Container schedulers are fast becoming the new base infrastructure on which we run cloud native applications. While schedulers such as Kubernetes, Mesos or SwarmKit can run application components much more effectively than traditional hypervisor-based cloud technology, we haven’t figured out yet how to use these technologies’ full potential. Infraloop is an attempt to show the way forward by putting programming back into programmable infrastructure. While most of the scenarios for which we use schedulers today are aimed at keeping a static infrastructure running and updated, we aim to show more sophisticated scenarios like multi-cluster autoscaling and failover. To this end we have created a framework for writing complex scheduler logic in a scheduler-agnostic way. This enabled us to implement an application and a set of policies that leverage multiple clusters of different types to achieve maximum efficiency and robustness. Infraloop is not a finished product but a glimpse into the future, a showcase of what is made possible by the combination of lightweight containers and schedulers.