A combine is already well beyond the ability of an individual grain farmer to replicate. The best they can do is buy commoditized combines. And those are only commoditized now because of the vast area devoted to cereal crops. The robots required for more specialized crops will be more expensive at first, until some patents expire, and even then the number of potential competitors will be limited by area under cultivation.

You cannot start a new cereal-crop farm without a hefty investment in capital. You just can't. At best, you would be renting other people's robots, and that might be costly enough to prevent you from ever buying your own.

An onion farmer may yet be able to build a better onion robot. If they do not, someone could invent one, and either eat up existing farmers' margins with rental fees, or simply start a new farm and undercut those existing farmers from lower labor costs. But that startup would need to have the robots.

A trunk-shaker harvester robot might be cheaper than human pickers, but they can damage trees and may cause damage to the harvested fruit. A robot equipped with a camera and a vacuum-assisted hand might outperform it for apples, but be useless for olives. The olive farmer might need a robot with flailing rods. It is still possible for an individual farmer to innovate their own robots specialized to a particular niche crop.

We're already seeing farmers in India committing suicide because they can't keep pace with the gene-edited seeds and robotic combines from industrialized farms. It seems inevitable to me that companies like Monsanto and John Deere will be able to squeeze individual farmers on a technology-yield treadmill such that if they ever miss a step, they fall off. And once they fall off, they can never find a way to get back on. Because they are not the people building the robots. They get left behind just as fast as everyone else not building newer technology.