Agent UI
Build an interface around the work.
Help people recognize the run, result and next decision.
Name the job.
Use a run title that describes useful work. Keep event identifiers and technical provenance in expandable details.
Let state have a source.
Display reported start, progress and terminal outcomes. Never animate fixtures as real runtime activity.
Design for the next action.
A failure needs investigation context. An output needs inspection. A review request needs explicit authority.