I think we should judge a technique in regard to the problems it solves and the ones it creates, not in regard to our understanding of the said technique.
No, it is not the same as using "inline styles" (just think about it). Yes, it reduces CSS scope and bloat in styles sheets. Yes, it breaks the Separation of Concern principle. No, it does not create bloat in the markup. Yes, it forces you to style via markup rather than via a style sheet, etc.
Like with any other technique, choosing to use "Atomic CSS" in a project should be considered in relation to the problems authors have to solve. CSS is a complex matter, there is no one size-fit-all solution and there are not that many solutions either so I think we should think twice before disregarding a tool, any tool.
PS: Yes, I know universal.css is a joke but it is mocking a real/serious approach to big CSS problems.
> [How do you] improve on it after it has been built?
It depends on what you need to achieve. You can rely on "variables" to set values that can then be easily changed via the config (http://acss.io/guides/atomic-classes.html#variables), you can have classes that live in one single place (one single file to edit), or you can simply go "old school".
"Atomic" (Atomic CSS via Atomizer) can be mixed with any other kind of architecture as it has no footprint and it is not content opinionated.
CSS is all about trade-off - you gain here, you lose here. The architecture you choose depends on your requirements. Atomic CSS leads to very small style sheets and allows to style via markup. If you have no interest in those 2 "features" then it has not much value to you.
No, it is not the same as using "inline styles" (just think about it). Yes, it reduces CSS scope and bloat in styles sheets. Yes, it breaks the Separation of Concern principle. No, it does not create bloat in the markup. Yes, it forces you to style via markup rather than via a style sheet, etc.
Like with any other technique, choosing to use "Atomic CSS" in a project should be considered in relation to the problems authors have to solve. CSS is a complex matter, there is no one size-fit-all solution and there are not that many solutions either so I think we should think twice before disregarding a tool, any tool.
PS: Yes, I know universal.css is a joke but it is mocking a real/serious approach to big CSS problems.