I agree with this. But I really see this from a business perspective. I think k8s works better at companies with large, complex engineering teams and infra. A lot of companies simply aren't ready, nor will they benefit from, introducing k8s into their stack.
GKE/EKS/AKS lower the bar by a huge lot. There is a ton of value in having just one platform, toolchain, deployment language, set of templates/blueprints, set of dashboards, set of alerts, target for hardening/securing, etc. pp. One of the big upsides of this for me is that it becomes a lot more tractable to get ordinary devs and others involved since you only have to polish things and train people once.
I've found it's quite neat for e.g. data scientists to be able to toss a quick yaml into git and have their thing running to their specifications in non-prod. Things like flux and Kyverno make sure they are boxed in tightly enough that they can't cause a lot of damage, and if that thing works out, someone adds CI in front of the yaml and some templating to the yaml itself and off we go to prod, and the result is still quite comprehensible, so people are enabled at least to ask the good questions to the right people. That's what I see as one of the killer applications of Kubernetes, to build a lingua franca interface for all things ops that doesn't change (much) for different service levels and even totally different needs, like frontend devs and data scientists.
It's probably not a great choice for a three-person startup trying to gain traction by moving fast, but IMO the threshold where k8s starts making sense is a lot lower than many here appear to think.
I agree with this. But I really see this from a business perspective. I think k8s works better at companies with large, complex engineering teams and infra. A lot of companies simply aren't ready, nor will they benefit from, introducing k8s into their stack.