The Notification Fatigue Trap
TRIGGER
Agents working on tasks naturally generate many status updates and intermediate states, but surfacing every action as a comment or notification creates noise that fatigues users and trains them to ignore agent communication entirely.
APPROACH
Cursor's team explicitly discussed comment frequency as a design decision during Linear integration. Rather than commenting whenever the agent does something, they limited updates to meaningful progress milestones. For longer-running tasks (hours), they implemented periodic check-ins rather than continuous streaming, with updates framed like 'a new hire explaining what they did so the reviewer can quickly decide if it's ready.'
PATTERN
“Users trained to ignore agent notifications stop reading them entirely—including the ones that matter. Throttle updates to meaningful progress milestones, not every tool call, because users judge agent communication against how a teammate would report, not how a CI system emits events.”
✓ WORKS WHEN
- Agents perform multi-step tasks where intermediate states exist but aren't all user-relevant
- Updates flow into notification systems that compete for user attention (email, Slack, in-app badges)
- Tasks run long enough (>30 minutes) that users need progress indicators but not play-by-play
- Users will review agent output rather than watching it execute in real-time
✗ FAILS WHEN
- Users are actively watching agent execution and want real-time feedback (interactive coding sessions)
- Tasks complete in under a minute where any update is effectively the final update
- Compliance or audit requirements demand logging every action
- Agent actions have immediate consequences that users must approve before continuing