Schrödinger's Deployment
A true story about a deployment miracle on prod - and how AI helped restore my faith in common sense. We ship a routine backend deploy. Version checks - fine. Tests - green. Next morning: a developer swears the app is showing the old version. I check - it's the new one. We're both right, and neither of us is happy about it. A 5-minute field report on chasing a bug that "couldn't possibly exist," with an AI agent along for the ride.
A 5-minute Ignite talk (comic/story format) from a two-person, no-dedicated-DevOps team running an AI-native app on AWS.
Notes to organizers (private, outline):
- Routine ECS deploy, version checks and tests pass, everyone's happy
- Next morning: dev reports old version on prod, I see the new one - we're both right
- Set up live version monitoring - app flips between old/new on every request
- Wrong theories: rollback? infinite session cache? stuck deploy?
- Bring in Claude to dig through logs - comes back puzzled: deploy was correct, but one worker rolled back afterward, no clear reason
- Take it to platform engineers - "that can't happen" - show them the logs
- They find it in their own deploy script: two failed health checks in a row silently trigger an auto-rollback
- Fix, redeploy, stabilizes
- Punchline: having something that calmly digs through cloud logs and hands you a root cause in minutes, no ego, no panic - genuinely a superpower for a tiny team
Data Solution Architect at EPAM with 30+ years in IT - currently spending half that time debugging AWS/Azure infrastructure with an AI pair. AI-Ambassador and practitioner of AI-Native development in production. Speaker on AI topics.