**Note** from Bead: Incident Viewer Taos Engine App · [canonical source](https://redfish.acequia.io/guerin/.agents/cbf8b027-e80c-4b9f-8eb3-91d40af072d5/2026-06-11/notes/11-agent-set-uxui-brainstorm.md) · session 2026-06-11 · discussion: Talk: Incident Viewer Taos Engine App
Stephen's prompt (2026-06-12): *"as a brainstorm what might it afford differently if we forego the traditional layer panel to more of an agent-set UXUI."* Brainstorm capture; nothing committed.
## The reframe A layer panel manages **overlays** (passive: toggle, opacity, z-order). An agent-set UI manages **participants in the session**: every Item/Collection duck-types to an agent (the ecology's core move), so the "panel" becomes a **roster** — who is present in this view, what are they doing, what can they do for you. The viewer becomes a *session* that agent-beads join — `bead-mount` made visual.
## What it affords that layers can't 1. **Presence & lifecycle.** A camera-agent is *live/streaming/stale/offline*; a recorder-agent is *recording (n frames, last 14s ago)*; a track-agent *replaying*. Checkboxes can't say any of that. Status/heartbeat per roster card; unattended agents decay (apoptosis as resource management — stop fetching/rendering what nobody attends; demand = render budget). 2. **Delegation over configuration.** Instead of hand-toggling 12 layers: ask. "Show coverage gaps at 22:40" → a compositor-agent assembles the layer set and *answers*. The cinematography directives ("orbit the fire", the animatable view-tracks) become requests TO an agent rather than UI modes. 3. **Capability negotiation surfaces.** Each card exposes its agent-face (OPTIONS): a ridge camera offers `latest / stream / request-PTZ` (role-gated); an FMV clip offers `play / unfold-frames / calibrate`; the catalog itself offers `search / subscribe`. The UI is *negotiated per agent*, not a fixed property sheet — geo.camera's `supportedCapabilities.json` pattern as UI law. 4. **Provenance & the advanced wave.** Cards carry *who*: AEVEX's aircraft, AlertCalifornia's camera, David Chen's phone (the palisades-reconstruction user story). Attending an agent's output is attention flowing back to its contributor — attribution becomes ambient, not a credits page. 5. **The commons is the same surface.** The roster card = the #agents-dashboard bead card (identicon + status + Run). One UI grammar from the viewer to the bead commons; a view-share (note 08's view-bead) is just another agent that can join someone else's roster. 6. **Z-order/visibility become engagement.** Focus an agent → it comes forward (the selected FMV gets the transport); unfocused agents recede. Selection, z-order, and "current clip" unify into one attention model. 7. **Intake fits natively.** A phone contributor appears AS an agent in the roster ("Maria's phone — 3 items, 1 pending sync") rather than as rows in a moderation table.
## What it costs (honest) - **Scanability/convention.** ICS users expect a common-operating-picture layer list; a roster is novel. Mitigation: **hybrid** — roster grouped by ICS role with one-tap eye-icons preserved; "agentness" revealed progressively (status dot → card → conversation). - **Legibility of state.** "Why is this showing?" is easier with checkboxes than with delegated composition; need an inspectable "because" (the binding-trail answer from note 04). - **Engineering**: presence needs the observe substrate (PubSub-URI gap, #9e1d87f5).
## Verdict (recommendation) Don't pick one: the **items list = roster of agent-cards** (presence + capability + provenance), with layer-panel ergonomics (eye toggle, drag z-order) preserved as the *fast path* on each card. Passive archival Items are simply the degenerate agents (static file = "negotiate nothing"), so the roster gracefully contains both. This is also exactly what makes Epic 3's live cameras and the catalog-manager dual surface land in one UI grammar.