Session Name: AI and Infra-as-Code: Private Repos Mean Poor Training
Infrastructure is unique to the organization creating it, and the code to create it is, for security and propriety reasons, usually private. But AI assistants like GitHub Copilot are trained on public data. For application code, this isn't a problem; there's a lot of great public application code. But when the only public sources of infrastructure-as-code are the examples in the docs, it severely limits the usefulness of AI assistants for platform/ops engineers! This talk will look at specific examples of the difficulty of generating infrastructure code when setting up a Kubernetes cluster using AI assistants, including the Cloud vendor code and the resources for in-cluster tools like certificate managers and service meshes, and why that difficulty exists. It will also go over ways to work around these limitations, using custom code samples to generate boilerplate and implementations specific to your organization.
Dave Sudia went from DevOps to DevRel; in both roles he has had to stand up infrastructure in repeatable but constantly evolving structures, and is the world's biggest fan of Infra-as-Code. By day you'll find him enabling developers to do their best work and by night you'll find him hanging with his kid, whose hobbies are now Dave's hobbies.