Agent Status & Monitoring
What is Agent Status & Monitoring?
When an agent is working on a long-running or multi-step task, users need to know what's happening without being forced to watch constantly. Traditional loading indicators like spinners and progress bars don't work for agentic tasks that may take minutes or hours, involve multiple parallel activities, or require occasional user attention. The design challenge is keeping users informed without demanding their attention. This pattern provides four status layers: ambient status (persistent unobtrusive badge), progress status (glanceable panel available on demand), attention status (interrupting notification when input is needed), and summary status (completion report). The system supports multiple concurrent tasks, provides estimated completion times, and auto-dismisses completed items while keeping them accessible in the audit trail.
Problem
Traditional loading indicators don't work for agentic tasks that take minutes or hours, involve parallel activities, or need occasional user input. Users need to stay informed without being forced to constantly monitor agent activity.
Solution
Design a layered status system: ambient badges for background awareness, expandable progress panels for detail, attention notifications only when input is needed, and completion summaries when tasks finish. Support multiple concurrent tasks with estimated times.
Real-World Agent Status & Monitoring Examples
Implementation
When to use Agent Status & Monitoring, and when it backfires
Use it when
- The task runs long enough that watching it is wasteful: minutes to hours, often in parallel, where the user should be doing something else while it works.
- The agent will occasionally need a genuine decision (a confirmation, a missing input, a fork in the plan) and the cost of missing that moment is real.
- There is a clear hierarchy of what matters: most updates are 'glance if curious,' a few are 'you must look now,' and the system can tell them apart.
Don't, or minimize, when
- The task finishes in seconds. A spinner says everything a four-layer status system would, with none of the overhead.
- Every update is equally urgent in your design, which means none of them are. If you cannot rank what deserves an interrupt, you are not ready to build the interrupting tier.
- You are tempted to ping the user on progress milestones (25%, 50%, 75%) because it 'feels responsive.' That is not keeping them informed, that is taxing their attention for no decision.
The trap
Over-escalation, the cry-wolf dashboard: every event gets routed to the loud tier because shipping a notification feels like good UX. Within a day the user has learned that your interrupts mean nothing and swipes them all away on reflex, including the one that actually mattered ('about to email the full client list, confirm?'). You spent the whole interrupt budget on 'task 50% done,' so when a real decision arrives there is no attention left to spend. It is worse than no status system at all: a plain spinner never trained the user to ignore you.
Take it into your own product
- 1
Match the tier to the stakes, not to the event.
Most agent activity is 'glance if you care,' a little is 'you must look now.' Ambient badges for the first, an interrupting notification for the second, and never the reverse. The whole design is a sorting rule for how much of the user's attention each event has earned.
- 2
Guard the interrupt budget like it's finite, because it is.
Every notification you fire spends a little of the user's willingness to look at the next one. Blow it on progress milestones and the one alert that matters arrives to a user who has already learned to swipe you away. An interrupt is for a decision, not for a status.
- 3
Status is a window, not a door.
Let users check in on the work without disrupting it: a panel they pull open on demand, never a thing that pauses the agent or demands acknowledgment. If checking progress costs the user a click and the agent a stall, you have rebuilt the spinner with extra steps.
- 4
Don't make them babysit it.
If the user has to keep the status panel open and watch it to feel safe, your ambient layer has failed. The promise of this pattern is that they can walk away and trust they'll be pulled back exactly when, and only when, they're needed.
- 5
Auto-dismiss the noise, keep the record.
Completed tasks should clear themselves so the surface doesn't silt up with finished work. But 'dismissed from view' is not 'gone': everything the agent did belongs in the audit trail, where the user can reconstruct what happened after the badge is long gone.
Add Agent Status & Monitoring to your product
Copy the prompt below into Claude Code or Cursor in your repo. It encodes the four moves on the left and asks Claude to find your AI decision surfaces and update them. Claude reports what it changed and asks before adding dependencies.
Check if your product already has this pattern
Upload a screenshot. We'll tell you which of the 36 patterns your AI interface uses and where the gaps are.
Audit My Design
