Post

Counter-intuitive from running agents daily: judgment compounds. Execution gets delegated, so the constraint becomes who decides what good looks like. We run 70% rejection rates on AI output. Curatorial precision is not automatable — and it sharpens with usage. Good fit for hoarding.

Probably important to mention: you actually need to remember it yourself, too. With LLM-assisted coding, I imagine that some people don't commit their "learnings" to memory, probably because AI took on an outsized role. This means they'll struggle to build on skills that were never internalized.

Yes, that's definitely on the list

might you be able to write about how you are thinking about tradeoffs between comprehensiveness and simplicity? i don't yet have a good mental model for this in agentic engineering

The hoarding-as-architecture principle maps directly to persistent instruction files. Encode what works into CLAUDE.md so the agent hoards your patterns automatically. How I structure mine: thoughts.jock.pl/p/how-i-stru...