In the late 1970s, developmental psychologist John H. Flavell formalized metacognition: knowledge and cognition about cognitive phenomena — the active monitoring and regulation of thinking itself.
His frame still maps cleanly onto serious agent systems:
- Metacognitive knowledge — persons, tasks, strategies (what / how / when and why)
- Metacognitive experiences — confidence, doubt, feeling-of-knowing
- Goals and tasks — what the activity is for
- Actions and strategies — what is actually done
Around those sits the regulation loop: plan → monitor → evaluate → control.
From psychology to this stack
- Persistent memory as metamemory — decide what deserves to persist, consolidate, prune, retrieve with awareness of gaps
- Gating as control — write/read priority from utility, novelty, uncertainty, goal alignment
- Recursion as self-reference — store why something was kept; let reflection become new experience (events → episodes → themes → meta-strategies)
Why this attracts builders
Scaling improves object-level fluency. Metacognition improves judgment about answering — and about remembering. Passive you want AI-first websites that compound instead of reset, you need more than RAG demos. You need a meta-layer over memory that you can own, inspect, and govern.
That is the research and engineering center of Persistence of Memory — made runnable with LanceDB, MySQL, FastAPI, PHP, and multi-model cognition.