Your Architecture Diagram Is Not Your Resilience
Microsoft, Wednesday, September 23rd, 2026
Microsoft's Mark Russinovich argues resilience must be continuously validated, since diagrams can look sound while operations drift.
This first article in a Microsoft/Azure resilience series draws on a conversation with Mark Russinovich about how resilience is changing in the AI era.
It argues resilience commonly fails through unnoticed drift, such as a workload deployed across availability zones but with a health probe pointing to one dependency, even as the architecture diagram still looks correct.
It cites Satya Nadella noting organizations must be able to continue operations if a given AI model becomes unavailable, arguing that AI models, inference endpoints, and retrieval pipelines are now critical dependencies that traditional, infrastructure-focused disaster recovery planning doesn't account for.
The series argues resilience must be continuously proven rather than treated as a one-time project.