The script was called deploy.sh. It was 247 lines of Bash. It had been written by a human—a former engineer who had left StrongDM two years before the factory existed—and it had deployed StrongDM's infrastructure reliably for longer than anyone cared to count.
Justin knew every line of it. He'd maintained it himself through three major revisions. He'd added the rollback logic in 2022 after an incident that nobody liked to talk about. He'd added the health check in 2023. He'd added the notification hook in 2024, so that the Slack channel would know when a deployment was in progress.
The script worked. It had always worked. It had the battle scars of a tool that had survived contact with reality—commented-out sections that recorded failed experiments, a retry loop with a sleep interval that had been tuned by trial and error over years, a logging function that was more detailed than some entire applications.
And today was its last day.
The factory's deployment pipeline had been running in parallel with deploy.sh for eight weeks. A shadow deployment—every time the script ran, the factory's pipeline ran alongside it, and Jay compared the results. Eight weeks. Fifty-six deployments. The factory's pipeline matched the script's behavior exactly, every time. And in four cases, the factory's pipeline had caught issues that the script would have missed—a stale configuration file, a dependency version mismatch, a certificate approaching expiration.
"It's better," Jay said, and he said it reluctantly, the way you say something true that you wish weren't.
Justin opened the script one last time. He read it from top to bottom. Line 1: #!/bin/bash. Line 247: echo "Deployment complete." Between those bookends, years of accumulated knowledge, encoded in a language that nobody would write a deployment script in if they were starting fresh today.
He added a comment at the top of the file:
# Retired 2026-01-30. Replaced by factory deployment pipeline.
# This script served StrongDM well. Thank you.
Navan, reading over his shoulder, said: "You're thanking a Bash script."
"I'm thanking the person who wrote it," Justin said. "And every person who maintained it. Including me." He committed the comment. He moved the script to an archive directory. He updated the deployment documentation to point to the factory pipeline.
Then he ran the factory's pipeline for the first time as the sole deployment mechanism. The Cedar policies evaluated. The scenarios validated. The canary deployed. The metrics held.
Jay watched the deployment complete. "How does it feel?"
"Like cleaning out a closet," Justin said. "You keep things because they served you. You let them go because something better serves you now. The letting go is the hard part."
The deployment pipeline hummed. The Bash script sat in its archive directory, retired and commented and still. It had served well. But the factory was better, and knowing when to move on was a skill too.
Justin adding "Thank you" to a Bash script comment made me feel feelings I was not prepared to feel about infrastructure automation. That script was someone's craft. Acknowledging that matters.