The topic is the subject of analysis in other disciplines, especially around Business Administration and Economy.
The use of the word agents is interesting is mostly a coincidence, it is used today in a sense that didn't quite exist 2 years ago (and that definition isn't yet formalized). We know that in economy the term agent was used to refer to people or organizations, possibly to programs especially in trading, but usually in the context of purchasing decisions or simulations. Of course it was used in an adjacent sense before, but in a way that isn't different to other similar words like "entity", or "decision-maker", or "being".
We can see that agents are used in this sense "The three largest nodes may represent countries, or buildings, or software agents"
In the context of agents that are computational, this has been discussed as well, especially in OOP, early OOP texts from Kay make parallels between Objects and cells, or create examples of Objects as office workers with specialized knowledge.
The phenomenon talked in this paper makes me think more of "the algorithm" as used in common parliance, rather than modern LLM agents. While these algorithms were usually controlled by a single company, this mode of analysis would consider a company as an agent as well, but it interacts financially with consumers, clients and in the case of public companies, through stock exchanges (which are connected to global markets at high speeds through HFT).
The math goes over my head, but I would say that if someone looks into it because of the current agent craze, it might be worth it to look into the broader intersection with economics, and look into the classical etymology of agents, rather than diving deep into this article just because of a deceiving word coincidence that gives the appearance of prophetical.
Agents were a thing in AI research decades ago. See for example the volume "Designing Autonomous Agents" from 1990 or the mountain of works on agent-based modeling. The phrase "multi-agent systems" goes back to the 1990s or earlier.
I'm sorry but it's wild to me that you could write so much about "agents" without recognizing their long, established history in computer science (especially in AI) outside of OOP. Agents are basically the entire framing of Norvig and Russel's "Artificial Intelligence: A Modern Approach" [0] (originally published in 1995, but drawing from much earlier work).
Not specifically AI, but not unrelated either, Agents play a major role in how we understand concurrency and mobile communication. The author of this paper, Robin Milner, is responsible, among many other things, for establishing the π-calculus (1992), which defines a formal language to describe agent communication.
If you want to go closer to the source you can take a look at Hewett's "Actor Model" [2] 1973. Which is when the field first started to formalize the idea of software agents.
The current use of the word "agent" is basically a marketing buzz-word that largely ignores the decades of research in the field of computer science around how to design intelligent interacting agents to accomplish tasks. Which is a bit of a tragedy because I personally think current LLMs could gain a lot of value if thought about in the traditional agent sense.
0. https://people.engr.tamu.edu/guni/csce625/slides/AI.pdf
1. https://www.pure.ed.ac.uk/ws/files/16426053/A_Calculus_of_Mo...
It's only been submitted twice, not "many times".
