Tell your agent what's worth interrupting you for

An always-on agent is only useful if it knows when to speak up. Give it three bins (tell me now, save it for the morning brief, or don't tell me at all) with real examples from your business, plus quiet hours. You stop getting pinged about nothing, and you stop missing the one thing that mattered.

How your agent does it

1. You fill in the three bins with real examples: emergencies, failed payments and bad reviews now; enquiries and paid invoices in the brief; newsletters and receipts never
2. You set quiet hours and the one thing allowed through them
3. Anything it isn't sure about goes in the brief as "Not sure if this matters", never guessed
4. For the first month you tell it which messages were noise and which brief items should have been a message, and it updates the rules

Example prompt

Here are my interrupt rules. TELL ME NOW: [e.g. a customer says emergency, leak or cancel; a payment over £[x] fails; a 1- or 2-star review; anything from (key clients)]. MORNING BRIEF: [e.g. new enquiries, paid invoices, supplier emails, diary changes]. NEVER TELL ME: [e.g. newsletters, receipts that match an order, automatic notifications]. Quiet hours are [21:00–07:00]: only 'tell me now' items that say emergency get through. If you're not sure, put it in the brief under 'Not sure if this matters'. In each brief, list anything over 3 hours old with no reply. To start, sort last week's emails and messages into my three bins and show me the ten you were least sure about.

Good to know

• Difficulty: easy
• Typical benefit: Fewer pointless pings, no missed emergencies
• Works best on: Claude, ChatGPT, Gemini, Grok Bot, OpenClaw, Hermes
• Agent job: Night-shift Assistant, with Super-Organiser and Straight Talker