I’d prefer a Neural Network that regularly evaluates outcomes and can make adjustments without spawning new objects.
Also that the Neural Network’s evaluation can be checked by the player as part of the scoring screen (for what you contributed to its pool for that playthrough) and on Featured Page for worldwide figures.
The idea is that the Neural Network pools successes and failures… noting what guards did and whether or not assassins succeeded or were killed.
The Neural Network can then adjust tactics and guard placement but cannot change guard types or spawn new items or guards nor can it re arrange other assets like cameras and recorders.
Eg: You check on Featured the Worldwide Security Monitor and note that Viktor Novikov chandelier kills in the main floor of the Palais has gone up such that 30% of the last 100 playthroughs worldwide has used this kill.
There is the probability that CICADA will increase security coverage of this area. But because the set of guards is finite it means other places may have fewer guards.
The AI also has a finite set of enforcers so moving one here will remove one elsewhere.
Note though that the Neural Network can only account for Crimes Noticed and other detected illegal behavior. Accidents, Poisons, and Undiscovered deaths will result in no evaluation.
Also remember that the game evaluates global results locally. So just because it happened in a playthrough of yours, doesn’t mean the same changes happened elsewhere even if someone plays HITMAN 2 simultaneously.
Btw the game already does this conditionally. One of the surgeons in GAMA Operating Theatre is sent upstairs if and only if 47 is disguised as Jason Portman.