Summary: Knowledge bases decay: completed projects keep resurfacing, abandoned plans show up as “50 days late”, and clipped files never get processed. The AI only knows what you tell it, so a living system needs routines that check freshness and loops that check whether the routines themselves still run, produce output, and are still wanted.
Key ideas
- Dashboards kept nagging about a project she had dropped and never told the system about (osim-tochna—second-brain-and-llm-wiki, dana-maman).
- Clipped files sat in the inbox unprocessed because no rule said to ingest from there; then a routine ran but she’d put files elsewhere. Check once, then 800 more files flow fine (osim-tochna—second-brain-and-llm-wiki).
- Routine health ladder: installed → runs → produces output → produces what I want, to my taste → do I still need it? (an idea-hunting job ran for days unread) (osim-tochna—second-brain-and-llm-wiki).
- Without a processing routine, raw captures “rot in the warehouse” (osim-tochna—second-brain-and-llm-wiki).
Disagreements & open questions
Takeaways
- Add a routine that checks the index is current and the raw inbox is fully processed (osim-tochna—second-brain-and-llm-wiki)
- Periodically review whether each scheduled routine is still needed (osim-tochna—second-brain-and-llm-wiki)