Kubernetes Executor Problems (AWS)

I am trying to use my AWS kubernetes cluster as the place where gitlab builds & runs stuff, but I'm experiencing some problems. I stood up the runner with this: https://gist.github.com/dts/cbb93aea7dbfec77ac70a8c24d436253. I did the following to "register" the runner:

concurrent = 1
check_interval = 0

[[runners]]
  name = "aws-kube-2"
  url = "https://gitlab.com/ci"
  token = "<snip>"
  executor = "kubernetes"
  [runners.ssh]
  [runners.docker]
    tls_verify = false
    image = ""
    privileged = false
    disable_cache = false
  [runners.parallels]
    base_name = ""
    disable_snapshots = false
  [runners.virtualbox]
    base_name = ""
    disable_snapshots = false
  [runners.cache]
  [runners.kubernetes]
    host = ""
    cert_file = ""
    key_file = ""
    ca_file = ""
    image = ""
    namespace = ""
    privileged = false
    cpus = ""
    memory = ""
    service_cpus = ""
    service_memory = ""

Which results in:

Running with gitlab-ci-multi-runner 1.6.1 (c52ad4f)
ERROR: Preparation failed: error connecting to Kubernetes: invalid configuration: no configuration has been provided
Will be retried in 3s ...
ERROR: Preparation failed: error connecting to Kubernetes: invalid configuration: no configuration has been provided
Will be retried in 3s ...
ERROR: Preparation failed: error connecting to Kubernetes: invalid configuration: no configuration has been provided
Will be retried in 3s ...
ERROR: Build failed (system failure): error connecting to Kubernetes: invalid configuration: no configuration has been provided

The docs seem to suggest that Kubernetes should auto-discover what's going on. Further, I want to be able to build against the Docker socket that is running on the machine itself, and limit it to one host in AWS (the same host that I have put the gitlab-runner pod on), so that caching happens in a sensible way. I suspect that this is all a pretty bad idea, and I should just spin up a dedicated host to do this (which will be my next path), but I figured I'd try it out since I already have a Kubernetes cluster spun up and going.