I don't regularly use AWS services, but it would seem to me that writing operational tooling (deployment descriptors and such) to the Kubernetes API allows more control over the management of your app and more portability to other cloud providers compared to ECS. On the other hand, ECS is a bit more "turnkey", doing more of the detailed control isn't required. It's worth someone with more regular usage weighing in, though.

Probably for two main reasons:

* Open APIs, if you care about being able to move away from AWS

* Tooling. People are likely to build a lot of tooling around Kubernetes, relative to Fargate/ECS.

That's my guess.