3 Robots walk into a room ...

"I call my LLM-based service objects robots instead of agents. Why? because robots do what you tell them. Agent have agency. That means they can make choices and do whatever they want to do. Like travel agents, real-estate agents, FBI agents. Robots are machines that follow instructions. That's what I want. My objects should follow my instructions... but what happens when you add a little bit of agency to 3 robots and put them into a room with a set of tools that allow them"

"to communicate with eah other through shared memory, broadcast message channels and direct message channels. Then you tell all 3 robots to do the same thing? They become a self-organizing group. SOGs have agency."

LLM-based service objects are labeled robots to emphasize strict instruction-following rather than autonomous decision-making. Agents are characterized by agency, meaning the capacity to make choices and act independently. Robots are defined as machines that follow instructions, which is the desired behavior for these objects. Introducing a small amount of agency plus communication tools—shared memory, broadcast channels, and direct messages—among three robots enables interaction. When all three are instructed to perform the same task, they can self-organize. These self-organizing groups (SOGs) demonstrate emergent agency through coordinated collective behavior.

Read at Rubyflow