Have an agent draft every reply in your inbox before you open it

Your agent drafts replies to every email before you read it, so you edit instead of writing from scratch.

For anyone whose inbox is the job · 8 steps · 7 min

How it works today.

You open your inbox and see seventeen messages that need responses. Each one requires you to read, understand the context, decide what type of reply it needs, then write it out. Even routine requests take three to five minutes because you are crafting each response from a blank page.

The reading happens fast, but the writing does not. You find yourself typing variations of the same explanations, scheduling language, and follow-up questions over and over. By the time you clear your inbox, two hours have passed and six new messages have arrived.

Before you start.

All of it has to be true, or step one fails in a way that is annoying to debug.

Email platform that supports rules or filters (Gmail, Outlook, etc.)
Access to an AI agent service that can read and write emails via API
A folder or label system to route messages through the drafting workflow
Sample emails from the past month showing your typical response patterns

The steps.

Set up email routing to trigger draft creation
Create a rule that moves incoming emails to a 'Draft Queue' folder and triggers your agent. In Gmail, use filters with labels. In Outlook, use rules with folder actions. The agent needs read access to this folder and write access to create draft replies.
02
Train the agent on your response patterns
Feed your agent 20-30 recent email threads where you provided good responses. Include the original message and your reply for each. This teaches the agent your tone, typical solutions, and how you handle different request types.
Paste this
```
Analyze these email exchanges and identify my response patterns. Note my typical tone, how I structure replies, what information I usually request, and how I handle scheduling, questions, and requests. Create a style guide I can reference: [paste your email threads here]
```
Configure the agent to create drafts automatically
Set the agent to monitor your Draft Queue folder and create a draft reply for each new message. The draft should appear in your email platform's draft folder, not as a sent message. Test this with a few sample emails first.

Create response templates for common scenarios

Build templates for your most frequent email types: meeting requests, information requests, project updates, and follow-ups. Each template should include your standard language and leave blanks for specific details the agent will fill in.

Paste this

Create email response templates for these scenarios based on my communication style: 1) Meeting scheduling requests 2) Questions about [your main work area] 3) Project status updates 4) Follow-up on pending items 5) Requests for information or documents. Use my tone from the previous examples and include [specific details] placeholders where the agent should customize.

Set up draft review notifications
Configure your email platform to notify you when new drafts appear, but not to send them automatically. You want to see that drafts are ready for review without them cluttering your main inbox view.
Establish your draft review process
Check your drafts folder twice daily at set times. Read each draft against the original message, edit for accuracy and tone, then send or delete. Most drafts should need only minor edits if the agent is trained properly.

Create escalation rules for complex messages

Train the agent to flag messages it cannot draft well: legal issues, sensitive personnel matters, or highly technical discussions. These should skip the drafting process and go directly to your main inbox for manual handling.

Paste this

Flag emails for manual handling if they contain: legal language, personnel issues, budget discussions over [your threshold], technical problems requiring deep expertise, or emotional/conflict situations. Instead of drafting, add a label 'MANUAL_REVIEW' and move to my priority inbox.

Monitor and improve draft quality weekly
Review which drafts needed heavy editing or deletion. Adjust your templates and agent training based on patterns. If the agent consistently misses certain types of requests, add specific examples to its training data.