"Managing" and "talking to an API" are not mutually exclusive though.
Yes, firmware has continuously become more complex. Yes, if you go back far enough (quite a long ways) there wasn't any.
Peripherals have always been a black box that increased in complexity over time. That increase in complexity does not imply a decrease in management complexity on the part of the kernel. Far from it! Modern device drivers are far from simple.
> They're not our devices anymore. They're intel's, nvidia's.
This is arguably true, but it is also a rather separate topic of discussion.
> They dictate how we use them.
That's largely only in theory. Now if you had said that Apple or Samsung were dictating how we use our phones I would have been inclined to agree. But I don't think gating certain features in the CPU or GPU for the purpose of market segmentation qualifies as dictating how I use my device. I don't like the practice, but I can't deny that I am able to use the APIs provided by the device in an arbitrary manner without it phoning home to the manufacturer or otherwise authorizing the specifics of their use.
> But the OS is not the one sending those signals.
Depending on how you define "sending those signals" and where you consider the boundary between sender and receiver to be you could reasonably argue that the OS never did that to begin with, or alternatively that it has always done so and still does. It's really quite arbitrary and depends entirely on where you consider the boundary of the device to lie.
I purchase a peripheral. It is a black box that implements some device or manufacturer specific API. The kernel has a device driver that abstracts over this and provides a generic userspace API that will (hopefully) remain relatively stable for multiple decades. That's the extent of the contract and that hasn't changed at all.
The device driver situation is already nearly unmanageable. Imagine how much worse it would be if the kernel needed to manage every last minute hardware detail down to the model and even sub-model variants. For example, for every USB mouse and keyboard, past and present. And that's before we even consider things like the firmware for the USB controller on the mouse, which in all likelihood is its own modularized unit from an entirely different manufacturer. But we're going to need to account for every last detail of that ourselves if we fully commit to the "all opaque firmware bad" route. After all, for the kernel to "truly" be in control of the hardware I suppose it will need to manually manage every last pin that falls under software control.
Technical accuracy and nuance is really quite important here. There are many different nefarious things happening at once. Conflating them only serves to confuse the discussion and leads people to (wrongly) believe that there's no need to worry about those weirdos ranting and raving in the corner.