Ramp and Stripe have both published impressive background agent metrics — Ramp's AI writes over half of all merged PRs, Stripe ships 1,000+ AI-authored PRs per week. But both companies spent years building bespoke infrastructure most engineering teams don't have: isolated VM environments, custom image registries, secrets management, network-level security, and parallel orchestration. The post breaks down each infrastructure layer required for production background agents (dev environments, security, network connectivity, automation/scheduling), explains why common primitives like Kubernetes and CI runners fail under agent workloads, and positions Ona as a managed platform that provides all these layers out of the box — deployable inside your own VPC on AWS or GCP.