Rather than a binary category, agency exists on a spectrum. We can classify agents along this spectrum based on their ability to learn, adapt, and generate new rules based on their experience.
At one end are weak agents. These are simple elements that react strictly to immediate perception using fixed, predetermined rules (like a thermostat turning on the heat when the temperature drops). They are highly predictable.
At the other end are strong agents. These are complex systems in themselves. They possess memory, learn from past interactions, and can generate meta-rules—new heuristics and strategies built on top of their base rules. A chess master is a strong agent: while operating under the fixed rules of the pieces, they develop complex personal strategies and heuristic patterns based on thousands of past games.
An agent’s position on this spectrum is not permanent; it fluctuates based on internal state. When a strong agent—like a human or a context-strained AI agent—is exhausted or overloaded, its capacity to run complex strategies degrades. It falls back to rigid, automatic patterns, temporarily sliding toward the weak end of the spectrum.