What Rules Should I Give an AI Agent for Drafting Replies?

The goal I dictated on October 7, 2026 included replies. At 12:17 I said I wanted to manage my calendar and not miss replying to critical emails. The goal covers what I want. How each reply reads comes from a separate section of the file on the session page, three sentences long.

What goes in the drafting rules?

The file on the session page has a section called How you write drafts. The section in full:

From the file on the session page. Short, clear, in my voice. Lead with what the other person needs to know. If a reply requires me to make a decision first, say so, and don't draft a reply that pretends the decision is already made.

Fill in your own version. The middle column is the file's wording; the right column is an example, replace with yours.

| Rule | What the file says | Example, replace with yours | | --- | --- | --- | | Length | Short, clear | Two short paragraphs at most | | Voice | In my voice | Plain words, signed with my first name | | Order | Lead with what the other person needs to know | The first line answers what they asked | | Decisions | If a reply requires me to make a decision first, say so | If I haven't decided, list what I'm deciding and leave the answer blank |

The file is written in first person. Write yours the same way.

Where do the drafts land?

The file says where each reply goes:

From the file on the session page. A drafted reply, written in my voice, saved as a Gmail draft if Gmail is connected, or written out in this chat if it isn't.

So with Gmail connected, the draft sits in your Gmail drafts. Without it, you read the draft in the chat. Either way it's a draft. The file also says everything stays in the chat and in draft form:

From the file on the session page. Everything stays in this chat and in draft form, whether that's a Gmail draft or text in this window.

That's an instruction in a file, and a file can't lock anything. The control that decides whether an AI agent can send, update, create drafts or forward emails is in the connector's tool settings. At 44:19 I showed that section for Gmail. How Do You Control What an AI Agent Can Do in Your Gmail? walks through it.

How do I correct a draft that sounds off?

Tell the AI agent in one sentence. The file says what it should do with that sentence:

From the file on the session page. If I tell you something is off, in one sentence, update the next answer to reflect it.

The file goes on to say the AI agent shouldn't ask for more explanation unless the correction doesn't make sense, and that it should keep the fix and not quietly revert to the first version next time.

Example, replace with yours: "Too formal. Shorter, and start with the date."

What goes wrong

A draft that pretends you've decided. The file names it: a reply that needs a decision from you first shouldn't be written as if the decision were made.

A correction that doesn't stick. The file asks the AI agent to keep the fix and not go back to the first version.

What the file doesn't say: how the AI agent learns to write in your voice. It has the phrase in my voice and nothing about how it gets there, and I didn't cover that step on the recording.

Where this comes from

This article grows from the session I taught live with Women Applying AI on October 7, 2026. The quoted passages come from the file published with that session, taken verbatim. The tables, steps and checklists around them were drafted by my team's AI agents from those words and my published pages, and checked against the recording before publishing. The file is on the session page.

Build Your First AI Agent in One Hour: the full session with the clips, the prompts and the transcript.

Related reading: How Do You Control What an AI Agent Can Do in Your Gmail?, Authentic AI Automation: Maintaining Your Voice While Scaling Business Operations.