What vertical farming and ag startups don't understand about agriculture

I think that framing agriculture's transition (hopefully) away from mono-culture into a more ecosystem focused idea seems like a tractable optimization problem. If we look at the reasons for mono-culture, I would argue part of the reason is that traditionally bigger yield is linked to bigger tools -- tractors are much larger than horses, spraying a chemical is easier when only one thing needs to survive. Monoculture makes it easy to apply big things, harvesting one row of corn is easy to scale to ten rows of corn just by making the combine harvester wider -- the harvester's problem statement is generic and scaleable in this way.

The hard problem, that you raised at the end, is how do we scale harvesting non-mono-cultures. The constraining variables are quite different when we need to perform a set of ten actions with no locality guarantees (Monoculture just guarantees locality of similar actions). I think one natural perspective is to look at how we do things non-locally at scale, which effectively reduces down to a distributed systems problem.

edit: few small changes

I will also note that in 2020, we have no shortage of leafy greens.

Sunlight is delivered as electromagnetic power (watts) proportional to surface area. Plants naturally grow on the surface of the earth, and therefore receive a small proportion of that power which they use to convert CO2 into sugars and eventually plant mass which we eat. Stacking a bunch of plants on top of each other cannot change that the lower plants must receive less power, and therefore cannot grow as much. And that's ignoring the added complexity and logistics (read: overhead) of maintaining a system that stacks plants on top of each other, which would surely obliterate whatever 2-digit% efficiency bonus you can eke out of stacking. The universe doesn't work like Minecraft.

Chemical and water use reduction seem to be a pretty good outcome, as well as being able to ignore seasonality.

I would like to see some numbers on farm equipment (in?)efficiency before throwing that out as a fact. Color me skeptical but it doesn't seem obvious at all that rebuilding a 10000-acre greenhouse every 20 years will necessarily produce less greenhouse emissions than running a few tractors. Or even that harvesting food in a greenhouse takes less energy than doing it with a tractor.

You want to know cropping ag's biggest problem? Too much data. Farmers are collecting all sorts of data - soil samples, weather station data, aerial infrared photos and yield monitor data to name a few. But there are few tools that give actionable information from all that data. Actionable in prescribing something that results in a positive ROI.

Now as an agronomist who soil sampled, walked the field multiple times every year and sometimes even rode the combine with the farmer I was able to do that - sometimes.

Someday it will happen but it's my opinion that AI is a long, long way from performing that job. But I do hope I live to see it.