TL;DR
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
The Pragmatic Engineer published excerpts from Sam Newman’s 2026 book, Building Resilient Distributed Systems, covering disaster recovery and the resilience of AI-powered software. The material outlines recovery targets and options, while warning that model-provider changes and constrained computing capacity can complicate planning.
The Pragmatic Engineer has published excerpts from Sam Newman’s 2026 book, Building Resilient Distributed Systems, setting out how software teams can plan for major outages and vulnerabilities in AI-powered services. The material focuses on disaster recovery targets, recovery patterns and AI provider uncertainty, offering engineers a framework for deciding how much downtime and data loss their systems can tolerate.
In the excerpt from Chapter 10, Newman describes disaster recovery as a set of procedures for failures that disable a large part of a system, such as a site-wide infrastructure outage. He says teams should define a recovery point objective (RPO), the maximum acceptable period of potential data loss, and a recovery time objective (RTO), the acceptable time to restore service. For example, if data is backed up hourly, the potential loss could reach an hour, making the RPO one hour.
The report lists several recovery approaches: backup and restore, pilot light, warm failover and active-active operation across multiple sites. These approaches differ in how quickly systems can recover and in the cost and complexity involved. Newman’s excerpt says restoring a service can require more than recovering data: teams may also need to make computing resources available and redirect traffic to another site.
The report also addresses AI-specific resilience concerns, including dependence on model providers, changes to models over time and the need for observability. It identifies multi-model and multi-vendor strategies, transparent cost tracking and regular vendor analysis as ways to manage uncertainty. The report flags ongoing CPU and GPU shortages as another constraint: teams should not assume infrastructure will always be available and may need to prioritize critical services.
Recovery Targets Shape Resilience
The guidance matters because recovery plans involve trade-offs, not a single universal configuration. Shorter recovery times and less potential data loss can require more infrastructure, testing and operational complexity. Setting RPO and RTO targets makes those trade-offs explicit and gives teams a basis for choosing an approach that fits the service’s needs.
For AI-powered products, resilience also depends on services outside a company’s direct control. If a model provider changes a model or becomes unavailable, application behavior or service continuity may be affected. Observability and provider planning can help teams detect and respond to such changes, while capacity constraints mean backup plans need to account for whether compute resources can actually be obtained during an incident.
As an affiliate, we earn on qualifying purchases.
From Outages to AI Dependencies
The Pragmatic Engineer report says Newman’s book took three years to complete, rather than the two initially planned. Newman told the publication he revisited and reworked the book to address AI’s relevance to system design. The report presents excerpts from two chapters, including material on disaster recovery and uncertainty in AI services.
The book’s recovery discussion uses “site” as a general term for the location of computing infrastructure, whether that means a public-cloud region, a data center or another environment. The report says the patterns discussed can apply to both public- and private-cloud settings. Its publisher is O’Reilly Media; the excerpts are reproduced with permission.
“The recovery point objective (RPO) defines the maximum amount of time that can have passed since the data was last backed up or synchronized.”
— Sam Newman, in the book excerpt published by The Pragmatic Engineer
AI system resilience monitoring tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Service Risks Still Vary
The excerpts do not prescribe one recovery target or architecture for every organization. Appropriate RPO and RTO levels depend on what a particular service can tolerate, and the report does not provide a cost comparison or performance measurements for the listed recovery patterns.
The report also does not identify specific model providers, infrastructure vendors or shortages with figures, nor does it quantify how often AI-related model changes cause incidents. Its discussion is guidance from a book excerpt, not evidence that a particular strategy will prevent outages. Teams’ actual exposure will depend on their systems, contracts and available capacity.
As an affiliate, we earn on qualifying purchases.
Teams Must Set Their Targets
The next step for engineering teams applying the material is to define acceptable data loss and downtime for each service, then choose and test a recovery approach against those targets. They will also need to review dependencies on cloud capacity and AI providers, including whether alternatives are available and whether monitoring can reveal model or service changes.
The report does not announce a follow-up study, product release or implementation deadline. It points readers to Newman’s book for the full treatment; the published excerpts cover only part of its 16 chapters.
cloud infrastructure backup services
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the news development?
The Pragmatic Engineer published excerpts from Sam Newman’s 2026 book, Building Resilient Distributed Systems, covering disaster recovery and AI-powered software resilience.
What do RPO and RTO mean?
RPO is the maximum acceptable period of potential data loss. RTO is the acceptable time to restore service after an incident.
Which disaster recovery approaches does the excerpt discuss?
It lists backup and restore, pilot light, warm failover and multisite active-active. Each involves different cost, complexity and recovery trade-offs.
What AI-specific risks does the report identify?
The report points to model-provider dependence, models changing over time and limited CPU or GPU availability. It discusses observability and multi-model or multi-vendor planning as ways to manage uncertainty.
Does the report recommend one recovery plan for every system?
No. The excerpt presents options and measurable targets, but says the appropriate plan depends on a system’s acceptable downtime and data loss, as well as its costs and constraints.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
