I've sometimes thought of deontological rules as something like a sanity check on utilitarian reasoning.

If, as you are reasoning your way to maximum utility, you come up with a result that ends, "... therefore, I should kill a lot of innocent people," or for that matter "... therefore, I'm justified in scamming people out of their life savings to get the resources I need," the role of deontological rules against murder or cheating is to make you at least stop and think about it really hard. And, almost certainly, find a hole in your reasoning.

It is imaginable — I wouldn't say likely — that there are "universal moral laws" for human beings, which take the following form: "If you come to the conclusion 'Utility is maximized if I murder these innocent people', then it is more likely that your human brain has glitched and failed to reason correctly, than that your conclusion is correct." In other words, the probability of a positive-utility outcome from murder is less than the probability of erroneous reasoning leading to the belief in that outcome.

A consequence of this is that the better predictor you are, the more things can be moral for you to do if you conclude they maximize utility. It is imaginable that no human can with <50% probability of error arrive at the conclusion "I should push that fat guy in front of the trolley", but that some superhuman predictor could.