Remember what creates future usefulness.
Commitments, preferences, recurring patterns, open loops, names the user confirms, and context they keep needing.
A tiny boundary-mapping tool for deciding what an ambient AI or wearable assistant should remember, summarize, ask about, forget, or ignore.
Designed by Chanelle Henry as a product-thinking artifact for Bee. Less “AI magic,” more trust, memory, consent, and human control. Basically: a tiny flashlight for the part of ambient AI that can get weird fast.
The product question underneath the product
Bee turns moments into meaning, often from an always-near wearable context. The design challenge is making sure meaning does not become overreach. This map explores what the system should keep, question, or let go before “memory” becomes another place users have to clean.
Commitments, preferences, recurring patterns, open loops, names the user confirms, and context they keep needing.
Health, family, legal, finances, workplace tension, and other people’s personal information need a consent checkpoint.
“That’s wrong,” “forget this,” “save this,” “temporary,” and “never infer this again” should be first-class controls.
A simple boundary model
The user should not have to wonder what the system heard, kept, inferred, or ignored. Every saved memory should leave a receipt: source, confidence, sensitivity, and how to correct it.
Try the prototype
Paste a conversation fragment or use a sample. The prototype sorts the moment into five buckets: remember, summarize, ask first, forget, or ignore.
Memory receipt
No raw transcript stored. Sensitive context is held for confirmation instead of being saved as memory.
Five product answers
These are the design questions that sit underneath personal AI memory. They are less shiny than feature lists, which is usually how you know they matter.
Remember chosen commitments and patterns. Summarize useful shape. Ask before sensitive context. Forget ambient debris. Ignore jokes, venting, and half-formed thoughts unless the user saves them.
Correction should be visible everywhere memory appears: wrong, forget, save, temporary, merge, and never infer this again. The system should show what changed.
Bee can remember my obligations without building shadow profiles of everyone nearby. Other-person context needs visible boundaries and ask-first behavior.
Insights should show source, date, context, confidence, and whether they came from user input, repeated behavior, conversation summary, or inference.
Use digest modes, urgency tiers, quiet states, “not now,” and “only show what changed.” Reduce open loops instead of creating new piles.
Not magical. Trustworthy. Calm. Correctable. Useful at the moment of need. A second brain with manners, not a tiny surveillance goblin.
Why this is my lane
I work where the system is messy, the human is overloaded, and the interface is only one symptom of a deeper workflow or trust problem.
Built RAG-informed research workflows to turn large messy inputs into usable insight while preserving human review.
Designed for healthcare, public-sector, accessibility, compliance, forms, and trust-heavy workflows.
Experience across mobile, wearable-adjacent thinking, enterprise platforms, Salesforce, prototypes, service blueprints, and design systems.
Patience, signal, timing, restraint, and knowing what not to add. Product strategy, but smoked.
For Bee
Ambient wearable AI is intimate infrastructure. The opportunity is not just better summaries. It is better boundaries, better correction, and better user control over what becomes memory.