Understanding AI

Your only job is the review

Ask for the email and the draft is in your drafts folder, not in the chat

What you'll walk away with

  • Three steps, and the first drafts sitting in your drafts folder by the end of them
  • The prompt that goes through your inbox and writes the replies you owe, without sending any of them
  • A second prompt that turns your corrections into an instruction it keeps, so the next lot need less changing

Connecting your email takes about two minutes. The first drafts land a few minutes after that.

Step One, Connect Your Email

Here from the video? Three steps, and this is the only one that is not a paste. In Claude you go to Settings and then Connectors, you pick Gmail or Outlook out of the list, and you log in the same way you log in to it anywhere else. Every other AI has a version of the same screen under a slightly different name. There is nothing to install and nothing to configure, and if your work account will not allow it, that is your IT department's setting rather than something you have done wrong.

Slow down at the login step. Whichever account you sign in with is the one it gets, and the only one it gets. If you have a work inbox and a personal one, check which of them the browser had you signed into already rather than clicking straight through, because that click is the actual decision here.

It asks before it acts. Reading your mail is one permission and writing into it is another. The first time it wants to put something in your drafts it tells you what it is about to do and waits for you to say yes, and you can keep it asking every single time rather than waving it through for the whole conversation. That is a setting rather than a personality trait, so decide it on purpose.

You can take it back. Every connection comes off from the same screen you made it on, and your Google or Microsoft account has a security page listing every service you have ever connected that cuts any of them off in one click. Nothing here is a one way door. Tools are how your AI stops being useless goes further on which apps to connect in what order, why read only gets you most of the value, and the four I would never connect at all.

Step Two, Paste This

Paste it in exactly as it is and let it run. It goes through your inbox, finds the five threads most obviously waiting on you, reads each one properly, and writes a reply into your drafts folder rather than into the chat. It will not send anything, and it will not guess at a price or a date.

Paste this into your AI
Go through my inbox and find the threads where the last message is from someone else and it is still waiting on a reply from me. Take the five most important ones, and tell me honestly how you decided which five. Read each thread properly before you write anything, the whole thread and not the preview. Then write a reply to each one and save it as a draft in my email, on the right thread, so I can find them all together in my drafts folder. You never send anything, now or later in this conversation, whatever I say to you afterwards. Where a reply would need me to commit to a price, a date, a scope of work or a legal position, do not guess it. Leave that one as a draft with one line at the top saying what you need from me. When you are done, list what you drafted, what you left alone and why, and anything you could not reach. Do not paste the drafts in here, because I am going to go and read them in my email.

Now go and open your drafts folder. That is the whole idea, and it is worth seeing before you read another word of this page. The same move works on your calendar once that is connected, so ask it to put the thing in your diary instead of telling you when you are free.

Step Three, Make It Keep The Corrections

The first drafts will need changing, which is not the thing failing, it is the thing not knowing you yet. Your edits are the most useful information it will ever get about how you write, so do not let them disappear into the sent folder. Read the drafts, send the ones you are happy with, then paste this. It asks you what you changed, works out which of those are rules and which were one offs, and writes the rules into an instruction it keeps and reads before it drafts for you again.

Paste this once you have read them
I have read the drafts you wrote me and I want the next lot to come back closer to what I would have sent. Ask me about them one at a time. For each draft, ask what I changed and why, and keep asking until you understand whether it was the words, the length, the tone, a fact you got wrong, or something you should have left for me rather than guessing at. Then sort what I have told you into two lists, the things I would want every single time and the things that were only true for that one email, and show me both so I can move anything you have put in the wrong one. Then write the ones that are rules into a standing instruction you keep and read before you ever draft for me again, and put the rule that you never send into it whatever else I said. Show me exactly what you wrote, in full, so I can correct it now rather than finding out next time. Ask me whether I am working in the Claude app or in Claude Code, because that decides where it goes. If it is Claude Code, save it as a skill in my project and merge it with anything already in there rather than replacing it. If it is the app, put it in a Project's instructions and tell me where to look for it.

Do that for the first dozen and the drafts stop needing it. Every correction you keep is one you never have to make again, and that instruction is the thing you actually own at the end of this.

The Idea

Almost everybody uses AI the same way, and it works. You go to it, you paste in what you have got, it does the work, and it hands the answer back to you. Then you finish the job yourself. You copy the email out of the chat and paste it into your email. You read the three times it suggested and type the meeting into your calendar. You take the write-up and save it into the folder it should have been in.

That last bit is small enough that nobody counts it, which is exactly why it is worth counting. It is the same few minutes every time, on every job, all day, and it is the part of the arrangement where you are doing the fetching and carrying for something that is meant to be doing it for you.

The review stage is what happens when the answer stops arriving in the chat. You ask for the email and the draft is already sitting in your drafts folder with the right person in the To line. You ask it to sort the meeting and the event is on your calendar. Nothing about the work is different and nothing about the AI is cleverer, the only thing that changed is where the finished thing landed. What is left for you is the part that was always yours, reading it and deciding whether it goes.

If you are the one moving the work between the chat and the place it belongs, you are the agent and the AI is a consultant you keep walking over to. Getting to the review stage is how that swaps round.

What It Looks Like On Real Jobs

The email is the one worth doing first because you feel it immediately. Here are the other two, and where the answer ends up once those accounts are connected as well.

  1. 1Reply to this one

    A draft in your drafts folder, on the right thread, addressed to the right person, unsent. You open your email in the morning and the reply you owe is already written.

  2. 2Sort this meeting out

    The event on your calendar at a time that is genuinely free, with the invite ready to go. Not three suggested times you then have to type in yourself.

  3. 3Write this up

    The document in the folder where that kind of document already lives, named the way the others are named, rather than a wall of text in a chat window you will lose by Thursday.

Do them one at a time rather than all three at once. One account connected and one job landing in the right place tells you whether you want the rest of it, and it costs you an evening instead of a weekend.

One Honest Bit

Connecting your email means the contents of the emails you ask it about get sent to the company that makes the AI, because reading is receiving and there is no version of this where it reads your inbox without your inbox leaving your computer. For most people writing ordinary business email that is a trade worth making, and it is the same trade you already made with whoever hosts your mail. It stops being worth it the moment the inbox holds other people's personal information, client records or anything your workplace has rules about, because that decision is not yours to make on your own and asking first is a two minute conversation.

There is a second question underneath it that people conflate with the first. Whether your conversations can be used to train the model is a separate setting from who is able to read them, it lives in your account settings, and the default is not always the one you would pick. Go and look at it rather than assuming, whichever way you then set it.

And the review is a real job, not a formality. A draft you did not read is worse than no draft at all, because at least a blank page makes you write something. The point of getting the work to land where it belongs is that the only thing you spend your attention on is the judgement, so spend it there and read the thing properly before you press send.

Where This Goes Next

There is a stage past this one, where you stop asking. The routine runs on its own at the end of the day, goes through everything waiting on you, and the drafts are already there in the morning without you having started anything. Your inbox already has the replies written is that one, with the prompt that sets it up and the rule it must never cross. And the meeting your AI books with you every morning is where the note about what it did lands, so you find out what is waiting before you open anything.

If the word agent is still doing a lot of unexplained work, an AI agent is a chat with a job is the start of the series and the plain English version. The brief is your agent's personality is how it knows the way you want things done, and the context is what your agent actually knows is the background it needs in front of it before any of this comes back sounding like your business rather than anyone's.

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.

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