I think author is totally ignoring the maintenance overhead that kubernetes comes with compared to ECS. Getting up and running with the most complex Kubernetes setup is easy for a long time now. For that, there are tons of CLIs that would boot strap a cluster in a cloud and if not that, you have very actively maintained Ansible playbooks and what not.

It is the upgrade path where the friction comes from.

True. upgrades is a huge overhead. But you also have K8s management tools for that where they manage upgrades for you. I know it's added cost, but worth the benefits.

Could you point me to some such tools? That would be an interesting prospect if it really works, every single time.