Finding Change That Passes Review Before It Breaks Production With Stag
25m
Stag's demo showed how a reasonable infrastructure change can pass review and still take a site offline. Rick Richardson ran a Terraform plan that tightened an AWS security group so web servers accepted traffic only from the application load balancer, a change most reviewers would approve. Stag analyzed the plan against its environment model and flagged it as unsafe: AWS health checkers don't run from the load balancer, so the new rule would block them, the servers would fail their checks within about 60 seconds, and the load balancer would pull them out of service. Rob Bird noted that AWS documents this behavior, but buried deep in the documentation, so a developer troubleshooting the outage would struggle to understand why the site went down.
Stag builds its model by collecting configuration metadata from cloud control planes using read-only access, and it extends into systems such as databases, where it reads schemas, table definitions, and index settings. It assembles that metadata into a graph that maps the causal relationships between components, such as a database's dependency on DNS and load balancers. Stag supports AWS, Azure, and Kubernetes, whether on-premises or in the cloud, and today covers what Bird called roughly ISO network layers two through seven, with application graphs built from code inspection coming soon. Teams can run simulations while writing infrastructure as code, again in the continuous integration and continuous delivery (CI/CD) pipeline, where Stag can block unsafe changes, and once more after deployment to confirm the result matches the intent. Stag rescans the environment every few minutes, so it also captures changes made by hand through a console or script, and evaluates each change for availability, performance, operational toil, and security.
When asked to compare Stag with tools such as Prisma Cloud and Wiz, Bird drew a line between Stag and tools that look for misconfigurations. The problems Stag finds usually involve settings that are each correct on their own but cause failures in combination, so it relies on no predefined rules or reference architectures and is designed for environments full of legacy systems and tech debt. Bird described Stag as a causal digital twin that works like a physics simulation, with causal relationships in the graph standing in for physical laws, and said the number of possible configuration combinations is too large for brute force or sampling. Stag recommends fixes specific to the customer's environment but does not apply them automatically, because it cannot yet tell whether a risky change was intentional, a typo, or the work of an attacker.
Presented by Rob Bird, CEO, Stag, and Rick Richardson, Principal Engineer, Stag. Recorded live at KubeCon North America 20026 in Salt Lake City, UT, on November 10, 2026. Watch the entire presentation https://techfieldday.com/appearance/stag-presents-at-tech-field-day-at-kubecon-north-america-2026/ or visit https://techfieldday.com/event/kubecon26/ or https://stag.io/ for more information.