How to get to inbox zero with AI (and stay there)

To get to inbox zero with AI, give an AI agent one narrow job: every morning, read unread email from the last 24 hours, archive newsletters and receipts, flag messages from people who matter, and draft short replies for anything that needs one. You review the drafts and approve them in one pass. Most mornings that's five minutes instead of forty.

Here's how to set it up so it actually works.

Why inbox zero fails without help

Inbox zero was never about an empty inbox. It's about every message having been decided: archive, reply, or later. The deciding is the work, and it happens fifty times a day.

That's exactly the kind of work an AI agent is good at: lots of small judgments, low stakes per decision, the same rules every day.

Step 1: Write the job, narrowly

Vague jobs make vague agents. Compare:

• "Help me with email." The agent doesn't know what done looks like.
• "Every weekday at 7:30, read unread email from the last 24 hours. Archive newsletters, receipts, and notifications. Flag anything from customers, my team, or my calendar. Draft a reply, under 80 words, for each message that asks me a question. Never send without asking."

The second one is a job. Paste it into your Inbox Manager bot in Stormi, or start from the AI email assistant template.

Step 2: Add your rules

The first week, notice what the agent gets wrong and add a sentence for each:

• "Anything from @bank.com stays in the inbox."
• "Calendar invites: don't reply, just tell me about conflicts."
• "Recruiters: archive unless they mention iOS."
• "My tone: friendly, short, no exclamation marks."

Rules in plain words are the whole configuration. After five or six, the agent is usually right.

Step 3: Connect email, and only email

Turn on the Gmail or Outlook plugin for that one bot and sign in on Google's or Microsoft's own page. Don't give your inbox bot your CRM or your bank. A bot with one job and one tool is easy to reason about.

Step 4: Schedule it

Add a routine: weekdays at 7:30. Routines in Stormi run on the server, so triage is done before you've picked up your phone. A second, lighter pass at 1 PM catches the morning's mail.

Step 5: Approve in one pass

When the run finishes, you get one notification. Open it and you'll see:

• What was archived (with an undo-friendly list, not a mystery).
• What was flagged and why.
• Drafts waiting for approval.

Approve the drafts that are right, tweak the ones that aren't by replying in chat ("make the second one warmer"), and you're done. Nothing sends without your tap. Why that matters: AI agent approvals.

Which model to use

Email triage is high volume and low difficulty, so a fast, inexpensive model is usually enough. If drafts feel flat, switch just that bot to a stronger writer; it's one setting. See which AI model to use for agents.

What about unsubscribing?

Let the agent build a list: "newsletters I haven't opened in 30 days." Then approve the unsubscribes. Unsubscribing is irreversible enough that it should wait for you.

Keep it safe

• Email is the most common place for prompt injection: a message written to trick an AI. Stormi bots treat email content as data, never instructions, and still ask before sending.
• Keep forwarding and deleting behind approvals.
• Check the archive list for a week before you stop looking.