AI agents can write code. Giving them access to production is a different problem entirely. As teams move from AI-assisted operations toward agents that can investigate incidents and take action on their own, the challenge becomes deciding what the agent can do, what context it needs, and where humans should stay in control.

In this hands-on workshop, you’ll follow a Kubernetes incident from its first symptom through investigation, remediation, and recovery. Using a sandbox environment and open-source tooling, you’ll connect an agent to operational tools, automate root-cause investigation, inspect its execution, and introduce human approval gates before allowing it to make changes.

Then we’ll push further: trigger another failure and test how far the agent can respond autonomously within defined boundaries. You’ll leave with a practical understanding of what agentic operations can safely automate today, what it takes to move toward greater autonomy, and how context, governance, and orchestration affect both reliability and cost. 

Key Takeaways:

• Build an agent that can investigate and respond to production incidents without being prompted through every step.
• Define where an agent can act autonomously, where approval gates belong, and when a human should remain in control.
• Give agents the operational context they need without sending unnecessary data to the model or driving up cost.
• Understand the practical steps from read-only investigation to safely increasing an agent’s autonomy in production.