The project: AI Concierge on AWS / Bedrock / Python / React.
Backend team: 2 people + AI (Claude Code).
No dedicated DevOps. Platform team has no bandwidth for us.
CI/CD situation - let's just say it's complicated.
But the system is live in production. And we own it.

So we invented our own ops helper.
Over the past year, it has helped us with four things we had no formal training for:

• Incidents - a Jira-watching agent pairs with a CloudWatch collector, they analyze together, ask clarifying questions back through Jira, and prepare recommendations
• Statistics - regular pulls from S3, Athena, and CloudWatch; anomaly detection; trend analysis; report generation
• Monitoring - CloudWatch Dashboard updates and redeployment
• Data problems (Athena/Glue) - someone pushed a migration with an incompatible schema. Together with AI we investigated the cause, patched the structure, found a workaround, fixed and tested the code. Later - solved it properly.

Four real stories. Ordinary days of AI-Native development.