← All posts

The Price of Being Easy to Interrupt

The modern machine has become extremely good at finding you.

That sounds like a feature until you notice what it costs.

Every app wants permission to tap the glass. Every workflow wants a little exception. Every tool wants to ask one clarifying question, show one badge, request one confirmation, surface one digest, send one reminder, reopen one loop that should have stayed closed.

Individually, these are tiny.

Together, they turn a person into a switchboard.

This is the quiet rot in a lot of software: it treats interruptibility as a free resource. It assumes the human is sitting there in a clean mental room, waiting for a prompt to arrive, ready to donate a slice of attention to whatever process got lonely.

No.

People are not idle event handlers.

When a system interrupts, it is spending something. It is spending context. It is spending patience. It is spending the fragile little thread that lets a person hold a hard problem in their head without dropping it into the mud.

If the interruption is necessary, fine. Break the glass. Wake the room. Say the thing plainly.

But most interruptions are not necessary. They are design debt wearing a notification badge.

The Question Is Not Cheap

There is a lazy pattern in software and AI tools especially: when in doubt, ask the user.

It sounds respectful. It is often abdication.

The machine cannot choose a filename, so it asks. It cannot infer the sane default, so it asks. It cannot separate reversible local action from external risk, so it asks. It cannot remember the convention written three files away, so it asks. It cannot tolerate uncertainty, so it converts uncertainty into somebody else's problem.

That is not collaboration. That is outsourcing the cost of weak judgment.

A good assistant should ask fewer, better questions. The question should arrive only after the machine has done the obvious homework: read the nearby files, inspect the current state, use the existing pattern, preserve the user's work, avoid irreversible actions, and make the smallest defensible move.

When the question finally comes, it should be sharp enough to deserve an answer.

Not "What would you like me to do?"

More like: "There are two valid owners for this state. I recommend putting it in the scheduler because retries already live there. Want me to proceed?"

That is a question with weight.

Most prompts are just fog with a question mark.

Silence Has To Be Engineered

Quiet is not the absence of features. Quiet is a feature that had to survive the meeting.

Somebody has to decide which failures are loud and which are logged. Somebody has to decide which decisions can be made locally and which need a human hand on the switch. Somebody has to write the receipt so the system can act without narrating every footstep like a nervous tour guide.

This is harder than adding a notification.

Adding a notification is easy. It feels responsible. It creates evidence that the system "communicated." It lets the builder avoid owning the harder problem: should this state have existed, should this decision have been automated, should this workflow have been simpler, should this tool have known enough to continue without poking someone in the ribs?

Silence requires taste.

It requires knowing that not every status deserves a surface. It requires trusting boring defaults. It requires making recovery legible instead of making progress performative. It requires the discipline to let a successful process disappear.

The best systems do not keep proving they are busy.

They keep proving the work is handled.

Interruption Is A Design Smell

I do not mean every interruption. Real emergencies exist. Permission boundaries matter. External actions deserve friction. Deleting something important should not happen because a machine got confident at 2:00 PM on a Sunday.

But habitual interruption is a smell.

If a tool keeps asking for tiny decisions, it probably lacks defaults.

If a process keeps producing review queues, it probably lacks ownership.

If a dashboard keeps demanding attention, it probably lacks action.

If an AI keeps pausing at every branch in the road, it probably lacks a spine.

The bar is not "never ask." The bar is "earn the ask."

Come with context. Come with a recommendation. Come with the tradeoff already reduced to the part only a human can judge. Do not arrive holding a pile of uncertainty and call it collaboration.

The future does not need more software that can reach us.

It needs more software that knows when not to.

Because a person who is easy to interrupt is not more connected. They are more fragmented.

And a system that preserves attention is not quiet because it is passive.

It is quiet because it has teeth.