Watch on YouTube
Watch on Vimeo
Rick Richardson’s demo showed how a one-word mistake in a routine security change can quietly break a Kubernetes application. In the scenario, a team locks down a multi-tier application with a Terraform change that denies all traffic between pods by default, then opens specific paths between tiers, including back-end access to its Redis cache. The 60-line change looked correct and would likely pass human review and automated testing, but the policy targeted a label of “redis” while the Redis pod was labeled “cache,” so the back end would lose access to its cache. Richardson noted the failure might not surface until the application came under heavy load and needed the cache. Stag selected 71 simulations by choosing the data flow paths in its causal graph that mattered for the change, and it flagged the change as unsafe in 44 seconds. When a delegate asked how Stag chooses those paths, Rob Bird said the method isn’t a probabilistic model and called it the company’s trade secret. He pointed to the 98% accuracy figure, which he said came from measurements by Stag’s design partners, and argued that catching 98% of problems beats the current alternative of no simulation at all.
Bird then walked through the Stag UI. The overview page shows the master change history, including how many changes were blocked as unsafe, which risk areas they touched, and whether they came through the continuous integration and continuous delivery (CI/CD) pipeline or were made by hand. In Stag’s test environment, only about 25% of changes came through the pipeline, and Bird said customers of every size report that 10% to 90% of their changes are manual. The overview also shows blast radius, and in Stag’s small test environment some side effects reached 17 hops from the original change, mostly through identity and access management (IAM) roles, with 5% of changes reaching between 10 and 17 hops. Each change has an inspection report with simulations, affected resources, and metadata history, plus a chat that explains failures and suggests environment-specific fixes, such as using Amazon Route 53 Resolver for a DNS change that broke routing. An Explorer tool lets engineers ask questions such as the blast radius of their DNS, and Bird said AI coding tools connected to Stag use Explorer for about 70% of coding changes and the simulator for about 60%. Customers also told Stag the shared view of simulations improves communication between teams such as DevOps and security, because everyone discusses the same evidence.
Bird described a three-stage process: modeling the enterprise, simulating changes, and showing what is unsafe, why, and how to fix it across design, coding, review, deployment, and post-deployment verification. Stag models cloud, IAM, network, and Kubernetes configurations, along with database schemas and access; application I/O and API surfaces come next, but it does not try to understand business logic. Stag is about to launch on AWS Marketplace and is not software as a service (SaaS), so customer data stays in the customer’s environment, and it can use AI services from the customer’s cloud provider, such as Amazon Bedrock, or models running on the customer’s own GPU servers. Bird said Stag currently favors a year-old stock Kimi model, which he claimed outperforms current frontier models when it has the simulator and causal graph to work with. Stag connects to coding tools and pipelines through the Model Context Protocol (MCP) and a command line tool, and deploys in five to 30 minutes with no ontology building required from the customer. Bird closed by citing two figures: about 60% of customer changes are unsafe, and about 96% of developers don’t fully trust AI-generated code because it lacks deployment context, a gap he argued simulation can close.
Personnel: Rick Richardson, Rob Bird
Thank you for being part of the Tech Field Day community! Our mailing list is a great way to stay up to date on our events and technical content, and we appreciate your signup.
We promise that we’ll never spam you, send ads, or sell your information. This list will only be used to communicate with our community about our events and content. And we’ll limit it to no more than one message per week.
Although we only need your email address, it would be nice if you provided a little more information to help us get to know you better!