Writing on memory, knowledge, and agent continuity.
OKF and OpenWiki standardize linked Markdown for agent memory. Useful as a document format; weak as a knowledge representation when you need typed relations, temporal facts, graph queries, or write-time reconciliation.
Most AI systems invoke brain analogies metaphorically. Scaffold's is structural: episodic and semantic memory, hippocampal reconciliation at write time, cognitive maps in the graph, and sleep-like maintenance in the background.
A language model trains once and stops learning. Memory is not the feature that's missing. State is: a living, revisable record of everything the model has met since, and the discipline of keeping it navigable, current, and fluent.
Automation replaces; augmentation remembers. On building agent continuity as a coherence problem: vectors and graphs, write-time integration, local ownership, and a shared mind that travels with you across apps.