This is fantastic. Right now I use GCE autoscaling groups to spin up runners while I've got several kubernetes clusters running 24/7. It'll be nice to utilize those for builds (especially with how fast the scheduler is) going forward. Saves money, as well.