Agents do not just answer — they act. That creates a new UI category: ambient status for long-running, tool-using work that may touch files, browsers, or accounts.
Three layers of agent UI
- Presence — is the agent with me? (companion / mind state)
- Activity — what step is running? (tool timeline)
- Authority — what is it allowed to do? (permissions, confirms)
Teams often ship #2 and #3, then wonder why the product still feels lifeless. Layer #1 is what humans use to stay oriented while skimming.
Calm technology constraints
- Motion stays in the periphery unless approval is required.
- Sound is off by default.
- Captions are short; details live in the timeline.
- Destructive actions never hide behind cute animation.
Mapping agent phases to a mind
- Planning / retrieving → thinking
- Waiting on user approval → listening (or a dedicated “blocked” chip in the timeline)
- Narrating results → talking
- Parked between tasks → idle
Why a tiny library helps
Agent dashboards get heavy fast. A zero-dependency presence layer like Glim keeps the ambient channel cheap while your timeline and permission graphs stay purpose-built. Do not force the companion to visualize every tool call — that is the timeline’s job.
Try Glim free. Offline starter + MIT creatures. No fake checkout.