How about harvest? Machines are quite specialized, how do you see that working out?
Like these: https://en.wikipedia.org/wiki/Rice_transplanter https://sda-industries.com/equipment-parts/rice-harvester/
Compare that to "small" machines sold in the US. https://www.deere.com/en/harvesting/
But we don’t build computing like that any more. Instead of large centralized single devices we use a large amount of protean hardware with various adapts (software).
Hordes of small devices can flood through the fields, using sensors to decide at any point what to do, and can use different effectors for different tasks at different times of the year, like the humans used to do.
I mean an autonomous device that picks a row, then does the same on the next row on the way back, then maybe unloads into a hopper and continues with the next couple of rows. And you don’t one, you have 50 or a couple of hundred of them rather than one huge combine harvester.
And if you have 50 of these platforms designed to go up and down the rows they could have been doing the planting at the beginning of the season (or desuckering your vines in spring or whatever) using a different tool, so your capex isn’t tied up in a few huge single-application devices, but amortized over a whole season. And when you have a fleet, if one fails and needs maintenance it’s no more of a big deal than when one server in a datacenter fails.