CI/CD and Deployments · Product Fluency

CI/CD and Deployments · 3 min

Situation

The pipeline is green, but the release still needs judgment.

Automated checks pass after a billing update. The code builds and deploys successfully. A configuration mismatch can still stop real invoices from being generated.

Automation reduces repeated work and catches known failure classes. It does not replace product acceptance or operational readiness.

Process

Change → checks → build → deploy.

Continuous integration regularly combines changes and runs checks. A build turns source into deployable output. Continuous delivery keeps changes ready for release; continuous deployment automatically releases changes that pass the configured gates.

Teams use these terms differently, so ask which steps are automated and where human approval or rollout decisions happen.

Example

Several failures can stop the path.

A type error may fail checks. A missing file may fail the build. Missing runtime configuration may let the build pass but break requests after deployment.

A database migration can also fail or make old code incompatible with new data. The stage of failure helps narrow the investigation and identify who needs to act.

Recovery

Rollback needs preparation.

Rolling back code can restore a previous version, but it may not undo emails sent, charges made, or data transformed by a migration.

Ask whether the previous version can read the new data and whether disabling the feature is safer than reverting everything. A recovery plan must account for external side effects.

PM question

How will we know and recover?

Define the user-facing success signal, monitoring window, owner, and rollback or disable criteria before release. Confirm which business workflow receives a smoke test.

For billing, “the homepage loads” is insufficient. Test the relevant invoice path with safe data and verify the downstream result.

Remember this

Deployment is a controlled change to a running system.

Know the gates, the runtime dependencies, and the recovery limits. A green pipeline is evidence about checks that ran, not proof of every product outcome.