Adding an Amazon SageMaker icon on your architectural diagram to "do some machine learning" is no different than adding an EC2 to "perform some business logic". It does not tell the full story.
I’ve heard many conversations referring to SageMaker as a cherry on the top of any cloud system, doing the famous machine learning.
Why it may be misleading
It can hide the complexity and effort required to deliver meaningful results, leaving your customer with the false impression that it will be an easy job - it's just a single component after all!
If you've ever used any piece of business software, you probably know how complex it is and you can imagine how much time was spent building it in the first place. Yes, EC2s (or Lambda, Fargate - whatever) are necessary to run it - but they're only providing the compute power. Without the apps on top of them, they'd be pretty much useless.
The same relation applies in the machine learning domain. SageMaker provides dozens of building blocks that allow you to deliver machine learning solutions in any scale - and that's absolutely awesome. I’ve already tackled that in some of my previous posts - go back to it if you need a refresher on a ML-aware toolkit of SM.
However, like with business apps and EC2s, SageMaker is just a compute power / toolkit that only helps data scientists and machine learning engineers build their ML software. It does not magically solve all your problems via a single click or command. There is still a lot of work to do!
Are high level diagrams bad?
Oh - there's absolutely nothing wrong with high level diagrams. Just make sure that your narrative clearly states that a particular component is a simplification over something much more complex.
Whoever reads your diagrams or listens to your presentation, needs to be made aware of that simplification. That’s it. Tell them that SageMaker is not some magic one button to solve machine learning for them.
In short - the next time you sprinkle SageMaker powder on top of your slick diagram, make sure you don’t accidentally oversimplify something that’s not really that simple.