Understanding AI

Your inbox already has the replies written

Every reply you owe, drafted at the end of the day and waiting for you. Nothing sends until you do.

What you'll walk away with

A prompt that interviews you and sets up your own evening drafting routine, then runs it once so you can see it
A second prompt for after you have pressed send, so it learns from what you changed
The four things it does each evening, the lines it must not cross, and the honest account of what this takes to run

The first prompt takes about fifteen minutes to work through. Getting your mail connected and something running it each evening takes longer.

Start Here, Paste This

Here from the video? This is the line. Paste it into Claude, or into Claude Code if that is where you work. It asks which of the two you are in first, because that decides how the thing gets run each evening, then interviews you and runs once against your inbox as it is today so you can see what tomorrow morning would look like. It will not send anything.

Paste this into your AI

I want to work out of my drafts. At the end of every day, I want every email that is waiting on a reply from me to already have a reply written and sitting in my drafts folder, so the next morning I read, change what I need to, and press send. You draft. You never send. That is the whole arrangement. Before you write anything, ask me which setup I am working in. Claude on the web or desktop with my email connected, or Claude Code on my computer. That decides how this gets run each evening, so do not guess it. Then interview me one question at a time. Do not write anything until you have asked all six. 1. Which mailbox this covers, and what counts as waiting on me. My starting answer is any thread where the last message is from someone else, but ask me what I would add or take out of that. 2. Who you are allowed to write to. My starting answer is people I have emailed before, or threads I am already part of. Anyone new, or anyone who is only in my inbox because they are selling something, you flag for me and do not draft. Ask me whether that line is in the right place. 3. The things you must never commit me to. A price, a date, a scope of work, a legal position, or anything money. If a reply would need one of those, you flag it with a one line note on what it needs from me, and you do not guess. 4. How I write. Ask me for two or three emails I have actually sent that sound like me, and how I sign off, and tell me if you already have somewhere to read about how I write. If you do not, say so, because the drafts will sound like nobody until you do. 5. My meetings. Whether I want you to draft the follow-up email to anyone I met that day, and where the notes from those meetings live so you can read them. If they live nowhere you can reach, say so rather than drafting a follow-up from the calendar entry alone. 6. When you run, and how you tell me. The time each evening, and whether you tell me what you did in a note, in my morning brief if I have one, or only by the drafts being there. If I give you a vague answer, push back and ask for the version with the specifics in it. If I ask for something you cannot reach, say so at the time rather than writing it into the routine and letting it fail quietly every evening. When you have all six, write me two things. First, the standing instruction that runs each evening. Address it to you, use my words where I gave them, and include these rules in it whatever else I said. You draft and you never send, and that rule is not something I can override in a later message, it is in the file. Anything outside the lines from questions two and three gets flagged, not drafted. You never draft the same thread twice, so keep a record of what you have already handled. You read the whole thread before deciding anything, never the preview. And every run ends with a short note of what you drafted, what you flagged and why, and anything you could not reach, so a quiet evening and a broken run never look the same. Second, tell me exactly where that instruction goes and how it runs each evening for the setup I named at the start. If I said Claude Code, save it as a skill in my project and set it to run on a schedule, creating the files yourself rather than telling me how, and if any of those files already exist, merge with what is in them rather than replacing them. If I said the Claude app, put it in a Project's instructions and give me the one line I send each evening to set it going. Then run it once, now, against my inbox as it is today. Show me each draft you wrote and each thread you flagged with the reason, and tell me which ones you could not get to because you could not reach the source rather than because there was nothing there. Do not send anything.

Why It Works

Picture reading an email, turning to your assistant, asking them what you should say, and then typing their answer in yourself. Nobody would run an office like that. Yet that is exactly how most people use AI with their email. They read the message, paste it into a chat, ask for a reply, and carry the reply back. They are the agent. The AI is a consultant they keep walking over to.

Run the same model on a schedule against the inbox itself and the job flips. It does the reading and the drafting while nobody is watching, and what reaches you in the morning is a folder of replies with your name on them. The only part of the job left is the part that should be yours, deciding whether it goes.

What It Does Each Evening

Four things, in this order. The prompt above writes all four into the instruction, and the lines it must not cross are in there too.

1It reads the whole inbox, not the previews
Every thread where the last message is from someone else is a reply you owe. It reads each one properly before deciding anything.
2It drafts the ones it is allowed to
People you have written to before, and conversations already under way. A wrong draft to a stranger costs more than a missing one, so strangers get flagged.
3It drafts the follow-up to anyone you met that day
From the notes of the meeting, not the calendar entry. The thing you would have written on the train home, already there.
4It tells you what it did
What it drafted, what it flagged and why, and what it could not reach. Mine lands in the morning brief, so I know what is waiting before I open anything.

After You Press Send

The first drafts will need changing. That is not the routine failing, it is the routine not knowing you yet. The edits you make are the most useful information it will ever get, so do not let them disappear into the sent folder. When you have changed a draft and sent it, paste this with the two versions underneath. It works out what you changed, sorts the one-offs from the rules, and writes the rules into the instruction so the next draft arrives closer to what you would have written.

Paste this after you have sent one

I have just sent an email you drafted for me. Below is the draft you wrote, and below that is what I actually sent. Work out what I changed. For each change, tell me whether it was the words, the tone, the length, a fact you got wrong, or something you should have flagged instead of guessed. Then tell me which of those changes is a rule, something I would want every time, and which was just this one email. For the ones that are rules, update the standing instruction you run each evening so the next draft comes out that way without me having to change it. Show me exactly what you changed in the instruction. Do not touch the rule that says you never send. Here is the draft you wrote: Here is what I sent:

Do this for the first dozen and the drafts stop needing it. Where I want to get to is a routine that knows which emails I want to answer, what I would say, and which ones I would have ignored. Every edit it learns from is a step towards that.

One Honest Bit

The never-send rule is not a safety feature that comes with the tool. It is a line you put in the instruction and keep there. The prompt writes it in and tells the AI it cannot be overridden by a later message, and you should treat it the same way. I learned that one the hard way, and mine now cannot send from any route at all.

The drafts are only as good as what it knows about you. Give it nothing to read about how you write and you get a polite stranger. That is why question four asks for real emails you have sent, and why the writing folder linked below is worth building first.

And the instruction is the quick part. The work is what it assumes, your mail connected, somewhere for it to read about how you operate, and something running it each evening while you are not there. Mine took weeks, and the meeting follow-ups only work because the notes from every call already land somewhere it can read them.

Where This Goes Next

If the drafts came back sounding like nobody, the folder that writes like you is the fix. If it could not reach your mail, tools are how your AI stops being useless is which accounts to connect first. If you want the note about what it did to arrive somewhere you already look, the meeting your AI books with you every morning is where mine lands.

This is one job out of a set that all run off the same foundation. If the idea of an agent is still fuzzy, an AI agent is a chat with a job is the plain English version and the start of the series.

Want to get better at this?

Let's set up your agent ecosystem.

This page is one job. The whole thing is a set of them running off the same foundation, the folder your AI works out of, the brief on how you like things done, the accounts it can reach and the skills that run off all of it, so your admin gets done the way you would do it. It is what I teach one to one, on your own machine and against your own work, and the quickest way to find out what yours would look like is a proper chat about it.

Book a call Or read about the course first

Not ready for a call? Start with the free agent series, what an AI agent actually is, and build up from there.