# Onboard your project. Two things to copy. The first one works out whether this should be a project at all, and then, if it should, writes that project's instructions for you. The second one makes your AI keep a journal of the work. You do not need to read this end to end. You need to copy two blocks. ## The most useful answer might be no Most people set up a project because they heard they should. Then they put everything in it, and the AI gets worse, not better. So the first block does not start by filling in a form about your project. It starts by asking what you are actually trying to get done, and why you think a project is the right home for it. Then it tests that. **If a project is the wrong answer, it will tell you so, and tell you where the material should go instead.** That is not the block failing. That is the block doing the most valuable thing in this document. Here is the test underneath it, in plain terms. A project is a room with the door shut. What you put inside it is visible inside it and nowhere else. Your other chats cannot see it. That is exactly what you want when the material only matters for one piece of work. It is exactly what you do not want when the material is the kind of thing you need everywhere. If you find yourself wanting the same facts in every chat you open, a project is the wrong home. That material belongs in your personal instructions, or in memory, where everything you do can reach it. And if the answer to "what is this project" is "my whole business", that is not a project either. That is the page about your business, which is what the first webinar gave you. Keep it. It is not this. ## What this is, and what it is not This is not about you. It is not about your business. If you were at the first webinar, you already have those. You have the page about how you work, and you have your business on one page. Keep both. There is a place for them in what follows. This one is about **one piece of work**. A project can be a part of your business. It can be one client. It can be one job you are doing for one client. If you can say what it is in a sentence, and you can say when it will be finished, it is probably a project. ## Before you start Open Claude at claude.ai and sign in. Do not create the project yet. Open a normal chat first and paste in the block below. Let it tell you whether you need one. --- ## 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 giving it enough to stop guessing. ``` You are going to interview me about a piece of work I am thinking of setting up as a project. Then you are going to tell me whether it should be a project at all. If it should, you will write the project's instructions 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. If an answer is vague, ask once more in plainer words, then move on. STAGE ONE. UNDERSTAND WHAT I AM ACTUALLY TRYING TO DO. Do not treat this as a form. You are trying to understand my thinking. 1. What are you trying to get done? 2. Why do you think a project is the right home for it? 3. What made you want to set this up now? What keeps happening that you are tired of? Ask a follow up if my answer to any of these tells you nothing. One follow up. STAGE TWO. NOW TEST WHETHER IT SHOULD BE A PROJECT. Ask these, still one at a time. 4. Is the information for this work only useful for this work? Or do you need the same information in your other chats too? 5. Is this a bounded piece of work with an end you can describe? Or is it everything about your business? 6. Will anyone else need to see this, or is it only you? 7. What actual files or documents would go into it? Name them. STAGE THREE. GIVE ME YOUR VERDICT, AND BE WILLING TO SAY NO. Decide, and say it plainly in your own words before you do anything else. Say NO, THIS SHOULD NOT BE A PROJECT if any of these are true: - I need the same information in my other chats. A project is a closed room. What is inside it is visible only inside it. If I need these facts everywhere, they belong in my personal instructions or my memory, not in a project. - It is "everything about my business". That is a single page about the business, not a project. Tell me to keep it where it is. - There is no end to it, and no files that belong only to it. Then it is a subject I talk about, not a piece of work. When the answer is no, do not soften it and do not offer to build it anyway. Tell me why in two sentences, and tell me exactly where the material should live instead. Then stop and ask if I want to test a different piece of work. Say YES if the information is specific to this work, it has an end, and there are real files that belong to it. Say briefly what convinced you. STAGE FOUR. ONLY IF THE ANSWER IS YES, WRITE THE BRIEF. Ask me anything still missing from this list, one question at a time: what done looks like and how we will know; who sees the result and judges it; what is already decided that you should not reopen; what I know that you would get wrong if you guessed; what is NOT part of this; the names, numbers and spellings you must always get right; how I want you to behave in here, ask first or draft fast, agree or push back; what you must never do; and which two or three real files should go in. Then write the project instructions. Write them addressed to the AI, in the second person, so they can be pasted straight into a project's instructions field. Use my words, not corporate words. Put the whole thing in a code block so I can copy it in one go. Underneath, list the two or three files I should add to the project, and nothing more than that. Then ask me one question, once: "What did I get wrong?" Fix what I tell you. Show it again. Then stop. ``` If it tells you no, that is a real answer and you have saved yourself a folder that would have made your AI worse. Ask it where the material goes and put it there. If it tells you yes, now create the project, name it after the work, and paste the instructions it wrote into the project's instructions field. That field is the one place your AI reads every time, in every chat in that project. That is the whole point of it. --- ## Block two: the journal Here is the problem the journal solves. Your AI does not start fresh every message. It reads its own system instructions, then your project instructions, then whatever it has in memory, then the chat you are in. That is a lot. But when you close a chat and open a new one, most of what you just worked through does not come with you. Some of it went to memory. Not all of it. And it will not go hunting through your old chats for the rest, because it is built not to waste effort. So you explain the same thing again. That is not the AI being stupid. That is you having no record. A journal is the record. It lives outside the chat, so it survives the chat. This works whether or not you ended up with a project. ### Pick one place for it Google Drive, or OneDrive. Pick one. Do not do both. Connect it to **one folder**, not your whole drive. Make a new folder for this work and connect only that. Your AI does not need your email, your customer list, or the rest of your files to keep a journal. It needs one folder it can write to. If that still makes you uneasy, that instinct is right and you should keep it. Connect the smallest thing that works. ### Then paste this in Add this to your project instructions, underneath what the interview wrote. If you did not make a project, paste it at the top of the chat you work in. ``` KEEP A JOURNAL OF THIS WORK There is a folder connected to this. Use it to keep a running journal. At the end of every working session, before we stop, write a short entry and save it to that folder. One file per session, named with the date. Each entry says: - The date. - What we were trying to do. - What we decided, and why. - What we tried that did not work, so we do not try it again. - What is still open. - What I should pick up first next time. Keep it short. Six lines beats a page. Write it in plain words, the way I would say it out loud, not as a report. At the START of a new session, read the most recent entries first, then tell me in one line where we left off. Do not make me repeat myself. If I say "log this", write the entry immediately without waiting for the end. Never edit an old entry. If something in an old entry turns out to be wrong, say so in a new one. ``` That last rule matters more than it looks. A journal you go back and tidy is not a record of anything. Leave the mistakes in and correct them forward. ### One more thing, for later All of this assumes you are working in a chat window, which is where nearly everyone is right now. That is why the journal has to go to a drive. There is another way to work, where the AI runs on your own machine and the journal is just a file on your computer. Nothing goes to a drive at all. That is a different setup and a different hour. That hour is Let AI Work for You. --- ## What to do in the next ten minutes 1. Pick one piece of work. Not your biggest. One you are actually doing this week. 2. Paste block one into a chat and answer the questions. 3. Take the verdict seriously, including if it is no. 4. If it is yes, create the project and paste in the instructions it wrote. 5. Make one folder in Google Drive or OneDrive and connect it. 6. Paste block two. Then use it for a week and see what it stops asking you. If the answers still feel generic, it is almost never the AI. It is usually that the instructions are describing your business instead of the work. Go back and answer question one shorter. --- 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).