Auto-Drafting Email Replies — Agent Writes, I Approve With One Word

The Problem

I get around 40-60 emails a day. About 15 of them need a real reply — not a template, not a canned response, but something that sounds like me and references the actual conversation.

Here's what that used to look like: open Gmail, read the thread, remember the context, type the reply, proofread, send. Repeat fifteen times. Thirty minutes minimum, usually more. And that's on a good day. On a bad day — when I'm between meetings or deep in code — those replies just pile up. People wait. Deals slow down. I look unresponsive.

The real cost isn't the thirty minutes. It's the context switching. Every email pulls me out of whatever I was actually doing. And the cognitive load of matching tone — formal for investors, casual for collaborators, direct for vendors — burns energy that should go elsewhere.

The Solution

Alfrawd, my master agent running on Mr.Chief, now handles the entire email reply workflow. He reads incoming emails via the gog CLI, drafts replies based on thread history and my writing style, and sends each draft to my Telegram for approval. I say "send" and it's gone. I say "edit: push the timeline to next week" and he revises and sends. One word. Done.

The Process

The setup uses Mr.Chief's Google Workspace skill with the gog CLI, connected to our workspace email. Here's how the pipeline works:

Step 1: Email monitoring via cron

# mrchief cron — runs every 15 minutes
schedule: "*/15 * * * *"
task: |
  Check Gmail for unread emails that need replies.
  For each: read the full thread, draft a reply matching
  Bilal's tone (direct, short sentences, no fluff).
  Send each draft to Telegram for approval.
channel: telegram

Step 2: Draft generation

# Agent reads the thread
gog gmail messages list --query "is:unread label:inbox" --limit 10
gog gmail messages get <message_id> --format full

# Agent analyzes: sender, subject, thread history, urgency
# Then drafts reply internally using context from SOUL.md tone guidelines

Step 3: Telegram approval flow

The agent sends a message like:

📧 Draft reply to Marc (Sequoia) re: Q1 Portfolio Review

"Marc — numbers attached. ARR hit €2.1M, up 34% QoQ.
Happy to walk through the detail Thursday if useful. — B"

→ send | edit | reject

Step 4: Execution

# On "send":
gog gmail messages send --to "marc@sequoia.com" --subject "Re: Q1 Portfolio Review" --body "..."

# On "edit: change Thursday to Friday":
# Agent revises draft, sends updated version for confirmation

# On "reject":
# Agent archives the draft, logs the decision

Step 5: Audit trail

Every sent email gets logged to daily notes:

## Emails Sent — 2026-03-12
- 09:14 → Marc (Sequoia): Q1 Portfolio Review reply ✅
- 09:22 → Sarah (Mr.Chief user): Support follow-up ✅
- 10:05 → Legal team: NDA revision — edited timeline ✅

The Results

The revision rate tells the real story. 88% of drafts go out unchanged. The agent learned my tone from months of sent emails — it knows when to be formal, when to be blunt, when to add warmth. The 12% that need edits are usually timing or commitment changes, not tone fixes.

Try It Yourself

1. Sign up for Mr.Chief and configure the gog CLI with your Google Workspace account
2. Set up the Gmail monitoring cron with your preferred check frequency
3. Define your tone in SOUL.md — the agent uses this as its style guide
4. Start with "draft only" mode — approve everything manually for the first week
5. Once trust is built, enable auto-send for routine reply categories (confirmations, scheduling, acknowledgments)

The key insight: you're not removing yourself from email. You're removing the typing. Every reply still gets your eyes and your judgment. You just stop being the keyboard.

I didn't hire an assistant. I built one that reads faster than I do.