But if you want to make pronouncements about how irrational agents actually act, a prerequisite to having any sort of scientific confidence.
People on HN, not being economists and usually in a polemical mood when this topic comes up, generally strike me as thinking that "people are not entirely rational -> people are entirely irrational -> all possible theories of economics have now collapsed to the very simple thing I already believed", especially convenient when the latter amounts to "the benevolent state should manage everything because people are dumb". ("And let's not think about the fact the benevolent state itself is made out of people.") But that's wrong; humans are not "generally" irrational, they are specifically irrational in very human ways, and the introduction of non-rational agents greatly expands the possibility space of economic theories, it does not contract it to an obvious point.
You could narrow it back down to roughly the same sized space as we started with if you could precisely characterize the exact ways in which the agents are irrational, but we can not currently do that. There's a long history of the very irrationality of humans getting in the way of determining what behaviors are irrational! Or, in other words, it is not necessarily obvious what is irrational. A simple example is loss aversion; is it irrational? Well, when you consider that in the real world, "0" can mean "starving to death", suddenly taking small, safe bets in preference to large, unsafe bets even when the latter nominally has a larger expected payoff can end up being rational, inasmuch as the former actually results in a longer expected lifespan. Determining "rational" is really quite hard, and I rather expect more "human irrationalities" to be determined to be "more rational than the putatively-rational simpler model" in the future. Along with a heavy dose of "yes, this is irrational, and we have this in the species because it was harmless up to about 100 years ago".
You can't map irrationalities to macroeconomics with anything less than this scale of simulation; the entire point of the rational agent was to make simple equations model things, the very point of abandoning that model is recognizing that the economy can only be modeled by more complicated models that aren't captured by those equations.
Unfortunately, the universe does not have a law in it that says we will have the power to know what it is we want to know in any practical manner.
And 100s of millions was just an example. Just as rough in the 1980s would have been the lack of practical programming environments. You can still do 100,000 agents today with a reasonable amount of state in a simply NumPy script that runs on your local desktop. That was not true up until fairly recently, because there would have been no solution as powerful as NumPy running on modern hardware.