Agents in Action: How Google is Automating Operational Toil
summary
Google builds a system using LLMs and agents to clear routine on‑call ticket toil for parts of the org, not for production emergencies, and runs it dry-run only until teams trust it. A cron job picks machine‑generated tickets the agent can handle, it fetches logs and metrics but never mutates production without human approval, and it prioritizes precision over coverage to avoid spurious actions. Auto‑generated skills are double‑checked to prevent overfitting, and teams of seven report two people per week freed from ticket queues.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:17Automating on-call ticket toil
- 3:02Agents handling on-call tickets
- 6:37Training agent skills
- 9:08Building trust with agents
- 11:15Avoiding overfitting in responses
- 12:59Cultural shift to responses
- 14:02Band-aid vs long-term solutions
- 16:09Key lessons learned
- 19:04Q&A
related talks
29mWhat exactly are AI Agents?
14m15 Months of building AI agents in 15 minutes
46mLeveraging AI and Automation in DevOps
25mHow Google DeepMind Runs Agents at Scale
30mAgents Building Agents
40mHow To Tame Your Agents?
36mAI Agents in Automation | 3 Real Production Cases with ROI
10mFrom Blank Page to Prod: A Year of Building AI Agents at Carrefour