Loop Engineering

A Better Way to Think, Create, and Work with AI - Second Edition

← All chapters

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

Loop Engineering by Fabio Correa, the Applied AI foundation volume on working with AI through framing, feedback, verification, memory, and human judgment.

Last updated 2026-10-25. This page hosts the memory-science grounding behind Chapter 7 (Tulving's episodic/semantic separation and Squire's declarative/procedural axis) and the Alex ACT memory-tier reference implementation of the consolidation pattern.

The question this chapter takes up

How does a loop get better between sessions instead of merely longer?

Put it to work

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.

Key sources

Chapter 7 cites in the print book: D1, D2, D9, D22, D23, D28, D29, D42, D43, D86, D87. 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

Dated notes

Dreaming. In May 2026, Anthropic announced Dreaming for Claude Managed Agents, one vendor implementation of the consolidation pattern this chapter describes: a dreaming job reads a memory store and past session transcripts and produces a new, reorganized output store, leaving the input store unchanged. As of October 2026 the feature remains a research preview. Sources: the launch post and the Dreams API documentation, both checked October 2026.

/compact availability. Chapter 7's print text sends readers to this site for /compact cross-surface availability, because that detail has a short half-life. That information is maintained in the Alex ACT ONE plugin, which tracks the surfaces it runs on, so this page points there instead of duplicating it: Alex ACT ONE releases.

Related resources

The full memory-formation discipline (PII filter, tier selection, when to promote) lives in the Alex ACT repositories linked above. The multi-instance extension is Chapter 15's subject: lessons now propagate across instances as reviewed releases in the Alex ACT ONE line, rather than through a shared memory bus.