My AI email triage, in eight steps.

Teach it your voice.

Everything starts here. I used Igor Pogany's voice prompt from the AI Advantage Bootcamp. Igor says it took him two years to build and refine, and it gives you a really detailed voice profile, about 6 to 10 pages.

It learns by reading your last 50+ substantive emails, and you can ask for 100, 200, 300 or as many as you want. You read the profile, tweak it, then save it as a voice skill. After that, every draft your triage writes sounds like you.

In the kitA voice-profile skill that teaches the method in our own words, a template and a worked example.

A free script keeps watch.

A plain Python script checks for the emails and messages that matter, as often as every minute. It costs zero AI tokens.

AI only wakes up when something real lands. Checks that find nothing cost nothing.

In the kitThe watcher script, a ready inbox example, and Windows and Mac schedules.

It drafts the reply.

For each email that needs an answer, it reads the whole thread and drafts a reply in my voice.

Anything it doesn't know goes in brackets for me to fill in. No made-up dates, prices or promises.

In the kitA reply-drafter skill. Drafts land in your drafts folder. Nothing sends.

Edit by voice. A critic checks it.

I talk my changes instead of typing them. Then a second agent, one that didn't write the draft, checks it before anything goes out.

"I usually have an agent critique it anyways. It catches mistakes."Joshua

So I don't have to keep re-reading every single draft.

In the kitA draft-critic skill with the full checklist: facts, promises, tone, private details, leftovers.

Nobody gets forgotten.

A missed-reply checker looks both ways: people still waiting on me, and people who haven't answered me.

After two business days of quiet, it drafts a friendly nudge.

In the kitAn inbox-triage skill that checks your sent mail too, not just unread inbox.

My AI chief of staff answers the simple ones.

Simple replies, or a fact I'd have asked AI to pull anyway, go out from my AI chief of staff with me cc'd.

That way nobody waits on me for a link or an address.

In the kitOff until you turn it on. You pick the reply types, a critic must pass it, you're cc'd and every send is logged.

A reading desk that holds still.

Most AI triage hands you a summary. I wanted to read the whole original email and the whole draft. But every time I dictated a change, the chat scrolled down and I had to scroll back up to find my place. Over 20 emails, that's a lot of scrolling. Two fixes:

• Split screen on desktop. All the emails and drafts open in a separate reading desk session. I drag it beside my main AI session. I read on one side and dictate on the other, and the reading side never moves.
• Phone as the remote. I read the emails on my desktop and dictate from my phone through Claude Code's remote session. My phone scrolls, my desktop doesn't, so I go one by one and tap send until I'm done.

In the kitA reading-desk skill, the split-screen setup and the phone-as-remote steps.

A charge watcher.

It flags charges and renewal receipts that land in my email, like a subscription that renewed when I didn't want it to. It asks me whether each one is legitimate.

If it isn't, it drafts the refund or cancellation request for me to approve and send. I've used it to get an unwanted renewal refunded.

In the kitA charge-watcher skill and a free watcher example. Drafts only. It never contacts a merchant, never moves money and never puts card numbers in a draft.