# After a system launches

_Published 2026-08-28._

The Engineering Notes page lists six principles. This essay is about the one that is hardest to keep: **production is part of development**.

## What "done" means after deploy

A system is not done when it ships. It is done when it can be operated, monitored, backed up, migrated, and recovered from failure by someone who did not build it. Until then, the build is still in progress, even if the deployment has happened.

This is why engagements include the operational layer — logging, health endpoints, migration paths, and failure recovery — as part of the build, not as a follow-up. A system that cannot be operated is a system that will be rewritten or abandoned.

## What survives after launch

The parts of a system that survive after launch are rarely the parts that were most visible during the build. The deployment script, the migration tooling, the monitoring, and the runbook outlast the original architecture. They are the part of the record that the next operator reads first.

A case study that ends at launch hides the part of the work that lasts longest. The current state section on each case study is the answer to this: it records what the system is doing now, years after the build, so the record stays honest about what survived.