In some respects K8s is a bit of a step backwards from Heroku or AppEngine. It's complicated and requires you master its learning curves to deliver benefits. This looks like a lot of cloud products which are becoming more complicated to sell to bigger companies where they have the ops teams dedicated learning these products. You get an 'easy' to setup demo with a lot of copy-paste or bash|sh but you'll need a PhD for running your own apps in production. And each product is a different learning curve and they don't always play nice.

The next paradigm is composability and simplification. I should be able to define an application, its configuration, what it needs to talk to, and level of desired spend for which availability/performance SLO and it should mostly just work. Security should just work and each perimeter is automatically configured - without managing separate configuration for admin APIs, end user auth, peer auth, k8 network policies, firewall policies, etc. I shouldn't have to care about runtime - GKE vs Cloud Run vs Lambda or optimizing Node shapes or whatever new product launched yesterday. If I push a production change there should be a consistent way of simulating, canarying, and slowing rolling that change out across all products.

Ideally, it should basically be a marketplace of commodity network/compute/storage and the providers are dynamically bidding for my business. If you can run my application as fast or as reliable with less spend, you get the business - if not the app magically moves the provider that can.