I'm always worried about the "Midas effect" with something like that. Having to be careful not to trigger actions you don't want to.

Yep, and IMHO that's why eye tracking even though is amazing when it does work is still very limited. We do a lot of small things without even realizing and stopping to do movements we are not aware of is very very hard.

Does it really have enough fidelity of distinguishing gestures for that today?

I was under the impression the set is pretty limited and maybe not completely robust (like per person setup/calibration etc.)

I think the inertia of familiar input devices is way too high for something like that to gain any sort of traction. For better or for worse, the mainstream user is probably not leaving his QWERTY keyboard and his touchscreen, even in use cases where those are highly suboptimal.

Keyboards are wildly suboptimal. Inertia is only an issue when you replace suboptimal with only slightly better suboptimal. The transition period from blackberry (keyboard-based phones) to iphones did not take long at all (relatively speaking of course).

suboptimal for which tasks and what's better then?

I spend a LOT of time with different interfaces (eye tracking, 6DoF controllers, pens, touchpad in keyboard, ergonomic mouse, etc) and... basically if there is 1 thing I'm confident about is that input methods are very task dependent. If you want to input text, especially hard to pronounce keywords and symbols, keyboard is great, if you want to move a pointer at arbitrary position in 3D keyboard sucks. It also depends on the context, behind a desk vs on the move is totally different.

So yes quite curious about such a wide statement.

IMHO inertia isn't the biggest factor, training requirement is.

I switched recently from mouse to touchpad and ... next day my great ergonomic mouse was basically in a drawer. Meanwhile when I tried to use Tap Strap the first step was to learn combinations with basically no gradient of feedback, it was either success or failure.

Inertia isn't such a big problem if the investment is small but if you have to learn an entirely new way then it's a huge risk.