My AI Agent Triages 200+ Emails a Day — I See 5

The Problem

I was spending 45 minutes a day scanning my inbox. Not replying — just scanning. Reading subject lines, opening newsletters I'd never finish, marking spam, starring things I'd forget about later.

Two hundred emails a day across investor updates, product alerts, team threads, cold outreach, newsletters, and actual important communication. The signal-to-noise ratio was maybe 3%. But finding that 3% required reading 100%.

The real cost wasn't 45 minutes. It was the context switches. Every time I opened Gmail, I lost whatever I was working on. Four times a day, that's four derailments. An entire creative hour, gone.

I needed a system where email comes to me only when it matters. Everything else should handle itself.

The Solution

An Mr.Chief agent running on a cron schedule, using the gog CLI to access Gmail. Four times a day, it reads every unread email, classifies it against a rubric I defined, applies labels, archives junk, and only sends me a Telegram notification for truly urgent items.

The agent isn't just filtering — it's understanding. It knows that an email from my co-founder about runway is urgent. It knows that a Substack digest is FYI-at-best. It knows that "Quick question" from a stranger is almost certainly cold outreach.

The Process

The backbone is a cron job that fires four times daily:

# mrchief cron config
- name: email-triage
  schedule: "0 8,12,16,20 * * *"   # 8am, 12pm, 4pm, 8pm UTC
  channel: telegram
  task: |
    Read all unread emails using gog. Classify each into one of five categories:
    - URGENT: requires my response within 4 hours (investors, co-founders, legal, financial deadlines)
    - ACTION: requires response within 48 hours (team requests, partner emails, scheduled follow-ups)
    - FYI: informational, no response needed (newsletters I subscribed to, product updates, receipts)
    - SPAM: unsolicited outreach, marketing I didn't opt into
    - ARCHIVE: automated notifications, CI/CD alerts, calendar confirmations

    For each email:
    1. Apply the corresponding Gmail label
    2. If SPAM or ARCHIVE: mark as read and archive
    3. If FYI: mark as read, leave in inbox with label
    4. If ACTION: leave unread, apply label, add to daily digest
    5. If URGENT: leave unread, send me a Telegram message immediately with sender, subject, and one-line summary

    After processing, send a single summary: "Processed X emails: Y urgent, Z action, W archived"

The classification prompt is where the intelligence lives:

You are triaging email for a startup founder. Context:
- Co-founders: [names]. Always URGENT.
- Active investors: [list]. URGENT if about funding/legal. FYI if newsletter/portfolio update.
- Team leads: [list]. ACTION unless flagged as blocking.
- Known spam domains: [patterns]. Auto-SPAM.
- Newsletters I actually read: [list]. FYI.
- Everything else: classify by content, not sender.

Override rules:
- Any email containing "wire transfer", "signature required", "deadline today" → URGENT
- Any email from domains I've never interacted with + contains "schedule a call" → SPAM
- Receipts from Stripe, AWS, Render → ARCHIVE

The gog CLI handles the Gmail operations:

# Read unread emails
gog gmail list --unread --max 50 --format json

# Apply label
gog gmail label <message-id> "URGENT"

# Archive
gog gmail archive <message-id>

# Mark as read
gog gmail mark-read <message-id>

The Results

The counterintuitive result: I respond faster to important emails now. Because I'm not wading through noise, the urgent ones hit my Telegram immediately. Before, an urgent email at 10am might not get seen until my next inbox scan at noon.

Try It Yourself

1. Sign up for Mr.Chief and configure the gog skill for Gmail access
2. Set up the classification prompt with your VIP senders and spam patterns
3. Start with a 2x daily cron (morning + afternoon) and review the classifications
4. Tune the prompt based on misclassifications for a week
5. Scale to 4x daily once accuracy hits 95%+

The key insight: don't try to build a perfect classifier on day one. Start permissive (classify more as URGENT), then tighten as you trust it.

I used to live in my inbox. Now my inbox lives in my agent. The emails that matter find me. The rest handle themselves.