no such thing as ignorance here - hopefully we are all still learning!
this is what I see:
The AGC operates on downstream signals, after the IF filtering. The sensing part of the AGC does NOT see the AM (medium wave) signals.
The AGC does its attenuation with MOSFET devices. My experience with devices is that their linearity does vary with bias - I am thinking they can be less linear when given an in-between bias. So I think your logic is best - to put the filter first so that the AGC attenuator does not experience large out-of-band energy.
Of course - if your ubitx is not operated anywhere near a many-kW AM station -- then it may not matter at all.
PS - Mr Magoo here (me) with poor vision is straightening out the signal-strength meter wiring connected to a VK3YE circuit. I like this AGC, and the meter is now working decently. Instead of Peter's configuration in parallel with the LED, I have it in series via a shunt resistor. It better represents signal strength than the parallel implementation. You guys can snicker at this 'minority' AGC circuit but it works nicely. Best wishes with whatever circuitry you all are using - this is a cool rig!