AI SRE: A Talk on Incidents, Open Source, and Engineering Teams in 2026

Production incidents have not changed. What has changed is who responds to them first.

This talk is about AI SRE. Using open source AI agents to investigate incidents, find root causes, and triage alerts in production. Built on the same stack this devroom runs on. Prometheus, OpenTelemetry, Kubernetes, Loki. These agents are already running inside real engineering teams in India and globally.

What this talk covers:

1. What AI SRE actually means at 3 AM, not the conference version
2. How Prometheus, OTel, and Loki become the data layer for AI investigation
3. The ReAct loop, how agents reason and act during a live incident, and where they fail
4. Real numbers, teams cutting MTTR by 60-70% and what that looks like on the ground
5. The honest jobs question, what happens to junior on-call engineers when AI handles tier-1
6. The deskilling risk, why automating everything fastest is not the same as winning
7. Which teams survive, the ones who use AI as a tool and keep engineers sharp enough to catch when it is wrong
8. 
   

Why IndiaFOSS:

Every tool in this story is open source. The AI layer sits on top of the stack, it does not replace it. Every on-call engineer in this room has lived the problem this talk is about.

Not a vendor pitch. Just 10 minutes of what is actually happening in production right now.

Resources : https://arxiv.org/abs/2210.03629

yisusvii.github.io/posts/ai-sre-news-2026

openobserve.ai/blog/opentelemetry-for-llms

sre.google/sre-book/postmortem-culture

1. How open source observability tools become the data foundation for AI incident investigation
2. What the ReAct reasoning loop looks like during a real production incident
3. The difference between AI that helps engineers and AI that replaces the skills engineers need
4. Which engineering teams are winning with AI SRE and what they are doing differently
5. How to think about junior engineer development in a world where AI handles tier-1 on-call