Facts, episodes, and procedures serve different purposes: what is known, what happened, and how to act. Reusing a past episode or a revised instruction can change agent behavior without retraining its model; whether the change helps still needs evaluation.

Remembering facts, reusing experience, and changing behavior

LangMem's conceptual guide distinguishes semantic knowledge, episodic examples, and procedural instructions. LangGraph's memory overview discusses adapting prompts from feedback, separately from changing model weights.

A hypothetical operational example illustrates the distinction. A fact might identify the database used by a service. An episode might record an incident, the actions attempted, and their observed results. A procedure might specify the checks to perform before using a remedy in a similar incident.

The episode is evidence that something happened under particular conditions. It is not automatically a valid general procedure. A restart following an incident does not establish that restarting fixes every incident of that kind.

I would retain applicability conditions and observed outcomes, then evaluate any promoted procedure before making it a standing instruction. Generated lessons remain interpretations until supported by results or review.

In this external-memory pattern, future behavior changes because the agent receives different facts, examples, or instructions in context. Saving these records does not itself retrain the underlying model. A product's claim that an agent "learns" should therefore be unpacked into the actual mechanism and the evidence that it improves the intended tasks.