Google has announced a completely rewritten second edition of their seminal book Site Reliability Engineering: How Google Runs Production Systems, ten years after the original book was first published. The book updates thinking on site reliability engineering (SRE) to encompass modern cloud architectures, platform engineering and AI workloads, whilst keeping the original emphasis on applying good software engineering techniques to operations work.

Announced by Google's SRE programme manager Betsy Beyer and software engineer David Huska, the new edition is positioned as a response to a production environment that has changed substantially since 2016, when the first edition appeared. The authors argue that the central problem is no longer adopting cloud infrastructure, but maintaining reliability as systems become more distributed, interdependent and continuously changing.
"Modern production environments operate under constant change. System boundaries are more complex, dependencies are deep, and failure modes are more complex and damaging than ever before."
Betsy Beyer and David Huska, Google Cloud Blog
The original SRE book became influential because it gave a public account of how Google operated large-scale services. It described practices such as service level objectives (SLOs), error budgets, toil reduction, incident management and postmortems. InfoQ's 2016 review noted that the book presented SRE as the application of software engineering to an operations function, with engineers expected to automate repetitive operational work rather than simply respond to it.
The second edition does not present SRE as a fixed set of practices. Instead, it describes how those practices have changed as organisations have moved from early cloud migration and monolithic applications towards microservices, multi-cloud deployments, serverless systems and internal developer platforms.
One of the clearest changes over the years concerns SLOs and error budgets. Google says that teams once relied on basic availability metrics and manually applied error-budget policies. The revised guidance describes user-centred, multi-tiered SLOs, dynamic burn-rate alerting and automated policy enforcement. Reliability practice itself has evolved away from infrastructure uptime alone and towards measures that describe the effect of failures on users.
Observability has also evolved, and the book's new edition talks this through. The first edition's world of static dashboards, basic logs and manual alert triage has given way, in Google's account, to OpenTelemetry, distributed tracing and automated anomaly correlation. The aim is not merely to collect more telemetry, but to reduce the time between an alert and a credible explanation of what has failed.
Automation is the third major area of revision. Google describes a movement from custom scripts addressing isolated tasks towards declarative infrastructure as code, internal developer platforms and self-healing systems. This aligns with the platform-engineering model now common in larger organisations: reliability capabilities are built once into a platform and consumed by product teams, rather than being reimplemented by every team.
Reliability testing has also changed. Google contrasts scheduled game days and synthetic load tests with continuous resilience checks, automated pipeline testing and live production canary verification. The practical implication is that resilience becomes part of the delivery pipeline rather than an occasional exercise performed before a major release.
"Rather than relying solely on scheduled game days and synthetic load tests, modern engineering embeds continuous resilience checks, automated pipeline testing, and live production canary verification."
Betsy Beyer and David Huska, Google Cloud Blog
Capacity management is another area where the book has been updated. Google says that manual provisioning and static forecasting have been replaced by dynamic auto-scaling, cloud FinOps alignment and optimisation for GPU and TPU clusters. This matters because AI training and inference workloads can concentrate demand on scarce accelerated computing resources, making capacity planning, networking and cost control more operationally significant.
Generative AI inevitably appears in the second edition, in two facets. First, SRE is presented as a discipline needed to operate AI infrastructure. The book covers the reliability of large GPU and TPU clusters, network topologies, non-deterministic model outputs, latency behaviour and the pipelines used for training, fine-tuning and data processing. These are different operational problems from those posed by a conventional stateless web service.
Second, AI is presented as a tool for reliability engineers. The post welcomes generative AI support for incident triage and postmortem writing, natural-language queries over telemetry, and predictive models that identify resource bottlenecks before they cause an outage. The examples in the announcement mostly place AI in a supporting role, gathering and organising evidence for engineers, rather than claiming that AI can replace operational judgement. Recent industry activity appears to support this, with InfoQ reporting in October 2026 that Datadog had released Scout, a specialist model intended to investigate production problems inside its Bits AI SRE product. Other vendors are exploring similar territory, including Honeycomb's Canvas Agent, Splunk AI SRE and New Relic's SRE Agent. InfoQ also reported in February 2026 on Google Cloud SREs using the Gemini CLI during real outages; using AI to help engineers work through incidents, rather than as an autonomous replacement for on-call staff.
The announcement also revises the organisational side of SRE. Google describes a move beyond dedicated SRE teams attached to particular applications, towards hybrid models, platform-engineering integration and reliability champions embedded in product teams. This has come about because many organisations cannot staff a central SRE team for every service, so reliability expertise has to be distributed more widely.
Despite the technological updates, Google argues that the underlying culture has not changed. The authors restate the importance of reliability-focused culture and expand the book's coverage of the human aspects of systems engineering. David Blank-Edelman made a related point in a presentation, You Are Asking the Wrong Questions (about Reliability and SRE), delivered at InfoQ Dev Summit Boston in 2025. He argued that reliability work should start from the customer's perspective rather than the component's perspective, and described SRE as fundamentally collaborative. He also challenged the conventional focus on a single root cause, favouring an examination of the contributing factors behind a failure.
The second edition is available in physical format through Google Books and in digital format from O'Reilly. For teams already practising SRE, its value is likely to lie less in introducing familiar concepts than in showing how those concepts have been adapted for platforms, distributed systems and AI-heavy infrastructure. For organisations earlier in their reliability journey, it remains a substantial reference on how to make operations a software-engineering problem rather than a collection of manual tasks.