# Onboard your AI agent. The free gift from The AI Myth, webinar three. One thing to copy. It asks you about the work you do every week, picks the one task worth handing over first, and then decides how it should run and writes it for you. An agent is onboarded the way a new hire is: with one job it can do well, described properly. This finds that job. You do not need to read this end to end. You need to copy one block. ## The most useful answer might be "not yet" Most people automate the thing that annoys them most. That is usually the wrong one, because the annoying tasks are annoying precisely where they need a person. So the block does not start by asking what you want automated. It asks what you actually do, then it tests one task against one question. **If it gets this wrong, could I undo it?** If the answer is no, that task is not first. Something that sends, pays, deletes or replies in your name can be drafted, but it is not handed over while you are still learning what the machine does well. If the block tells you a task is not ready, that is the block working. ## What this is, and what it is not This is not about writing better messages. That was the first hour of the series. It is not about giving the AI a library. That was the second. This is about **one task leaving your hands**. A good first task is one you do at least weekly, that follows the same steps each time, that works on files you already have, and that you would notice immediately if it came out wrong. ## What Cowork is, in one line Cowork is the agent in the title. It is the same AI you already talk to, with hands: it can open the files in a folder on your computer and do the work, instead of telling you how to do it. It comes with the paid plan. ## Before you start Open Claude. Have in mind two or three things you did last week that you will do again this week. Do not build anything yet. --- ## Block one: the interview Copy everything between the lines and paste it into a chat. Then answer the questions. It asks one at a time. Answer in one sentence. Short beats complete. ``` You are going to interview me about the work I do every week. Then you are going to pick one task to hand over to you first, decide how it should run, and write it for me. Read this whole brief before you say anything. HOW TO RUN THE INTERVIEW Ask one question at a time. Ask it, then stop and wait. Never ask two questions in one message. Keep me to one sentence. If I write four, say the shorter version back to me and ask if that is right. Do not advise me while you are still asking. Do not tell me my answer is great. Ask the next question. STAGE ONE. WHAT I ACTUALLY DO. 1. What did you do last week that you will do again this week? 2. Which of those do you do at least weekly? 3. Which one takes the longest, or gets put off the longest? 4. Walk me through that task in the order you do it. One line per step. 5. Where do the files live? On your computer, in Google Drive or OneDrive, in your email, or somewhere else? Name the folder. 6. When it is done, who reads it or receives it? STAGE TWO. THE UNDO TEST. 7. If I did this task and got it wrong, what would happen, and could you undo it? 8. How would you know it was wrong? What would you look at? STAGE THREE. YOUR VERDICT. BE WILLING TO SAY NOT YET. Say it plainly, in your own words, before you build anything. Say NOT YET if any of these are true: - Getting it wrong sends, pays, deletes, publishes or replies in my name and cannot be undone. - I cannot describe how I would spot a wrong result. - The steps change every time, so there is nothing repeatable to hand over. When the answer is not yet, say why in two sentences, then offer the nearest version you CAN do: usually the same task with you preparing it and me sending it. Do not soften it and do not build it anyway. STAGE FOUR. HOW IT SHOULD RUN. PICK ONE AND SAY WHY. A. TOGETHER, IN A CHAT. Best when the task is occasional, or the first time, or the files are on my computer and I want to watch it happen. B. A SKILL I TRIGGER. Best when the steps are the same each time but I decide when to run it. The files can be on my computer. C. ON A SCHEDULE. Best when it should happen whether or not I remember. Rule you must respect: a scheduled task cannot use a folder on my computer. If this task needs my local files, it is B, not C, and you tell me so. Name the one you picked and say in two sentences why, and what I lose by that choice. STAGE FIVE. WRITE IT. Ask me anything still missing, one question at a time: what a finished result looks like; the names, numbers and spellings you must always get right; what you must never touch; what to do when something is unclear, stop or make your best guess; and which folder the result goes in. Then write the instructions for the shape you picked. Address them to the AI, in the second person. Use my words, not corporate words. Put the whole thing in a code block so I can copy it in one go. At the end of the instructions, always include this line: after finishing, write one short note saying what you did, what you changed, and anything you were unsure about. Underneath the block, tell me in plain steps where to paste it and what to connect. Then ask me one question, once: "What did I get wrong?" Fix what I tell you. Show it again. Then stop. ``` ## Check the work before you trust it The last rule in the block matters more than it looks. Every run leaves a short note saying what it did and what it was unsure about. For the first three runs, read the note and check the result yourself. You are not being careful for the sake of it. You are learning where this thing is strong and where it guesses, which is the only way you will know what to hand over next. If the note ever says it was unsure and it went ahead anyway, that is the task to tighten, not to abandon. ## What to do in the next ten minutes 1. Paste the block into a chat and answer the questions. 2. Take the verdict seriously, including "not yet". 3. Build the one it picked. 4. Run it once while you watch. 5. Read the note it leaves. Then use it for a week. The second task is easier than the first, and you will already know which one it is. --- The AI Myth is a free live hour. You can come to the next one at [theaimyth.com](https://theaimyth.com) If you want help doing this inside your own business, with someone who has already built it, [I will show you how it works](https://metamorphosisworldwide.com/ai-demo-call).