TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The Pragmatic Engineer published excerpts from Sam Newman’s 2026 book, Building Resilient Distributed Systems, covering disaster recovery and AI-powered software reliability. The material discusses recovery targets, failover approaches, infrastructure constraints and uncertainty around model providers; it does not present new outage data or a tested industry-wide solution.
The Pragmatic Engineer has published excerpts from software engineer Sam Newman’s Building Resilient Distributed Systems, a 2026 book that addresses disaster recovery and the reliability challenges of AI-powered software. The material offers engineering guidance on setting recovery targets and planning for model-provider changes, but the source does not document a particular outage or provide new measurements showing that AI has caused more incidents.
Newman’s excerpt from Chapter 10 focuses on disaster recovery when a large-scale failure—such as a cloud infrastructure problem or a site-wide outage—disrupts services beyond the reach of routine protections. It frames recovery planning as a trade-off: reducing acceptable downtime or data loss can require greater cost and complexity.
The excerpt defines two measures for setting requirements. A recovery point objective (RPO) sets the maximum acceptable gap between the latest backup or synchronization and an incident, and therefore bounds potential data loss. A recovery time objective (RTO) sets the acceptable period of service unavailability. Newman notes that restoring data, making compute resources available and redirecting traffic all contribute to recovery time.
The newsletter says the book also covers disaster-recovery patterns including backup and restore, pilot-light and warm failover, and active-active operation across multiple sites. Its AI-related topics include provider resilience, model changes, observability and ways to manage uncertainty, such as considering multiple models or vendors and examining vendor arrangements regularly. These are topics and approaches described in the source, not evidence that any one pattern suits every system.
Recovery Plans Must Include AI Dependencies
The guidance matters to teams whose services depend on both conventional infrastructure and external AI providers. A recovery plan can account for a failed data center yet still leave a product impaired if a model provider is unavailable, a model changes, or the system cannot observe how that change affects its behavior. Newman’s framing asks engineers to treat those dependencies as part of reliability planning rather than as separate procurement concerns.
The practical value is in making risk and service expectations explicit. Teams can decide how much data loss and downtime they can accept, then weigh those targets against the cost and operational demands of their chosen design. The newsletter also flags limits on GPU and CPU availability as a planning concern; it does not quantify shortages or establish how often they prevent recovery. That makes vendor capacity and the priority of critical workloads questions for each organization to investigate.
disaster recovery backup and restore software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Site Failures to Model Risk
The Pragmatic Engineer report presents the excerpts as part of a podcast and book discussion with Newman. It says he spent three years writing the book, rather than the two he had initially planned, and reworked sections to account for AI. The book, published by O’Reilly Media, has 16 chapters; the newsletter says its excerpts cover only part of the material.
Newman uses “site” as a general term for the location of computing infrastructure, whether that means a public-cloud region, a data center or another deployment environment. That terminology lets the disaster-recovery patterns apply across different setups. The newsletter introduces the AI discussion as covering vendor and model uncertainty, alongside the more established problem of recovering services after a large-scale infrastructure failure.
“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 an excerpt from Building Resilient Distributed Systems
multi-site failover infrastructure
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limits of the Published Evidence
The source is an excerpt and editorial report, not an independent survey of outages or a comparative test of recovery strategies. Its statement that more AI-written code is associated with more deployments and bugs appears in the newsletter’s framing, but the supplied material gives no supporting dataset, time period or comparison baseline. It should not be read as a quantified finding that AI has caused a rise in production incidents.
The report does not specify which recovery pattern is best for particular workloads, how often model changes cause failures, or what level of GPU and CPU scarcity teams should plan for. It also does not provide implementation results, costs or service-level data for the approaches discussed. Those details remain dependent on each system’s requirements and vendor arrangements.
As an affiliate, we earn on qualifying purchases.
Teams Must Set Their Own Targets
The immediate next step for engineering teams is to define acceptable data loss and downtime for each critical service, then evaluate whether their backup, failover and traffic-routing arrangements can meet those targets. The excerpt points to weighing recovery speed against added cost and complexity rather than adopting a single default design.
The newsletter says its selected material spans disaster recovery and AI resilience, while noting that it covers less than a full chapter. Readers seeking the full treatment can consult Newman’s book. For teams applying the guidance, further work includes reviewing provider dependencies, monitoring model behavior and checking whether replacement infrastructure and vendors would actually be available during a disruption.
As an affiliate, we earn on qualifying purchases.
Key Questions
What development does the report describe?
The Pragmatic Engineer published excerpts from Sam Newman’s Building Resilient Distributed Systems, covering disaster recovery and resilience issues in AI-powered software.
What do RPO and RTO mean?
RPO is the maximum acceptable period of data that could be lost after an incident. RTO is the maximum acceptable time a service can remain unavailable while it is restored.
Does the report prove AI is causing more outages?
No. The newsletter raises that concern in its introduction, but the supplied source includes no data, timeframe or baseline to establish an increase or its cause.
What AI-related risks does the excerpt raise?
It identifies dependence on model providers, changes to models, observability and uncertainty about vendor capacity as issues for resilience planning. It does not quantify how often these risks lead to service failures.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
