All that possible and it just up to you which route you choose and which solution you choose to use.
We have about 100 billion neurons and 1000 billion glial cells(insulators and other functions).
The whole neural network of a fly is about 100'000 neurons, so you have just enough to replicate virtual fly swarm for a million flies in one king of the flies. And if you decide to place a direct control in glial cells you have for 10 of such swarms if your density is 1:1.
But you probably need to tamper a thousand or few thousand of neurons in a fly to create in those flies some urge to do stuff you need and for them to act. There are guys which were capable to live with a half brain so you have plenty of space in that regard.
Yes, bandwidth can be a problem in that situation, but neurons usually fire at rates of 10's-100's Hz, so we may talk about 1000Gbit throughput. We do have links at 100Gbit, and it is not a limit.
A little optimization, you do not need to control each firing of the network of neurons, send patterns once a second and it gets you to 10-100 times reduction, 100Gbit in our case, you do not need to control each neuron, so another 100 times reduction probably, and we talk about just about 1Gbit.
And as AlexP correctly pointed about the hierarchical system, control one, and let others creatively repeat, another reduction(maybe 10 times, maybe 100 times).
Also, why not to use their neurons for some positive activity, which helps you to manage them, so they create some biological processing unit swarm with visual recognition and awareness of the goal and the target.
So we may reduce it further, This mind-reading AI can see what you're thinking - and draw a picture of it - we already have systems which "read the mind" of a human
So you need a slightly improved version of that which works on human batteries(if it is your goal, sort of natural appearance of the capability) and creatively translates your wish/command it to commands, patterns and swarm optimizations for the units you have the control over.
And the system supplies you with useful feedback, you really do not wish to know what a fly thinks about Kant, and so it goes not better to know what each flies thinks about Kant.
But reports about tired, losses, overall recreated/processed picture can be useful.
And your urge for the task can be a thing to suppress their need to feed themselves, or if you have time you can be more gently with the swarm. All that can be translated to voices pictures feelings sense in the matter you prefer, in a matter which is more convenient, including LSD like warping of perception. So you absolutely free here in that regard. You can dive in any level of insanity - like symbiosis, a man dream to be a butterfly, a man dreams to be a devil possessing bitterly's and jumping between bodies(which not necessarily represent actual insects, but their virtual projection of typical problems they have atm if you need a finer control over swarm), a man small like ant talks to them and shows them what to do.
The form which feedback takes is entirely up to you.
Managment of the swarm, and practical implementation, it is the slave of optimization from an informatical standpoint of view. But you definitely can go away with 1-10Mbit throughput, or even less, and size of swarm does not matter that much in the case, but you can lose granularity of control, but it depends how sophisticated systems you use, how effective are your bots at utilizing insects functions, how effective are your algorithms.
but potentially we can have a million fly swarm as smart as a human or even better if we turn the table around, so you can set the task for it like for a human, and it can be fine in controlling the executions of it, like the human you initially willing to update in the question.
More than that, it can be a projection of the human to the swarm, sort of its digital copy on a biological media of the swarm, tampered in the way to execute the goal - and you even can depict it in a way as - if there is a need to catch a desire of to do something, or to meditate - you really have no bounds in depicting stuff how it may look like.
Basically, anything you can imagine how you would manage the swarm(on your own, being the swarm) can be done. Any state you may imagine to trigger that behavior may indeed be the trigger.
There are some limits - really you have to imagine stuff as if would be real to prevent you crossing those limits, to feel it. You can't imagine listening to millions of voices(you can think to be able to do so, but you cant actually imagine that - if it makes sense) - your limit it probably 10 flows, maybe it is better for the guy 100-1000 but that's it. Those flows can represent groups of insect clustered by certain demands(by a system which controls and manages them for you), simple as simple those animals are - hunger, loss, light, moving, ... - a choir of voices which merges in a choir of voices you can distinguish individually. Like in those stories when forest and trees talk to a human by their noise of their leaves in the wind.
Their initial states and inputs are processed by the network of your bots in them and compiled in higher abstractions which require less bandwidth and which are recognizable by the guy.
How many different types of primitive life you control, it does not matter as long as you have proper software for that, proper algorithms suitable for each kind. You can teach your algorithms on the fly to cooperate, but sort of imagining how they do that, and it may be considered as programming of the swarm. But also you need to have a robust invasion bootstrap system, which determines capabilities of host insect, does it fast, and determines the kinematic model, does adjustment translation of typical commands. It also can be part of an interaction of a human with some virtual insect or feeling like that insect. And there should be an even deeper base for that actual bots - some high-level programming, assembler and actual bot as device out of matter.