Contextual Assistance — Books
Exploring how AI and good product design can surface better next steps without feeling like marketing, upsell, or interruption.
Why this experiment exists
In complex enterprise products, users often already have the tools they need — they just don’t always know the best or safest way to use them.
This experiment explores a simple question:
What if the system could notice when a user is “done” with something, and help with what usually comes next — without popups, tours, or nagging?
The focus here is Books: a surface where admins frequently archive, export, or share materials after a meeting concludes.
The idea
Instead of treating actions like Archive, Export to PDF, or Send to Data Room as isolated commands, this experiment treats them as signals:
- Archiving usually means the meeting is finished
- Exporting usually means materials will be shared externally
- Sending to the Data Room usually implies an audit or record-keeping need
The system can respond in two modes:
- Coaching — suggesting better or more complete follow-up actions
- Agentic — performing those follow-ups automatically, with clear receipts and undo controls
What this prototype illustrates
- Ambient, contextual messaging (not global notifications)
- A clear difference between suggesting and acting
- Guardrails that preserve user trust (review, undo, cautious modes)
- A system that can be a little proud of itself when it saves time
This is intentionally not a production-ready UX — it’s a way to reason about timing, tone, and trust.
Explore the prototype
→ Open the Context Assist — Books experiment
/now/ca-books
As you click around, pay attention to:
- when the system chooses to speak
- when it stays quiet
- and whether the assistance feels helpful or intrusive