AFK Infrastructure Ralph Loops Meet Crossplane Tobias Kässer, Yury Tsarev

Converting Terraform modules to Crossplane Configuration packages is tedious, error-prone, and deceptively complex: mapping HCL to XRDs, translating resource graphs into Compositions, ensuring functional equivalence. For terraform-aws-vpc, this takes weeks. This talk shows how Ralph — an autonomous AI agent loop — can do this largely unattended. Ralph spawns fresh AI coding agents against a PRD, with progress persisting in git, not the LLM's context. Each iteration is guided by type checks, unit tests, and reconciliation against a real Crossplane cluster provisioning actual AWS infrastructure. We'll walk through converting terraform-aws-vpc to a Crossplane Configuration — still three weeks, but with dramatically less hands-on time. Attendees learn how to structure PRDs for autonomous infrastructure migration, wire feedback loops to Crossplane's reconciliation model, and where AI agents reliably struggle. A field report — honest about both promise and limitations.