How cute is that?
git tag 202501201430
git push origin 202501201430
GitHub Actions handles this automatically - check their docs if you need the setup details.
Things That Will Bite You (If You're Not Careful)
If staging and production aren't similar, you'll end up debugging "ghost bugs" that only show up in production. These suck. Try really hard to keep things close in functionality. This means, no month-long binge coding "accidents". When I'm working I push to prod multiple times per day. This shouldn't be strange.
Handle secrets like an adult. Replit Secrets are fine for staging. In production, use AWS Secrets Manager (or at least something designed for production). Never hardcode creds. That's not a shortcut, it's a future incident report.
Test in staging before you ship. I'm not saying you need perfect test coverage. I'm saying you should at least run the thing in an environment that looks like production before you let customers run into it first.
Automate database backups. "Manual backups" is startup-speak for "we lost everything."
Have a rollback plan, and actually try it. Murphy's Law is real and that deployment will fail at the worst possible time.
Why This Works (In Real Life)
Replit lets you move fast. GitHub keeps your changes trackable. Lightsail is reliable without needing a PhD in cloud architecture. And date-based tags tell you exactly when something deployed.
The beauty isn't in complexity. It's in how boring the process is. When you're shipping a lot of apps, boring is the feature.
-Sethers