Skip to main content

Procedural Memory

Procedural memory is memory of how to do something — implicit knowledge tied to learned routines and skills. It originated in cognitive science as the third pillar alongside episodic and semantic memory. In LLM agents, procedural memory shows up as learned tool-use patterns ("when the user says 'do the usual,' run this 4-step sequence"), prompt templates the agent reaches for in specific situations, and few-shot example stores the agent retrieves before generating. Most agents fake procedural memory by stuffing static examples into the system prompt; a real procedural-memory layer learns and updates patterns from interaction over time.

Example

A developer-productivity agent learns over weeks that when the user asks for a "PR summary," they want bullet points organized by file, with a short risk callout at the bottom. The pattern is stored in a procedural-memory layer and retrieved (not invented from scratch) on every PR-summary request — making the agent faster and more consistent than rebuilding the format from system-prompt instructions each turn.

Put this into practice

Build polished, copy-ready prompts in under 60 seconds with SurePrompts.

Try SurePrompts