I was thinking of a whole bunch of funny replies, but I think a more serious answer would be deserved too.
First, there's the problem of inputs. A conventional AP uses the inputs that are found in the conventional sixpack: some mechanical gyros and the pitot-static system. Maybe a flux gate compass or something like that and some input from the GPS. In aerobatics, conventional mechanical gyros are out of the question in any case - you need something solid-state like for that, with unlimited freedom. Laser-based maybe? I'm not an expert in those. But that would tell you your attitude - important for all those ZLA figures. Oh, and the flux-gate compass would need to work properly in all attitudes, even 90 degree up or down, to do those 1/4 rolls in downlines properly.
Another input you'd need would be for a g-meter so that you can enter your loop with a fixed value, say 4G, and reduce the Gs over the top to 2Gs.
And the third additional input you'd probably need would be an Angle of Attack sensor, so you know exactly when to pull the stick all the way back to enter a spin or snap roll. (The alternative would be if the AP would know the exact aircraft mass, density altitude and such, to be able to calculate the exact moment of stall based on pitot-static data only.)
Of course, I'm assuming here that the AP has direct GPS input as well to know the exact location of the box, to calculate wind drift and so forth.
The second problem is one of output. You would need proper three-axis control by direct servo. Pitch, roll and yaw, with servos that can whip those controls from stop to stop instantaneously. None of these indirect controls where the AP slowly winds a trim wheel up or down. You'll want autothrottle too, especially if you have a fixed pitch prop.
And you'll definitely want to program the airframes limits into the AP. Not just the normal g limits, but also limits on rudder loads (snap rolls) and wing root bending. And maybe some limits on what the occupant (let's not call him pilot in this case) can sustain.
If you have all those, and have gotten it certified, I'm sure a clever person could write a program that would take your Aresti sequence, allows you to input a few tradeoffs in that sequence between altitude and speed, and then come up with an AP recipe.
Question is, would that be fun? I used to play chess but I gave up when I wrote a chess program during my computer science studies, and it beat me.