These last few pages regarding autopilot use has been quite thought provoking.
Despite considering myself reasonably experienced, I am acutely aware that there is much I don't know and have spent some time considering how I fly and whether it could be improved upon. I currently fly a modern 4 axis AP machine similar to those employed on the NS, but in the corporate role.
So some thoughts:
1. When I fly a 3 axis NPA, I do so with airspeed (AS) hold engaged and I control the RoD directly with collective. It would appear that most of you on the NS, fly with VS engaged, so I have considered why I am at odds with all of you and the following comes to mind:
a) I have tried vertical speed (VS) and have found that the rate of change of VS is too slow (I experimentally tried it on a PAR and there was no way I could control the RoD at the rate the controller required)
b) The controls work in the "traditional" way - ie collective (flown by me) controls RoD. The cyclic (via AP) holds my desired speed.
c) It fits historical expectations - on an initial IR and subsequent LPC's, the examiner asks you what speed you intend to fly the approach, implying a steady speed. By using AS hold, you are getting the AP to do that for you.
d) Finally, and here my knowledge is a little ragged, and my aircraft RFM is of little help - I believe that if you fly an APP uncoupled, BUT with the Flight Director engaged, that the collective demand bar is telling you what to do to maintain GP (on an ILS for example) and the pitch command bar is showing you what to do with cyclic to hold speed - ie "traditional" flying
2. I am one of those pilots who occasionally, when autopilot hasn't done what I expected, have decoupled and flown myself (in fact I did so on my last LPC). Generally, I have regained parameters, and then recoupled. Again, I have asked myself why and come to the following conclusions:
a) Controlling the AP when everything is going as you expect is straightforward - when things go awry (and I admit, most often because of operator error), recovery can be slow.
b) I have NEVER, EVER been taught Unusual Attitude recovery by using the autopilot have you? And at what attitudes/speeds etc?
c) I treat the AP as my co-pilot ("George"). If George doesn't do as I expect (again, most likely because I've incorrectly briefed him - ie pressed wrong buttons/sequence/set wrong datum) and its not possible to quickly correct through autopilot manipulation, then I take control - that's why I monitor it/him.
Going back to the first point, I understand that collective can have marked effects on other axis, but smooth application can still result in rapid response with easily controlled secondary effects.