Importance vs. relevance vs. attention
Three words that get used interchangeably, and why a memory system has to keep them apart.
When people talk about what a system should "remember," they tend to collapse three different ideas into one. A memory architecture cannot afford to.
Three different questions
- Relevance asks: does this relate to what I'm doing right now? It is about the present query or intent.
- Importance asks: does this matter beyond right now? It is about durability — the conclusions worth carrying across sessions.
- Attention asks: what am I actually spending capacity on this moment? It is the scarce resource being allocated.
Something can be relevant but unimportant (a passing detail that fits the current task). Something can be important but not currently relevant (a standing goal that isn't active this second). And attention is finite regardless of either.
Why it matters for design
If you treat these as one axis, you get a system that either hoards everything "relevant" or promotes everything "important" — and drowns. Keeping them distinct is what lets an agent surface the right memory for the current intent while still holding onto the handful of conclusions that should shape its behavior over time.
This is a working position, not a solved problem — exactly the kind of open question the Vision page is about.