Google SRE Prodcast The One With AI Agents, Ramón Llamas, and Swapnil Haria
Jul 23, 2025
In this installment, Swapnil Haria, a Google Software Engineer specializing in AI agents, and Ramón Llamas, a seasoned Staff Site Reliability Engineer, delve into the transformative impact of AI on production management. They discuss how these agents can summarize alerts, detect hidden errors, and even prevent outages. The duo highlights the balance between human expertise and AI capabilities, the complexities of evaluating non-deterministic systems, and the importance of structured postmortems in enhancing incident response.
AI Snips
Chapters
Transcript
Episode notes
Agentic AI's Dynamic Problem Solving
- AI agents can operate without fixed scripts, dynamically planning steps to achieve a goal.
- This flexibility means inputs transform unpredictably to outputs, expanding traditional deterministic algorithms.
Enforce Agent Action Guardrails
- Restrict AI agents' ability to modify the world and use sandboxes for code execution.
- Require human permission for any impactful actions to ensure safety and control.
Agents Pre-Process Alerts Fast
- At Google, agents analyze alerts ahead of humans, quickly identifying causes or ruling out issues.
- This saves humans time and still allows them final decision-making control.