The AI SRE Handbook
Operating Production Systems in the Age of AI
Production systems have always been getting more complex. AI has accelerated it: teams ship more code into systems with more dependencies, faster than the workflows built to operate them. Even organizations that instrument everything still lose hours per incident assembling evidence the system already had. The bottleneck is no longer visibility. It’s reasoning.
Gartner forecasts that 85% of enterprises will use an AI SRE platform by 2029, up from less than 5% in 2025. But a frontier model pointed at your observability stack is not an AI SRE. This handbook covers what's changed since the Google SRE Book, what self-driving production requires, and how to evaluate any AI SRE platform, including your own.
What you’ll take away
- Why incident response still takes hours despite a decade of observability investment
- The L0–L5 maturity ladder, and how to place your organization on it
- Five questions for evaluating any AI SRE platform, backed by benchmark data from real Sev1–3 incidents
