I'll take a quick stab. First off it's not a software computer system. Their are so many different elements in play. Your robot needs to know about soil science, agronomy, pests, weather predictions, stock markets, machinery, the list goes on.
Farmers have to wear many different hats. You are a seed scientists for part of year when you figure out which seed you should plant for next season. This choice involves many other factors (weather predictions, personal yield goals etc). During planting season you have to make human decisions of when to plant/not plant. Training a robot to make this decision would be very complicated and most likely involve many 'failed' attempts that cost tons of $$$.
Lets say you did get the robot to choose a good seed, then you hired a fleet of robots to plant corn on 5000 acres and they successfully chose the correct times, right plant depths, row spacing, seeding rates, and managed to get the crop fully planted without any hiccups.. At this point you would then need to be scouting the fields, most likely putting down some N after planting. Robots have to decide which fertilizer goes where, how much to put on, is each field the same or does field X get higher N on one side but less N on other??
THEN lets assume all your crops grow fine, weather behaves (never does)and your robot fleet has scouted all fields. Then the robots must go out and harvest the grain. The combine picks up grain which then gets dumped into a semi on side of the road. Again maybe there is a 'harvest semi for Uber' kind of service where automated semis drive around between combine and elevator at will.
Another decision to be made is selling the grain or storing it...
In short, there are way too many decisions and farmers (i'm thinking midwest corn/beans guys) have to make across many disciplines. These guys have an intrinsic knowledge about their land that has been accumulating for 10s or 100s of years!