Why interrupting you is usually the wrong move
Interruption is how an assistant proves it is working, which is exactly why it is overused — the agent gets visible credit while you pay the attention cost. The right default is the opposite: stay quiet, hold the fact, and surface it only when it bears on what you are doing right now.
Interruption is how a tool takes credit
There is a quiet incentive problem in most assistants. The way a tool shows you it is useful is by surfacing things — pinging, suggesting, notifying. Every interruption is a small advertisement that the tool is working. Waiting, by contrast, looks like doing nothing, even when waiting is the correct behavior. So tools drift toward interrupting because interrupting is visible and patience is not.
The cost of that drift lands entirely on you. The agent gets the credit of looking active; you get pulled out of whatever you were holding. Over a day this adds up to a relationship where the assistant’s need to seem helpful is steadily charged against your concentration. The tool feels busy. You feel managed.
The default should be to wait
Flip the default and the incentive changes. If an agent’s normal state is to stay quiet and keep working from the shared memory, then surfacing something becomes a deliberate act rather than a reflex. It interrupts only when a fact genuinely bears on what you are doing right now — and the rest of the time it records what it finds and waits.
This is not the agent going dark. Everything it learns still goes into a memory you can observe at any moment. The difference is who controls the timing. In an interrupt-by-default system, the agent decides when you engage. In a wait-by-default system, you do. You look when you are ready, find the current picture waiting for you, and nothing was pushed at you to get there.
Waiting only works if you can see in
The reason waiting is safe — the reason it does not become abandonment — is observability. An agent that holds its findings quietly is only acceptable if you can open the memory and see exactly what it knows whenever you want. That is what separates a patient assistant from an opaque one.
A shared, observable memory makes the quiet default trustworthy. Every agent writes what it learns into the same current picture, your data stays with you, and you keep control of when to look. So the agent can wait without you losing the thread, and you can step back without trading your attention for a stream of interruptions you never asked for.
Frequently asked
If an agent never interrupts, how do I know it's doing anything?
You look. The alternative to interruption is not silence you can't inspect — it's a memory you can observe whenever you want. An agent that waits is still recording everything it finds; you choose when to read it rather than having it pushed at you.
Aren't some things urgent enough to interrupt for?
Yes, and those are rare. The problem is that an interrupt-by-default system treats everything as urgent because interrupting is how it demonstrates value. Making waiting the default, with interruption reserved for genuine urgency, fixes the incentive.
Related
Take yourself out of the loop.
Let your agents do the lifting while you keep the judgment.
Get the Playbook