Agricultural robotics and employment
At least with regard to agriculture, the effect of robotics upon employment depends on the approach taken. If your goal is to further reduce the number of people deriving an income from farming, and you are willing to accept any other sort of expense to that end (autonomous tractors for instance), then you can probably manage to reduce the percentage of the workforce engaged in agricultural production to an even smaller fraction of 1%.
If your goal is to maximize the production of those crops that are easily produced and handled in bulk and survive long-term storage well, in the interest of generating return on capital investment and foreign exchange, and only care about how it’s done insofar as that impacts the bottom line, you might conclude that capital expenditures to further minimize payroll would generally not be cost effective, that it would cost more to replace the remaining workforce than to keep it.
However, if you’re interested in guaranteeing the sustainability of production far into the future, despite climate change, while also halting soil loss, ending the use of poisons, preserving remaining diversity in both crop and native genomes, and rebalancing production for healthier diets, you may need both more sophisticated machinery and all the people you can recruit.
Such a complicated goal implies complex operations, and complex operations imply a large variety of tasks, some easily mechanized and others common enough to make mechanization worthwhile, even though challenging. Those that are neither common nor easily mechanized will fall to human workers, farmers and farmhands, who are far more adaptable than any machine.
At some point in the future it may become possible to build machines adaptable enough to take the place of a farmer, but until the annual cost of ownership of such a machine drops below the annual cost of one human worker, it won’t make economic sense to deploy them, and without an infrastructure to drive down the cost of robotics, that may never happen.