# brgsk (brgsk.xyz)

Personal blog at [brgsk.xyz](https://brgsk.xyz). The through-line is taking a piece of AI-systems vocabulary that gets used loosely — "memory" being the clearest example — and separating the word from what's actually built underneath it. The method is to name each component precisely, ask what engineering the borrowed term would require, and then check which production systems actually implement it versus which ones just apply the label. The result is writing that's skeptical without being dismissive: the point isn't that the tools are bad, it's that the imprecise naming hides the real unsolved problems.

Style: lowercase prose, references to primary cognitive-science sources (Tulving, Baddeley & Hitch) alongside named practitioners (Letta, LangChain), and concrete library comparisons rather than abstract taxonomy.

## Ingested articles

- [[agent-memory-anatomy]] — agent "memory" libraries ship autobiographical semantic memory while borrowing the full episodic/semantic/procedural/prospective vocabulary; decomposes the system into extractor, store, and retriever