Is AIOps the Future of Operations? Real Use Cases From the Trenches
summary
Adobe operates 70+ clusters and 200,000+ services and builds an agent marketplace that deploys agents as MCPs to Azure functions/foundry, indexes GitHub runbooks into Azure AI search, and routes alerts via Slack with Octa token propagation. Results: 550 automated alerts/month, ~91 engineering hours saved, and 71% RCA accuracy. Agents use curated metric reports, Splunk/Prometheus tools, and a manually curated golden dataset; deployment takes about half a day to spin up an agent. Human-in-loop approvals enable progressive autonomy and safe remediations.
AI-written summary · about
chapters (13)
- 0:00AIOps at Adobe
- 1:27Scaling operations requirements
- 4:12Autonomy spectrum in agentic AI
- 5:35Agent marketplace infrastructure
- 8:23Agent structure in marketplace
- 9:50On-call agent workflow and RCA
- 12:22Key findings on tools and knowledge
- 15:33Curating accurate knowledge bases
- 16:58Automation evaluation with golden datasets
- 18:48Client application integration
- 19:52Autonomy from RCA to remediation
- 21:40Strategic takeaways
- 22:43Q&A
related talks
44mThe Future of IT Ops: AI-Driven Tools You Can Build Today
02mAIOPS – How can machine learning help in IT operations
24mAI Ops and its challenges
29mOps in the AI Era: Why Managed Services Beat Going Solo
25mAI-Powered O11y in Microservices Architectures
33mNext Generation of DevOps AIOps in Practice
29mAutomation first, People centric
17mAI TechOps: Data and Infra for AI Foundation to Scale and Manage 1000s of Cohesive Agents