Post #1490137

@abucci @ngaylinn I'm using it in a behaviorist sense. We set up an environment in which an agent might do well or very poorly, like a coding environment. Then we place system X in that environment and observe what it does. It might behave very unagentic, like a rock, or a next-token predictor. Or it might behave "more agentic" like an RL-trained black box. That's not proof of agenticity, but there are things we can observe.