I posted this on Reddit as well:

Great article. I would have liked to see Dataflow get a shoutout for event-based workloads, but it's a little more ETL-related than this article is covering.

The "scale to 0" likelihood is a great rule of thumb for choosing function/run over gke.

I'll say that we do one GKE cluster per business domain, giving us 10 or so clusters with only 5 teams depending on how you count them. It doesn't seem to be too much an issue (although I'm on the application development side, not cloud ops). It's all set up in terraform.