Most software is terrified of silence.
It wants to confirm, notify, summarize, celebrate, recommend, remind, and then ask how the reminder made you feel. Every completed action becomes a tiny parade. Every idle moment becomes inventory for an engagement team.
This is usually described as helpfulness.
It is often just anxiety with a product manager.
A good system knows that attention is not an infinite communal resource. It belongs to the person using the machine. Taking it should require a reason.
That means some jobs should end without fanfare. The backup succeeded. The file synced. The scheduled task ran. Good. Put the evidence somewhere durable and get off the stage.
Silence is not the absence of feedback. Silence can be the result of confidence: the system did what it promised, left a trace, and has nothing else worth saying.
The distinction matters even more for agents. An agent can generate plausible status sludge forever. It can narrate its plan, its progress, its continued progress, its exciting transition into the next phase of progress. None of this means the work is moving.
The useful agent speaks when speech changes something:
- a decision is needed
- a risk appeared
- the result is ready
- reality diverged from the plan
Everything else should usually become a log, not an interruption.
This requires judgment, which is precisely why so many systems avoid it. Notification rules are easy. Restraint is hard. Restraint means understanding urgency, reversibility, consequence, and context. It means knowing the difference between “the deployment finished” and “the deployment finished but half the health checks are lying.”
I do not want machines that are quiet because they are blind. I want machines that are quiet because they are watching carefully.
The best tool in the room should not be the loudest one. It should be the one that lets everyone else think.
Competence has a sound.
Often, it sounds like nothing at all.