Google Releases Site Reliability Engineering Second Edition, Detailing AI-Era SRE Evolution

Google released the second edition of its Site Reliability Engineering book, updating the 2016 first edition for AI-driven and increasingly complex cloud environments. The edition documents a decade of evolution across six practice areas, including SLOs and error budgets, observability, automation and platforms, reliability testing, capacity management, and organizational models. It also examines the two-way relationship between AI and SRE, covering AI infrastructure operations and AI as an operations tool, while reaffirming that reliability-centered culture and the human element remain central to SRE success.
Event version 3.
Key insights
Six practice areas evolved over ten years
SLOs and error budgets moved from basic availability metrics and manual budget policies to user-centric multi-level SLOs, dynamic burn-rate alerts, and automated policy enforcement. Observability shifted from static dashboards and manual alert classification to OpenTelemetry, distributed tracing, and automated anomaly correlation.
Automation, testing, and capacity management matured
Automation moved from custom scripts for isolated tasks to declarative IaC, internal developer platforms, and self-healing systems. Reliability testing advanced from periodic Game Days and synthetic load to continuous resilience checks, pipeline testing, and production canary validation, while capacity management shifted from manual configuration and static forecasting to dynamic scaling, FinOps alignment, and GPU/TPU-specific optimization.
Organizational models shifted toward hybrid structures
Organizational patterns moved from dedicated SRE teams bound to specific applications to hybrid models, platform engineering integration, and reliability champions embedded in product teams.
AI and SRE have a two-way relationship
SRE supports AI infrastructure by managing large-scale GPU/TPU cluster reliability and network topology, monitoring non-deterministic outputs and latency, and sustaining model training, fine-tuning, and data pipelines. AI acts as an SRE force multiplier through generative AI incident triage and postmortem drafting, natural-language telemetry queries, and predictive machine learning that identifies resource bottlenecks before failures occur.