Lots of PaaS offerings have been doing this for a long time, RedHat OpenShift being the original one; they contributed a lot of 1.0 k8s code and provided a lot of key early architectural feedback by implementing a PaaS using k8s. Another early one was Deis, who were at the first KubeCon.
All this is to say that you're at very least certainly half-correct, in that k8s is a very flexible tool that can be used to build a very simple, elegant, and ergonomic PaaS.
I'm not sure I agree that it's inappropriate for most businesses though, unless you think that only a PaaS like Heroku is appropriate for most businesses; the analogy I'd suggest is "Heroku vs. running your own VMs" circa 2010. Heroku is great for getting started, and lets you move fast by abstracting away a bunch of infra. But it's also restrictive; you can't pick and choose your components freely. As you grow past a certain point you'll almost certainly need the flexibility (or just cost-effectiveness) that you get from running your own infrastructure.
K8s is an improvement here because you can run a managed cluster on something like GKE, which takes away most of the operational toil, while still giving you a lot of flexibility on what components / pieces to include. The k8s domain API does a great job of abstracting away true infra concerns like volumes, compute, scheduling, load-balancing, etc, while making it really easy to package, use, and iterate on the stuff that sits on top of that infrastructure.
I'd probably not encourage a seed-stage startup to use k8s unless you're very familiar with the tool; a PaaS like Heroku would likely be more appropriate. However at the point that you'd usually graduate to running your own VMs (wherever that is on your own company's developmental path), I'd say that using k8s is now a better choice.