AO units are fundamentally limited by the isoplanatic path, which is just few arc seconds across in the visible band (most of the time, seeing being a stochastic problem by nature).
Most professional AO systems work in IR bands where it is "easier" to deal with seeing and "larger" isoplanatic angles.
A key figure of merit is the ratio N=D/r0 between the scope aperture D and r0 known as the Fried’s parameter, or Fried’s coherence length, which is a diameter (not a radius). Typically r0 stretches from 5cm to 10cm, with r0=40cm or more few exceptional sites.
r0 = 7cm being an average value for most of us.
If N<=1 the scope is essential diffraction limited, for N>1 it becomes seeing limited. For N <3 roughly the seeing is mainly of tilt/tip in nature (the guide star wander for short exposures), a large N requires higher order wavefront correction (beside tilt/tip). In any case the isoplanatic angle is quite small, unless atmospheric turbulences are limited to the boundary layer (near ground). The AO corrections rate needs to be in the few 100 Hz to truly correct seeing inside the isoplanatic patch (again most of the time). Unless you try to resolve double stars few arc seconds apart, AO for amateur astronomers should be seen as an imager stabilizer device for mitigating mount and other left over system errors, correlated across the all scope FOV.
Here is a link with an interesting example related to the isoplanatic angle from the Palomar AO system (credit R. Dekany, Caltec):
http://www.innovationsforesight.com/...eing-tutorial/
The best way to limit seeing in auto-guiding is to use long exposures (>10 to 30 seconds). Mount tracking error, flexure and atmospheric refraction can be deal with PEC and with system modelization such as Tpoint and Protrack.
However nothing truly replaces a good mount, AO helps to mitigate short time, “high” frequency, mount mechanical errors (noise) not corrected by PEC. On the other hand AO may require using quite bright guide stars (for short exposure) and may lead to more risk of “chasing” the seeing outside the isoplanatic angle, which in turn may blur the image more than without any AO.
Bottom line the seeing will eventually limit the guiding quality.
However NIR offers a unique opportunity to improve the tracking accuracy.
Longer wavelengths are less sensitive to seeing and its effect scales as the 6/5th power of the wavelength.
Let’s use 550nm for the average visible wavelength (classical guiding) and 850nm for the NIR one (for instance ONAG guiding), this leads to an improvement of:
[ (850/550)(6/5) -1] x 100 = 69%
Which translates to smoother tracking.