What happens when agents become numerous?

Imagine an inbox receiving requests from ten agents. A person can read the messages and decide which matter. Now give that person ten thousand machine correspondents. Their attention budget has barely changed. The rules for reaching them begin to matter as much as the quality of any individual message.

An AI agent is software that can take steps towards a goal: checking a calendar, booking a service, or passing work to another system. Much of the discussion asks how capable one agent can become. Simulacra asks what happens when many of them meet. Whose requests get through? Which promises can be trusted? Who sets the rules?

Similar pressures could appear wherever agents share resources. They may compete for access, form reputations, delegate work, or rely on intermediaries. I want to use small simulations to study how those relationships develop. In a bounded world, we can change one rule, hold other conditions steady, and examine the consequences.

How to read this notebook

Think of this collection as a notebook of experiments I want to try. The inboxes, towns, and services give each question a setting; the proposed comparisons explain how we might learn something from it. They are imagined studies, with results still to come. Longer essays develop the argument; shorter notes work through one question. The historical field note explains how I arrived here.

The trilogy has three parts. Games asks what the rules allow and reward. Ecology follows agents as they compete, cooperate, and learn from one another. Genus asks what makes agents different and what a replacement can inherit. You can read them in order or begin with a situation you recognise. Each note introduces the terms it needs.

What the ecological language means

Ecology is the organising analogy of this project. Its terms point to observable relationships: a niche is a role sustained by the surrounding population; selection is a difference in which configurations get copied or retained; a lineage is a chain of agents connected by what they inherit. Each term needs a test that could show the pattern is absent.

Using this vocabulary makes no claim that software is alive. Its value lies in helping us compare relationships across different systems. Where the analogy stops explaining behaviour, I will use the more literal description: shared resources, interaction rules, copying, or memory.

Civilisation names a longer-term question about institutions. A market, a rule, or a shared archive might persist after every agent that created it has been replaced. Studying that persistence gives the larger idea something concrete to answer to.