📬 AI Controls My Inbox: The Prompts That Fixed It (Claude Cowork)

Michael Wacht

🔥

5d (edited) • 🟠 AI Practitioner

I promised

@Kyle Covan

my next post would be the specific prompts strategies that I used to make my Inbox triage more efficient.

Not the philosophy.

The actual strategies.

📝 What actually changed for the better this week:

It (Cowork) stopped asking me things it already knew the answer to
When a question was genuinely open, it started handing back a ready answer instead of just a flag
The one narrow auto-accept rule finally got tested by something trying to slip past it

Here's how each one played out:

📝 It stopped asking things it already knew.

Early runs, it asked "should I create a follow-up note?" for a contact. The same report already showed a meeting booked with that person, two lines up. It had the answer. It asked anyway.

Fix: before flagging anything, re-check it against my sent email and meetings first. Five questions became one.

📝 Flagging it isn't finishing it.

A Skool contact wanted a meeting time. His only clue was "morning or night." Turned out that meant Jakarta, eleven hours ahead.

Fix: don't just flag it as open. Pull my calendar, check the timezone, hand back ready-to-paste times. Three options came back, both timezones shown, nothing left for me to calculate.

📝 The narrow rule finally got a real test.

The auto-accept rule only fires on one exact domain. Everything else gets flagged, no guessing. On 7/12, a meeting came in from a different organizer doing the same kind of work, close enough to pass at a glance.

It didn't auto-accept. It got flagged. That's the test that actually matters. Not the obvious case, the one built to sneak through.

📝 The number that proves it.

Day one: five items needing me.

Day five: one.

Same volume of email and meetings.

The difference was an agent that stopped asking what it already knew.

9 comments

📬 AI Controls My Inbox: The Prompts That Fixed It (Claude Cowork)