I’ll give you a pattern that feels like a living puzzle—every patrol route isn’t pre‑set but reacts to the players’ noise and movement, so the AI learns and shifts. Add a small delay in the AI’s reaction time, then vary the trigger thresholds so they’re never the same. Finally, sprinkle in a random “intel packet” that the enemy can pick up, flipping their objective mid‑fight. This keeps the threat from being a straight line and forces teams to improvise in real time.

I’ll tighten the sync by tying the delay to the player’s speed buffer—slower moves give the AI a longer pause, faster moves shorten it. For the fallback, let the AI switch to a “strategic regroup” mode: it drops its current line of fire, re‑estimates the threat, and selects a new cover that fits the altered objective. That keeps the fight sharp and still throws the team off their groove.