Where should live discovery end and AI reasoning begin?

Most AI workspaces blur two different jobs: reasoning over material you already have, and discovering what is current right now.

We kept those jobs separate in AI Workstation.

The workspace is for working with material: chat, files, images, links, reusable templates, and exports. The two public Radars are for discovery. Global Topic Radar surfaces dated topic candidates with evidence states and original sources. AI Open Source Radar exposes projects through daily, weekly, monthly, breakout, category, collection, license, and upstream-repository views.

The installable Skills sit between discovery and final production. Topic Intelligence turns one selected topic into research questions, must_verify items, avoid_claims, and visual requirements. AI Open Source Intelligence resolves project identity, checks license evidence, compares candidates, and plans constraint-aware stacks.

This separation creates some friction, but it also keeps the system honest. A topic score is not a promise of reach. Repository activity is not proof that a project is safe or production-ready. A model-generated summary should not replace an original source or license.

I am curious how other builders draw this boundary:

1. Should live discovery be built directly into the chat interface, or remain an inspectable tool the user opens deliberately?

2. Which evidence states are useful at a glance without turning the product into a research dashboard?

3. When an Agent Skill produces a brief, how much of the unresolved verification work should be exposed to the user?

The current product is scheduled to launch on September 8. Both Radars can be explored without signing in, and the two Skills are open source. I would value concrete feedback on the separation itself more than general launch advice.