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 resilience for AI-powered software. The material discusses recovery targets, failover approaches, infrastructure constraints and the need to monitor model and vendor changes; it does not provide evidence that AI has caused a measured rise in outages.
The Pragmatic Engineer has published excerpts from Sam Newman’s 2026 book, Building Resilient Distributed Systems, laying out disaster-recovery planning and resilience considerations for software that uses AI. The report focuses on how teams can set measurable recovery targets, choose failover arrangements and prepare for disruptions to infrastructure and model providers.
Newman’s excerpt describes disaster recovery as the procedures used when a major event—such as a fire, flood, power problem or failure in underlying cloud infrastructure—disables a substantial part of a system. It frames recovery planning as a series of trade-offs: reducing acceptable downtime or data loss can raise both cost and operational complexity.
Two measures are central to the approach. A recovery point objective (RPO) sets the maximum acceptable period of data loss, based on how recently data was backed up or synchronized. A recovery time objective (RTO) sets how long a system can remain unavailable. Newman’s account says RTO includes the work of restoring data, making computing resources available and redirecting traffic to the recovery site.
The report also identifies backup-and-restore, pilot-light, warm-failover and multisite active-active arrangements as patterns covered in the book. Its AI-related topics include model-provider resilience, changes to models and observability. The report says the book took three years to complete rather than the two Newman initially planned, as he revised its material to address AI. It does not publish comparative outage data or quantify AI’s effect on production incidents.
Recovery Targets Shape Resilience
The guidance matters because a recovery plan cannot be judged only by whether a backup exists. Teams need to decide how much data they can afford to lose and how long a service can be down, then choose infrastructure and procedures that can meet those limits. Those choices affect architecture, staffing and spending.
AI services add dependencies that may change outside an engineering team’s direct control. The report highlights provider resilience and model changes as design concerns, alongside observability. Monitoring can help teams detect when system behavior or availability shifts, while plans involving more than one model or vendor may be considered as ways to manage uncertainty. Such approaches also bring added integration and operating costs; the source does not claim they eliminate outages.
Newman’s excerpt also warns that recovery plans depend on infrastructure being available when needed. The report points to CPU and GPU shortages as constraints teams should factor into planning, and says they should understand vendor limitations and prioritize critical services. A failover design that assumes unlimited replacement capacity may not match what suppliers can provide during a broad disruption.
As an affiliate, we earn on qualifying purchases.
From Site Outages to AI Dependencies
The book excerpt appears in a Pragmatic Engineer report following a podcast conversation with Newman. It presents disaster recovery as a complement to handling failures of individual service instances: a site-wide or other catastrophic event can defeat protections designed for smaller, localized problems. Newman uses “site” broadly for the place where computing infrastructure runs, whether that means a cloud region, a data center or another setup.
The report covers selected material from two chapters, not the whole book. It names four recovery patterns—backup-and-restore, pilot light, warm failover and multisite active-active—but the supplied material does not lay out their detailed costs, recovery speeds or suitability for particular businesses. Newman’s book is published by O’Reilly Media, and the report says its excerpts are used with permission.
The report’s introduction links more frequent code deployments and production bugs to the growing use of AI coding agents, but offers no incident dataset or analysis establishing that relationship. That statement is part of the report’s framing, not a quantified finding. The specific material provided instead concentrates on planning for catastrophic failure and AI-related dependencies.
“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
As an affiliate, we earn on qualifying purchases.
Outage Evidence and Vendor Limits
The supplied report does not establish whether AI coding agents have caused an increase in outages, nor does it give figures for incident frequency, data loss or recovery performance. Its comments about more deployments and bugs are presented as the author’s framing, not as a measured result. It also does not specify which model providers or infrastructure vendors face particular risks.
The excerpt introduces recovery objectives and patterns but does not give a recommended RPO or RTO for a specific service. Appropriate targets depend on a system’s needs and the cost of meeting them. The report also does not quantify GPU or CPU availability constraints or say how well any particular multi-model, multi-vendor or failover strategy performs.
As an affiliate, we earn on qualifying purchases.
Set Targets Before Choosing Failover
The next practical step for teams is to define acceptable data loss and downtime before choosing a recovery design. They can then compare the available patterns against those targets, including the time and resources required to restore data, bring up workloads and redirect traffic. The report does not identify a future research release, regulatory deadline or scheduled milestone tied to the book.
For AI-powered systems, the topics raised point to reviewing provider dependencies, changes in model behavior, observability and infrastructure capacity as part of ongoing resilience planning. The source does not provide a rollout plan or claim these measures guarantee continuity; readers seeking specific implementation trade-offs would need to consult the full chapters or assess their own system requirements.
distributed system resilience hardware
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the news?
The Pragmatic Engineer published excerpts from Sam Newman’s 2026 book on disaster recovery and resilient software, including AI-related concerns.
What do RPO and RTO measure?
RPO is the maximum acceptable period of data loss, tied to the last backup or synchronization. RTO is the acceptable time a system can remain unavailable after an incident.
Which disaster-recovery patterns does the report name?
It lists backup-and-restore, pilot-light failover, warm failover and multisite active-active. The excerpt supplied does not compare their costs or performance in detail.
Does the report prove AI is causing more outages?
No. The introduction links more code deployments and bugs with AI coding agents, but the supplied report provides no incident data or quantified evidence to establish that AI has increased outages.
What AI-related risks does the book address?
The report says the book covers model-provider resilience, model changes and observability, as well as uncertainty-management options such as using multiple models or vendors.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
