Loop Engineering

A Better Way to Think, Create, and Work with AI - Companion site

← All chapters

Chapter 7 - Layer 2b: Long-term Memory and Consolidation

Last updated 2026-08-09. This page hosts the cognitive-load grounding behind Chapter 7 and the Alex ACT memory-tier reference implementation that operationalizes the consolidation pattern.

Companion activity

Take one useful lesson from a recent AI-assisted task and decide whether it belongs in user memory, session memory, repo memory, or an episodic post-mortem.

What this chapter argues

Working memory holds the active conversation; durable stores can preserve selected context beyond it. But persistence alone is not learning. A useful memory cycle captures an experience, distills what may generalize, stores it where it can be retrieved, and deliberately reinserts it into later work. The /compact instruction only compresses context within a session. Meditation is the separate consolidation practice that can promote lessons into skills, instructions, and prompts. The lesson becomes operational only when a later task retrieves and uses it successfully.

Memory tiers (Alex ACT reference implementation)

Key sources

Chapter 7 cites in the print book: D1, D2, D9, D22, D23, D28, D29, D42, D43. Entries below are the sources this page draws on; some are related literature cited elsewhere in the book rather than in this chapter.

Primary-source links

Chapter contents

Related resources

The full memory-formation discipline (PII filter, tier selection, when to promote) lives in the Alex ACT repositories linked above. A fleet, memory, and culture page on this site covers the multi-instance extension - how lessons propagate across heirs, and why the current constellation distributes them as reviewed releases rather than through a shared memory bus.