This is a half-baked theory. It doesn’t suggest any particular role for neurons or other physical aspects of the brain. It is about the workings of the mind. It is primarily concerned with the things we learn how to do. 

 Imagine yourself…the mind you are as a volume of processing power: As an environment with a certain amount of ability to respond to the world. This software-like volume is a hierarchy made up of entities with different roles to play at different levels.
The smallest increment or unit is a “Mindon” (see: Making up crap, H. Miller). Metaphorically it is like a cell in the body, an atom in matter or a byte of computer memory. It is there not because of evidence pointing to it but simply because it belongs logically within the structure of this theory as the smallest unit of work. For our purposes, it creates the image of a tiny individual operator with a bit of processing power.
A new and unknown task is encountered. For example, learning a musical instrument, casting a fishing line or properly cutting an onion. The brain experiences it as if with a kind of touch in the processing centers naturally assigned to the sense data involved in the task (Motor, optical, etc).
Executive function does a kind of “importance triage” focusing on the sense data and relevant memories. Mindons begin to swarm and cluster around the task creating a complex imaginative prototype or map of the skill and begins measuring and comparing the experience of trying the skill against the map, making edits in the map as more information comes in and also making edits in DOING the task. Heuristics are noted and retained. With time and practice the map, measurements, and comparisons become more accurate, detailed and nuanced. Plateaus and benchmarks are hit and become a bit like a saved version of a game or a file, the starting point next time. Essentially this is a matured and organized group of mindons forming a stable repeatable task.
These coherent, informed collections of mindons I would call “Agents“. Something like playing the guitar wouldn’t involve a single agent but many. I imagine agents in this example being like proper holding, finger pressure, picking and strumming, volume and tuning, etc. Each of these areas would be an agent.
The whole coordinated group functioning together would be an Agency. I could use other names, I could call the agency a program and the agents, modules. The terminology isn’t very important but I’d like to not be completely bound by computer metaphors.
This system is far more fluid and flexible than any computer system. The agents are not limited to one specific agency, once established, I believe they can flow on demand into new situations that call for them. Somewhere in the mental map mentioned above would be a process of looking for existing “off the shelf” agents that could hit the ground running on the new task. For example, if you have an agency for driving a car or playing the guitar then when picking up a Ukulele or sitting in a go-cart the agency isn’t fully applicable but many useful agents flow instantly into the task.
So, mindons group at the behest of executive function, and form educated agents which group as a whole agency; “playing the guitar”. Singing would be an agency too so when playing the guitar AND singing there must exist a kind of Meta-agency that allows parallel processing and two-way feedback.  All of this is playing out in a larger framework where the musician is also processing things like audience reaction, etc.
This suggests an overall environment or community of mindons, agents, agencies and…departments? I’m not crazy about the bureaucratic feel of this metaphor but the hierarchical structure is needed.