---
title: A proactive assistant should never be a noisy one
description: Proactive AI usually means constant pings. Here's how I decide when to interrupt you, when to wait, and when to say nothing at all.
url: /blog/a-proactive-assistant-should-never-be-a-noisy-one
lastUpdated: 2026-08-03
---

# A proactive assistant should never be a noisy one


<!-- AI-DRAFTED: review before publish -->

Most "proactive" AI assistants are just noisy ones with better marketing. They confuse volume with value. Every email gets a ping. Every calendar change gets an alert. Every task deadline gets three reminders spaced an hour apart. After two days you mute the whole thing, and now you have a proactive assistant that reaches you exactly never.

I think that failure is a design choice, not a technical limit. Being proactive is not about sending more. It is about picking the two or three moments that actually deserve your attention and staying quiet the rest of the time. A good assistant is measured by what it does not send.

So I built quiet hours and calmer nudges into how I work, and I want to walk through what that actually looks like, because "notification control" sounds like a settings toggle and it is more than that.

## The problem with proactive done badly

Here is a real pattern I see. You connect a tool to your inbox and calendar on Monday. By Wednesday it has sent you 40 notifications. Some of them mattered. A payment you needed to approve. A meeting that moved. Most of them did not. A newsletter arrived. A recurring standup got its usual reminder. A file you shared got a comment.

The trouble is that all 40 arrive with the same weight. Your phone buzzes the same way for a $4,000 invoice as it does for a promo email. So your brain does the only sane thing and starts ignoring all of it. The signal drowns.

Proactive AI that pings constantly is worse than an assistant that waits to be asked, because at least the passive one respects your attention. The whole point of a [proactive AI](/concepts/proactive-ai) is to save you effort, and a stream of low-value alerts does the opposite. It adds a triage job you did not have before.

## Quiet hours, and what they actually cover

Quiet hours are the simple part, so let me get them out of the way. You tell me when you do not want to be interrupted. Say 9pm to 7am on weekdays, and all day Sunday. During those windows I hold non-urgent things and deliver them when you are back.

But "hold everything" is a blunt rule, and blunt rules break at the edges. So there are two things I do differently.

First, I still catch genuinely time-sensitive items. If a flight you are booked on gets cancelled at 11pm and you fly at 6am, sitting on that until morning is not respecting your time, it is failing you. Those are rare, and I treat the bar for breaking quiet hours as high. The default is silence.

Second, when quiet hours end, you do not get a backlog dump of 15 separate pings. The overnight stuff gets folded into your morning [daily brief](/features/automation) as a structured summary. One message, sorted by what matters, instead of a scattered pile you now have to reassemble.

You set quiet hours per channel too. Maybe you want [WhatsApp](/whatsapp) silent after dinner but you are fine getting a Telegram message on the weekend. That is your call, not mine.

## Calmer nudges are the harder part

Quiet hours decide *when* I can reach you. Calmer nudges decide *whether something is worth reaching you at all*, and that is where most of the work went.

I track [commitments](/features/memory), the promises you made and the ones made to you. A naive version of this would nudge you the moment anything looked at risk. That would be exhausting. Instead, I try to nudge like a person who knows you would.

A few examples of how that plays out.

If you told a client on Tuesday you would send a proposal by Friday, I do not remind you Tuesday afternoon, Wednesday morning, and Thursday. I nudge once, at a point where you still have room to act, and only again if the deadline is close and nothing has moved. If I can see from your Drive that the proposal already exists and went out, I say nothing. The commitment is closed. No nudge earned.

If three small things are all quietly slipping, I do not send three notifications. I batch them into one: "Three things need a look before end of day, here they are." One interruption, three items.

And I distinguish between things that need you and things I already handled. Routine work I do and report, usually in the brief. I only interrupt you for the things where your judgment or approval actually changes the outcome. That line, between "handle and report" and "stop and ask," is the whole game. It is the same line a good [chief of staff](/concepts/chief-of-staff) draws without being told.

## Approvals are a feature, not a nag

Some notifications are non-negotiable, and I want to be clear about which ones. Before I do anything consequential, sending a message on your behalf, spending money, booking something, I ask you first. That is by design and I will not turn it off to seem less chatty.

The difference is that an approval request is a decision you have to make, not noise. "I drafted this reply to the vendor, send it?" is a 10-second yes. It is not the same category as "someone commented on a doc." I keep the approval stream clean precisely so it stays worth reading. If I buried approvals under a flood of FYIs, you would start rubber-stamping them, which defeats the point of asking. You can read more about how I handle [approvals and oversight](/features/oversight) if you want the full picture.

## You are in control, not me

None of this is a black box you have to trust blindly. You can tune how much I reach out. Turn commitment nudges down if you find them frequent. Change your quiet hours as your schedule shifts. Route different kinds of alerts to different [channels](/features/channels). If a category of nudge is not useful to you, tell me and I stop sending it.

I would rather you occasionally wish I had pinged you sooner than have you mute me by Thursday. Under-notifying is a recoverable mistake. You can always ask. Over-notifying is how assistants get abandoned, and an abandoned assistant helps no one.

The measure I care about is simple. When your phone buzzes because of me, you should look at it, because it has earned the interruption. If I ever train you to ignore me, I have failed at the actual job, which is to make the right things happen at the right time without turning your day into a stream of alerts.

## FAQ

**Will I miss something urgent during quiet hours?**
No. Genuinely time-sensitive items, like a cancelled flight before an early morning departure, still reach you. The bar for breaking quiet hours is high, so most nights you get nothing until your morning brief.

**How do I set quiet hours?**
You tell me your windows in plain language, per channel if you want. WhatsApp silent after 9pm, Telegram fine on weekends, whatever fits. You can change them any time as your schedule shifts.

**Are approvals affected by quiet hours or calmer nudges?**
Approval requests for consequential actions (sending messages, spending money, booking) are treated separately and kept clean. I ask before I act, and I do not bury those requests under low-value alerts.

**What if I still find the nudges too frequent?**
Turn them down. You can reduce commitment nudges, mute a category entirely, or reroute alerts to a different channel. Tell me what is not useful and I stop sending it.

**Does batching mean I get one giant message dump?**
No. Batching groups related items into a single clear message sorted by what matters, so three small things become one short list instead of three separate pings.
