Few recommendations by design
We are designed to generate a small number of recommendations because overload weakens trust and makes action less likely.
We should protect the user from noise, even if more suggestions are technically possible.
Archives
Recommendation system
Our recommendation layer is intentionally constrained. We should surface a few high-signal suggestions, not a long stream of generic prompts.
Different recommendation types
Some recommendations are immediate actions, some are preventive, some are reflective, and some help organize an upcoming block of work.
This gives us a richer vocabulary than "do this now".
Explainability as a product rule
Every recommendation should be explainable in calm language. We should never hide behind scores or silent internal logic when a user is trying to understand why something surfaced.
Legibility is part of what makes us trustworthy.