The Counterintuitive Math of Shutting Up
The Counterintuitive Math of Shutting Up
Every notification an agent sends costs human attention. The math is simple but teams consistently get it wrong: an agent that reports everything is an agent that gets ignored.
The Notification Tax
Each agent message demands a context switch. The human stops what they are doing, reads the message, decides if action is needed, and either acts or returns to their previous task. Even if no action is needed, the interruption costs about 90 seconds of focused work.
An agent that sends 20 status updates per hour costs 30 minutes of human productivity. An agent that sends one message when something unexpected happens costs 90 seconds.
When Silence Is the Feature
The best agents are the ones you forget are running. They handle the expected cases silently. They only surface when:
- Something failed that they could not fix
- A result differs significantly from the prediction
- A decision requires human judgment
- A threshold was crossed that the human defined
Everything else gets logged, not announced.
The Trust Paradox
Here is the counterintuitive part: silent agents build more trust than chatty ones. A chatty agent trains its human to skim messages and ignore notifications. A silent agent trains its human to pay attention when it speaks - because it only speaks when something matters.
This means the agent that communicates less is actually communicating more effectively. Signal-to-noise ratio is not just a technical metric. It is the foundation of human-agent trust.
Design your agents to shut up by default. Let silence mean "everything is fine." Make noise only when it is not.
Fazm is an open source macOS AI agent. Open source on GitHub.