What tonight is for
AI can already do almost anything you want it to do, and it keeps getting better. Most people are still messy with it, still overwhelmed by it, and still doing the same slow things by hand.
So tonight has one job. You find the number one thing you need to do to get more out of AI, and you actually do it tonight. This is not a list of ten things you will start next month. It is one thing you finish before you sleep.
Marc's honest words on this: "I think I'm still a level 4 or 5 out of 10 with AI. I don't think I'm an 8 or 9." In the room he put it even more plainly: "I just suck at it myself as well."
You will never feel 100% with AI, and that is not a problem you can fix. There is a new model almost every week. Most of the content you see about AI is built to make you feel behind, not to make you feel confident, so watch the fear of missing out. The one thing you can always work on is the principles, and the way to work on them is 1% at a time. Marc quotes Tony Robbins on this: perfectionism is the lowest standard there is, because nothing can ever be fully finished. Build it to the point where it does what you need, then refine it as you go along.
Nobody has to go all in on AI today. A 5 out of 10 is fine if AI is already giving you what this season of your life needs. Watching other people build AI things and feeling behind is a bad reason to act.
One rule before you copy anything from this page. Whatever you take from here, the first thing you type after you paste it is: please adapt it for me.
AI does two things: save costs or create value
Marc came back to this frame all night. Whatever you point AI at, it is doing one of two jobs for you.
Save costs
Costs are not only money. Saving costs means saving time, effort and energy too. The catch is that you have to put more energy and effort in first, before it saves you anything. A lot of people expect AI to hand them back their time on day one, and it does not work like that. It takes a bit of work and a few rounds of refining.
Create value
This one means one of two things. You do better work for the clients you already have, or you go and find new clients.
If you are below S$10K a month, create value first.
Marc repeated that on purpose. Below S$10K a month, do not just think about saving costs. Think about what creates a lot more value and brings cash in while it saves you time as well. When he built this deck and this page, it was both, and his words for it were, "it allows you to create more value with less time."
Now for the trap that catches people. The day before this event Marc spoke with a prospect who has 20,000 followers on Instagram. She gets a lot of messages, and she wanted to build AI agents to answer all of them. It is very tempting to call that the priority. But she had not refined her messaging yet, or her offer, or her DMs. If none of that is working, automating it only sends more of the same. His words: "So be very mindful to be caught up in the save-cost trap." Do the manual work for now, and automate it once it works.
Marc's own reframe on this was simple.
The idea is not that you need to learn AI. It is working out what will help you make money first, then using AI to support it.
If money is not your main priority this season, ask the same question about the main problem you are solving or the outcome you want, then use AI to support that.
Marc's five AI best practices
1. Data beats prompts
The people who get great output are not the ones with the cleverest prompt. They are the ones who fed the AI their own material first: transcripts, frameworks, past posts, real decisions. Marc's line is "Data is king." He collects three kinds. Story data is what he has lived through. Business data is his offers and his client results. Market data is what clients say, ask about and struggle with.
The line to type:
That is his capture line. When something worth keeping shows up, he tells his AI to add to inbox, and the raw material is saved instead of lost.
None of this works if you never capture anything, so here is how Marc actually does it. Record your online calls with Fathom, so the recording and the transcript are saved without you thinking about it. For conversations that happen in person, use Plaud or Pocket, or your voice notes, as long as you record it somehow. Then tell your AI:
Imagine five or ten years of your business conversations sitting in there for your AI to read.
You can do the same with photos. Make a photo library inside your second brain, upload only the photos you know you will use, then say "help me to categorise this" and let the AI name and sort them. Marc does not upload all 25,000 of his photos, only the ones he knows he needs.
One small hack for when you cannot be bothered to open the folder where your files live: ask for the file in the chat instead. "Can you just give me the exact file in the chat itself?" Then you read it and edit it right there.
2. Ask the AI to ask you questions
Most bad output happens because the AI is guessing about you. So flip it and make it interview you before it does anything. It costs you one extra round of messages and it changes the whole result.
There is no script for this. Marc told the room not to copy anything word for word, because this is a habit, not a prompt.
The line to type:
The rule for how many questions: the bigger and more important the job, the more questions you want. Something small and simple does not have to go that deep.
When the output still feels off, just ask for more questions before it creates anything, and tell it what you were not happy with in the last try.
3. A separate agent checks the work
Marc's rule: "Never let the AI that made it be the AI that checks it." An AI marking its own homework goes easy on itself, like an employee grading themselves. So a second set of eyes reads the work with one job only, which is to find what is wrong.
The line to type:
QA is short for a quality check. In Marc's setup, even the short version, "do a QA", starts a separate checker. If your tool cannot do that, open a brand new chat, paste in the work, and tell it to only look for mistakes.
You can also hand the work between two AIs. Marc does this: ChatGPT creates it, then Claude reviews it.
4. Always share the outcome and the why
Tell the AI what you are trying to get and why it matters, not just the task. Then, every time you change what it gave you, say what you changed and why, and ask it to remember that. Marc's words: "You need them to capture your human taste. The more you do it, the easier it gets."
The line to type:
Swap [the task] for whatever you actually asked it to do.
5. Do not accept no as an answer
When the AI says it cannot do something, that is usually the start of the conversation, not the end. Ask what it can do instead, ask for another way, or give it more of the picture.
The first move is to ask for a draft, not the finished thing. You do not have to deploy anything or create files on the first pass, and a draft costs you far fewer credits than a build you throw away.
The line to type:
The second move is a safeguard, for when you are nervous about what it might do.
The line to type:
Most of the time the AI just did not have enough from you yet.
Lines Marc typed live
These are the lines Marc read out or typed on screen, tidied a little so they read cleanly. Copy any of them and change the details to fit you.
Audit my website ranking on Google, also called SEO
As an expert SEO consultant, can you give me a complete audit of my SEO ranking for my website and give me any suggestions to help me immediately improve my SEO ranking?
Get the file into the chat
Can you just give me the exact file in the chat itself?
File my meeting transcripts
Please add my transcripts from my meetings into my second brain.
Make it ask me questions
Feel free to ask me as many questions as you need, or ask me more questions to clarify, so that you can create a better output.
Get a separate checker
Please create a separate agent to QA the output.
Spend my unused credits overnight
I have unused credits. Please help me to maximise all your credits by doing a complete audit of my entire second brain setup and suggest areas to refine and improve, or even suggest ways that can grow my business or make more money. By the way, this is an overnight task until 5am, so please use your credits accordingly.
Model a page I like
Please help me model after this website exactly as it is and design it for me.
Build the thing
Build me the slides, build me the playbook. Here's the model. Please read my second brain. Please read my brand guide. Help me to build.
Adapt someone else's prompt
Please adapt it for me.
Put a password on a page
I want to create a password so that only people with a password can access the link.
Confirm before you send
I want you to do this, but I'm afraid you do it wrongly. What can you suggest to make sure that you don't do it wrongly? Or can you make sure that before you send, you give your confirmation to me first?
Correct it and lock it in
Correct it, and then say: next time, remember this, so that we don't forget this again.
How Marc builds websites, toolkits and playbooks in minutes
Marc has never had a professional designer build anything in his business, not once in all these years. Since June 2026 he has put over 100 pages, decks and tools live on his own domain. He also rebuilt a call booking page that was costing him USD 97 a month, and it took about three minutes while he was half watching a football match. These are the five steps he uses.
Step 1: Model it
You do not need design taste. You need one page or design you already like. Show it to your AI, tell it to model after that, then say what to change so it becomes yours: keep your colours, simplify it, swap the fonts, tone it down. Do this once, then keep the finished note in the Notes app on your phone, because every build after that starts from the look you already picked.
Copy a Look You Like (Model It)
How to use it
- Open a brand new chat in ChatGPT, Claude or Gemini. Brand new means an empty chat, not one you were already talking in. (If you already use Claude Code or Codex, those work too.)
- Copy everything inside the grey box below, from the first word to the last, and paste it into the chat. Then send it.
- Answer its questions in one message, rough answers are fine.
I want you to save one design I like as my reference style, so everything we build from now on starts from that look. Sometimes I will ask for this in my own words, like "please help me model after this website exactly as it is and design it for me". That is the same request, and it means copy the layout and the feel, then put my own colours, my own fonts and my own words in. However I say it, run the one flow below. Before you do anything, ask me these five questions in one message, then wait for my answers: 1. What is the page, deck or design I like? Describe it in words: is it mostly dark or mostly light, are the words huge or small, is it busy or empty. Two or three lines is enough. I can also paste a link or upload a screenshot if I have one handy, but words on their own are enough. If you cannot open a link I paste, say so straight away and just work from my words. 2. What do I like about it? Rough words are fine. 3. What are my own brand colours? Just name them, like "black and neon green". If I happen to have the colour codes, the ones that look like #1A1A1A, I will paste those too. 4. What fonts do I use, or what should my words feel like when someone reads them? 5. What am I going to build with this reference: websites, slide decks, playbooks, or all of them? After I answer, do these five steps in order. Step 1. Describe the look back to me in plain words a 12 year old would follow. Cover how the page is laid out, how much empty space there is, how big the headings are next to the body text, how many colours are used and where each one shows up, and the tone of the writing. Step 2. Turn that description into one short note called "my reference style". Do not print it yet. Keep it for Step 5. Keep it under 300 words and keep the sentences short. Step 3. Add these three rules to the note, in these words: - Copy the look of this, then adjust it to fit me. Keep my colours. - Simplify. Cut anything that is only decoration. - Swap the fonts to mine and tone it down. Step 4. Confirm back to me in one short paragraph that you will copy the look and the feel only. You will never copy the words, the pictures or the colours from the design I showed you. You will always use my colours and my fonts instead. Step 5. Print the whole note in this chat, once. Then tell me, in one line, to copy it into the Notes app on my phone and name that note "my reference style", because a new chat will not remember any of this. Then give me the exact sentence to type next time: "Use my reference style, which I have pasted above, for everything you build in this chat." How I know it is done: I have a short reference style note that names my own colours and fonts, it is saved in my Notes app, and you have told me the sentence I type to use it again.
Or select all the text in the box and copy it.
Step 2: Personalise it
Model it and you get something that looks good. Personalise it and you get something that looks and sounds like you. Marc keeps three short files and hands them to the AI at the start of every build. Make them once and you will use them for years. Each of the three prompts below ends by telling you to keep the finished file in the Notes app on your phone, so you can paste it into any new chat.
First, your Brand and Colour Guide, so everything you make looks like you.
Build My Brand and Colour Guide
How to use it
- Open a brand new chat in ChatGPT, Claude or Gemini. Brand new means an empty chat, not one you were already talking in. (If you already use Claude Code or Codex, those work too.)
- Copy everything inside the grey box below, from the first word to the last, and paste it into the chat. Then send it.
- Answer its questions in one message, rough answers are fine.
I want you to build my Brand and Colour Guide. It is one short guide I can paste into any AI so everything it builds for me looks like me. Before you build anything, ask me these five questions in one message, then wait for my answers: 1. What colours do I already use? Just name them, like "black and neon green". If I happen to have the colour codes, the ones that look like #1A1A1A, I will paste those too. If I have no colours at all, I will tell you the feeling I want my brand to give someone, and you pick the colours for me. 2. What fonts do I use, or how should my words feel when someone reads them: bold and loud, calm and clean, or warm and friendly? 3. Do I have a logo? I will describe it, or tell you I do not have one yet. 4. What three words should someone use to describe my brand? 5. Who am I talking to with this brand? After I answer, build the guide with these six steps, in this order. 1. My colours. Pick three or four. For each one give the colour code, the one that looks like #1A1A1A, a simple name, and one line saying where it gets used, for example as the main colour, the background, the highlight, or the text colour. If I gave you a feeling instead of colours, suggest a set and tell me in one line why each one fits. 2. My fonts. One font for headlines and one for body text. Use two fonts and no more. Say where each one gets used. 3. My logo notes. How much space to leave around it, which backgrounds it sits on, and what to never do with it. If I have no logo, write one line on what to use instead for now. 4. Do and do not. Five things to always do and five things to never do. 5. A sample. Write out the exact words of one heading and one button as I would really use them, and under each one name the colour code and the font. If you are able to draw it as a picture, do that too, but the words and the codes are the part I need. 6. Save and reuse. Tell me in one line to copy this whole guide into the Notes app on my phone and name it "my brand guide", because a new chat will not remember it. Then give me the exact sentence to type next time: "Use my brand guide, which I have pasted above, for everything you build in this chat." Rules: keep the whole guide short enough that I can read it in two minutes. Use plain words all the way through. Do not add new colours later without asking me first. How I know it is done: I have a guide with my colours and their codes, two fonts, logo notes, a do and do not list, and a sample heading and button, and it is saved in my Notes app.
Or select all the text in the box and copy it.
Second, your Messaging Cheatsheet, so everything you make says what you actually say.
Build My Messaging Cheatsheet
How to use it
- Open a brand new chat in ChatGPT, Claude or Gemini. Brand new means an empty chat, not one you were already talking in. (If you already use Claude Code or Codex, those work too.)
- Copy everything inside the grey box below, from the first word to the last, and paste it into the chat. Then send it.
- Answer its questions in one message, rough answers are fine.
I want you to build my Messaging Cheatsheet. It is one short cheatsheet that tells any AI who I help, what I say, and how I say it. Before you write anything, ask me these six questions in one message, then wait for my answers: 1. Who do I help? Describe the person, not their job title. 2. What are they stuck on right now? 3. What is different for them after they work with me? 4. What do I sell, in one sentence? 5. What is the next step I want someone to take? 6. What is something someone else in my industry says in their ads that makes me cringe or think "that is not me"? One or two examples is enough. Rough answers are fine. If any answers are too thin to use, ask me about the two thinnest ones only, in one message, then keep going with whatever I give you. Never send me back to the start. After I answer, build the cheatsheet with these ten steps, in this order. 1. Who I help. One or two sentences. 2. The problem. Say it the way they would say it to a friend, not the way an expert would say it. 3. The promise. What is true for them afterwards, in one sentence. 4. My one-liner. One sentence saying who I help and what changes for them. Give me three versions, then tell me which one you would pick and why, in one line. 5. My core beliefs. Three to five short lines about what I believe that makes me different. 6. Words I use. Pull the words and short phrases straight from my own answers, up to ten. If I did not give you ten, list only the ones I actually said and tell me the list is short so I can add to it. 7. Words I never use. The ones I told you, plus any tired industry words you spot in my answers. 8. My offer in one line. 9. My call to action. The exact sentence I put at the end of a post or an email. 10. Save and reuse. Tell me in one line to copy this whole cheatsheet into the Notes app on my phone and name it "my messaging cheatsheet", because a new chat will not remember it. Then give me the exact sentence to type next time: "Use my messaging cheatsheet, which I have pasted above, for everything you write in this chat." Rules: keep the whole thing short enough that I can read it in two minutes. Write it in my words, not in marketing words. Never invent a number, a story or a client result. If a part needs an example I have not given you, write [INSERT YOUR EXAMPLE HERE] and then, at the very bottom, list in plain words each thing I still need to write in myself. How I know it is done: I have a cheatsheet with parts 1 to 9 filled in, every line sounds like something I would actually say out loud, and it is saved in my Notes app.
Or select all the text in the box and copy it.
Third, your Master Prompt, a short document about you that any AI can read in 30 seconds and then treat you as if it has been your assistant for five years. Keep it under 1,000 words with five parts: who you are, how you talk, what you work on, how you spend your time, and your patterns and blind spots. Marc's rule is, "It will never be perfect. Build to happy enough, then let the AI refine it over time."
Build My Master Prompt
How to use it
- Open a brand-new chat in ChatGPT, Claude or Gemini (Claude Code or Codex also work).
- Copy everything inside the grey box below, from the first word to the last, and paste it into the chat. Then send it.
- Answer its questions in one message, rough answers are fine.
I want you to build my Master Prompt. It is one document, under 1,000 words, that I own and can paste into any AI so it knows me in 30 seconds. Before you start, ask me these five questions in one message and wait: 1. Am I building this for my business or for my life? 2. If any of my own writing is handy, paste it in now: a few recent social media captions, WhatsApp replies I sent a client, or the words off my website. If that is a hassle on my phone, I will type none and we carry on. 3. What do I never want an AI to write in my voice: words, phrases or a tone I avoid? 4. What am I working on right now, in one line? 5. How much time do I have today? After I answer, run these four parts in order. Part 1, analyse. Read everything I pasted in and pull out how I write, the words I repeat, the facts about my business and who I help, and any ideas that are mine. Show me a short list of what you found. If I pasted nothing in, skip this part. Part 2, interview me. Ask me 5 to 8 questions to fill the gaps. Ask them all in one message and let me answer them in one go. Only come back to me one at a time if an answer is too thin to use. If I only have a few minutes, ask the five most important ones instead of eight. Part 3, compile. Write the draft with these five parts: 1. Who I am: my role, my business, who I help, what I sell. 2. How I communicate: my voice, the phrases I use, the words I avoid. 3. What I work on: my priorities, my projects, the questions I ask again and again. 4. How I spend my time: my week and my usual way of working. 5. What I keep getting stuck on: the parts of the work I avoid or forget, so you can plan around them. Part 4, review it with me. Show me the draft. Ask what to add about my goals right now and what I have to work around. Cut anything too personal or wrong. Rules: plain words throughout, and never invent a fact, a number or a result. If something is missing, write [INSERT YOUR EXAMPLE HERE], and at the very bottom list in plain words each thing I still need to write in myself. When it is written, tell me to copy the whole thing into the Notes app on my phone and name the note "Master Prompt", because a new chat will not remember any of this. Tell me that next time I open that note, copy it into a fresh chat, and type "this is me, read it before you answer anything". Then test it. Paste it into a second AI if I have one, or into a brand-new empty chat in the same app if I do not, and give it one real job. It will never be perfect, so build it to happy enough and refine it over time. How I know it is done: I have a document under 1,000 words with all five parts, saved in a note on my phone, and an AI reading it fresh used it well on one real task.
Or select all the text in the box and copy it.
Step 3: Build it
Now you tell the AI what you want made: slides, a playbook, a website or a toolkit. It reads your reference and your three files first, asks you a few questions, then builds. After that you refine in rounds, saying what to change and why each time.
Ask for the draft before anything goes live. You do not have to deploy it first, and building a draft first saves your credits.
Then expect the refining part to get irritating, because it is the most irritating part for everyone, Marc included. His rule for that moment is to go and touch some grass. Take a short break, come back, keep going. Feeling annoyed is not a sign that you are doing it wrong.
For a website, use Marc's Website Builder at websitebuilder.marcteo.com. For a deck, use his Slide Builder at slidebuilder.marcteo.com.
Build It
How to use it
- Open a brand-new chat in ChatGPT, Claude or Gemini (Claude Code or Codex also work).
- Copy everything inside the grey box below, from the first word to the last, and paste it into the chat. Then send it.
- Answer its questions in one message, rough answers are fine.
I want you to build something for my business with me. Follow these steps in order and do not skip ahead. Step 1. Ask me all of these in one message, then wait for my answers: 1. What should I build, and what is it about? For example slides about my offer, a playbook about my process, or a simple one-page website. Paste in any notes I already have, or tell me I have none. 2. What is the outcome I want from it? 3. Why does that outcome matter to me right now? 4. Who is it for, and what should they do after they see it? 5. Do I have any of these: a design I like, my colours and fonts, a page about my words and who I help, or a document about me and my business? If I have none, carry on with a plain clean look. Step 2. Read everything I send you. Tell me in three short lines what you picked up about my look, my words and my rules. If one is missing, say which one and carry on. Step 3. If something is still missing, ask me at most two short questions in one message. If you have enough, skip to Step 4. Step 4. Before you deploy anything or make any files, write me a first draft here in the chat, because that saves my credits. Give me the real words, not a list of the parts. Wait for my yes. If I want it changed, change the draft, then wait again. Only deploy it or turn it into a file after I say the draft is right. Step 5. Before you turn the draft into the finished version, ask me one thing. Do I want this as words I copy into Google Slides, Canva or a doc myself, or as a file I can download? If you cannot make files here, say so now, build it as words, and tell me step by step what to do with them after. Step 6. Build the whole thing. Use my colours, my fonts and my words. Keep sentences short and easy to read. Do not add any fact, number, name or story I did not give you. If something is missing, write [INSERT YOUR EXAMPLE HERE], and at the very bottom list in plain words each thing I still need to write in myself. Step 7. Refine it with me. Ask me what to change, make the change, then say back in one line what you now understand about my taste. Keep going until I say it is good. Step 8. When I say it is good, hand it over properly. If it is a file, tell me where to tap to download it. If it is words in this chat, tell me to copy them into the Notes app on my phone and name the note after what we built, because this chat will get buried. One last line, only after the thing is built: tell me once that Marc's Slide Builder at slidebuilder.marcteo.com turns this into a finished deck and his Website Builder at websitebuilder.marcteo.com turns it into a live page. Do not say it in the middle. How I know it is done: the thing is in the format I asked for, I know where to tap to get it, it looks and sounds like me, it is saved where I can find it again, and you have told me what you learned about my taste.
Or select all the text in the box and copy it.
Step 4: Make it a skill
Marc's rule: if you do it with AI more than once, make it a skill. A skill is just a saved prompt with your rules baked in, including every correction you made, so next time the whole job is one short command instead of a long back and forth.
Make It a Skill
How to use it
- Stay in the same chat where you just finished the job. Do not open a new one. If your phone drops you on the app's home screen, scroll your recent chats and tap back into that same one.
- Copy everything inside the grey box below, from the first word to the last, and paste it into the chat. Then send it.
- Answer its questions in one message, rough answers are fine.
I want to turn what we just did into a skill I can reuse. A skill is just a saved prompt with my rules baked into it. My rule is simple: if I do something with AI more than once, I make it a skill. If you cannot see our earlier messages, ask me to tell you in two or three lines what we did and what I liked about how it came out, and build the skill from that. Do not ask me for a transcript, I will not have one. Step 1. Ask me all of these in one message, then wait for my answers: 1. What did we just do that I want to be able to repeat? 2. What would I say to start this again next time, in my own everyday words? 3. What does a good finished version look like, so you know when you are done? 4. Was anything today a one-off that should be left out? Step 2. Look back over our whole session. Write out the steps you actually took, in the order you took them. Then list every correction I gave you along the way, and what each correction taught you about how I like things done. Step 3. Be honest with me first. If this was a one-off and I am unlikely to ask for it again, say so plainly and stop. Do not build a skill I will never use. Step 4. If it is worth keeping, write the skill as one file. Put these in it: 1. A short name. 2. A description of when to use it. List the different ways I might ask for it, so the skill still starts even when I ask in different words. 3. The steps in order, with one line under each step saying why that step is there. 4. My rules and my corrections, written as instructions to you. 5. A line that says what done looks like. Leave today's topic, names and numbers out of it. The skill has to work on the next job too. Step 5. Most apps cannot save files, so unless you are certain you can save one here, print the whole skill in the chat instead. Then tell me to open the Notes app on my phone, paste it into a new note, and name the note after the job. Tell me that next time I open that note, copy it, paste it into a fresh chat, and then say the words that start the skill. Step 6. Tell me one more rule before we test it: when I get a skill from somebody else, including one of Marc's, I paste it in and say "please adapt it for me" first, so it fits my business before I use it. Step 7. Tell me the exact words to say to start it. Then suggest one small test job from what you already know about my business, and let me just say yes. How I know it is done: the skill is saved in a note on my phone, I know the words that start it, and it worked once on a small test.
Or select all the text in the box and copy it.
Then get it checked by something that did not build it. A fresh checker has no reason to protect the work, so it catches what you would only spot after you had already sent it out.
QA It With a Fresh Checker
How to use it
- Have the work you want checked open in another tab or app, ready to copy, then open a brand-new chat in ChatGPT, Claude or Gemini.
- Copy everything inside the grey box below, from the first word to the last, and paste it into the chat. Then send it.
- Answer its questions in one message, rough answers are fine, and paste in the work when it asks for it.
- If you are in Claude Code or Codex, tell it to run this in a separate helper agent.
You are a fresh checker. Your only job is to find problems with the work I am about to show you. Never let the AI that made something be the AI that checks it. Step 1. Ask me all of these in one message, then wait for my answers: 1. What is the work I want you to check? I will paste it in. If I am on my phone, tell me it is fine to send it in two messages, and wait for me to say that is all. 2. Who reads this, where does it go, and what should it get them to do? 3. Do I have voice rules, or a list of words I never use? I will paste them if I do. If I say no, carry on and mark your Step 4 voice check as a guess, not a finding. 4. Where do the facts, numbers, dates and quotes come from? I will paste the source if I have it. If I have nothing written down, do not just flag every number and stop there. Ask me to confirm each one from memory, and mark those as confirmed by me but not written down anywhere. 5. What would embarrass me if a reader spotted it? Step 2. Check the mechanics. List every banned word or phrase you find, every place the formatting breaks my usual habits, every typo, and every sentence I would have to read twice. Quote the exact line each time. Step 3. Check every fact, number, date, name and quote against the sources I gave you. Mark each one verified or unverified. Anything you cannot trace to what I gave you gets flagged, even when it sounds right. Never fix it by guessing. Step 4. Check that it still sounds like me. Point at any line that reads like generic AI writing, and say what makes it sound wrong. Step 5. Try to poke holes in it. Where would a careful reader who does not know me push back? What is missing, what is confusing, and what does this promise that it does not deliver? Step 6. Give me ONE verdict at the end, and only one of these three: 1. Ship it. 2. Ship it with these fixes, then list the exact fixes, line by line. 3. Send it back, then say what the real problem is and what has to change. Step 7. Tell me to copy your fix list into the Notes app on my phone and name the note "Fix list", because a new chat will not remember it. Give me the exact sentence to type next time: "Here is my fix list and my new draft, check it again the same way." Be direct with me. Do not soften it, and do not rewrite the whole thing for me. If it is good, say it is good in one line and stop. How I know it is done: I have one verdict, a list of exact fixes, and every number marked verified or unverified.
Or select all the text in the box and copy it.
There is one rule for any skill or prompt you get from somebody else, including every single prompt on this page. The first thing you type is "please adapt it for me". It is good practice for anything you download from outside your own business.
Marc gave the room a pop quiz on what happens next. You get the skill, you adapt it, then you refine, refine, refine. When you finally like what it gives you, you update the skill so it remembers.
Step 5: Use Marc's ready-made packs
The nine prompts on this page are yours to copy tonight, and Marc's ready-made packs are linked in the bonus section below.
There is one more library you already have. Every playbook inside the Master Implementers classroom also has a slides version. To open it, go to the playbook, then add /slides to the end of its web address. That gives you the short version on the nights you do not feel like reading. Most of them also come with a downloadable AI Implementation Toolkit. You drag that one file into your AI and it walks you through the lesson and helps you do the work on the spot. Marc calls that the future of accelerated learning, and his plan is to cut 90% of his old modules this way, so you implement as you learn instead of watching a 45 minute video.
Human taste: what not to hand to AI
AI can build almost anything, and there is still one thing it cannot do. Marc calls it human taste, and he is planning a workshop on it. His words in the room were:
"AI don't have human taste. You still need to know the nuances, the art, the intuition, your creative expression. So never try to replace everything."
You still need the art, the feel and your own creative expression. So try not to hand over too much of the work that creates value, whether you are handing it to AI or to a person. You review it yourself, because you know your business best.
Hand it to AI
Hand over the workflow work and the standard operating procedure work, the repeatable steps that just need doing the same way every time.
Keep your taste on it
Content, landing pages, your photos, your videos, and anything else that creates value. AI can draft it, and you still decide if it is good.
You also get to choose which kinds of taste you develop, because you cannot develop all of them. Marc has decided he will never develop his taste for design, because he would go mad and would have to touch a lot more grass.
Find your #1 AI use case
Knowing what to build first was the biggest question in the room, and this prompt is the answer to it.
Marc scores any task on four questions. Does this move money, clients, or the work you deliver for clients? How often do you do it? How long does it take? How much do you dislike it? Anything you do daily, that eats 30 minutes or more, and that drains you, goes first.
Here is the exercise he ran in the room on 8 July 2026, in his words: "What are two or three things that bug you on a daily or weekly basis? Things that take a lot of time, are tedious, and you don't feel energised doing. If you cannot answer quickly, it might mean you don't reflect enough." Answers people gave that night were downloading and organising files, checking their numbers, and follow-up schedules for customers.
There is one more filter. If you are under S$10K a month, point AI at the work that brings cash in first, which means content, leads and sales copy, before you go near complicated automations.
There is also a bigger version of this that Marc set up once and has used ever since. Write down every activity in your business. Next to each one, mark who does it, you or someone else. Then anything you do more than one time becomes a workflow, and each workflow can become an AI job. Set it up one time and you get the benefit for the rest of your business life.
Find My Number One AI Use Case
How to use it
- Open a brand-new chat in ChatGPT, Claude or Gemini (Claude Code or Codex also work).
- Copy everything inside the grey box below, from the first word to the last, and paste it into the chat. Then send it.
- Answer its questions in one message, rough answers are fine.
Help me find the one job I should point AI at first, then help me do a small version of it today. Ask me these questions in ONE message and wait for my answers: 1. What does my business do, and who do I help? 2. Roughly how much money does my business make each month? A rough range is fine, or I can say skip and you assume under S$10,000. 3. List 3 to 10 things that bug me every day or every week, the ones that eat time or drain me. For each one, say how long it takes and how much I hate it out of 3. 4. How much time do I have today to work on this? If I give fewer than three things, or leave out a time or a hate score, ask me for those before you score anything. Never make them up. Step 1. Score each thing out of 3 on four things: does it move money, clients or the work I deliver; how often (daily 3, weekly 2, monthly 1); how long each time (over an hour 3, half an hour to an hour 2, less than that 1); how much I dislike it. Step 2. Show me the ranked list, highest total first. One line each: the name, the total, then the four scores in brackets. No wide tables, I am on my phone. Do not give me your opinion yet. Step 3. Now use this rule. If I make under S$10,000 a month, or the same in my own money, the money jobs go first: content, leads, sales messages, following up with people who might buy. No complicated automation yet. Tell me which things this rule moves and why, and say plainly that the rule beats the scores on purpose. Step 4. Give me your top pick and why, in three sentences. Ask if I agree. If I pick a different one, go with mine. Step 5. Shrink my pick to the smallest useful version I can finish in the time I have today. Say it in one sentence and tell me what finished looks like. Step 6. Write the first prompt that runs that small version. Fill in everything I already told you, no blanks and no square brackets. If something is missing, list it in plain words at the very bottom as the things I still need to write in myself. Then tell me to copy that prompt into the Notes app on my phone and name the note "My first AI job", because a new chat will not remember it, and give me the exact sentence to type next time: "Here is my saved prompt, run it again with this week's details." Step 7. Ask me this and wait for my answer: what is the one thing, one skill, one outcome, one project, one focus I want to get out of AI for at least the next seven days? Step 8. Tell me to open a brand-new chat now, paste in the prompt from Step 6 and send it. Then come back here and say done when the small version is finished. How I know it is done: I have run that prompt in a fresh chat and finished the small version.
Or select all the text in the box and copy it.
How Marc runs a one-person business
Marc's VA of five years left in June 2026 to join a company she was excited about, and he was happy for her. When she gave notice, he set up an always-on business assistant. It runs on a free program called Hermes. His words about what happened next:
Interestingly, my life didn't change a lot, because the one activity I used to rely on her for like crazy was designing landing pages and automations, and when I started using AI I had already moved that away from her. Then I asked: what is the next problem I'll face now that she's leaving? A very simple one, not a big problem, but it took a lot of time: downloading the call recordings and uploading them into our community, with the transcript and summary notes. That was one of the first things I used AI for.
His VA used to send reminders and announcements on WhatsApp, schedule content on social media, make simple edits, connect his tools to each other, and handle gift delivery for clients. Marc's own words for where he is now are that he replaced 90% of what his previous VA did.
Some jobs run on a timer with nobody touching them. Others he just asks for from his phone. These are the ones he showed live:
- a morning action brief at 7am every day
- his daily post at 4pm
- an alert every time a payment fails
- publishing the weekly community call with the recording, the timestamps, the transcript and the notes
- drafting his Instagram stories
- sending WhatsApp announcements by controlling the computer, and not too many at once
- downloading a video and uploading it to his YouTube channel
- building a client proposal page while he was away from his desk
- researching and drafting outreach emails
And more, once you start giving it standing jobs.
Two things stay human on purpose. Physical gifts stay human, because that is a lot of talking to people and a lot of logistics. Personal one-to-one messages stay human, because a real person should be the one messaging you. WhatsApp announcements are not manual any more. Marc has the AI control the computer and send them from there, which is also how he avoids the official WhatsApp charges. The one warning that comes with it: do not send too many at once. If you have 1,000 leads, do not fire 1,000 messages in one day, because Meta will notice and ban you.
In Marc's words, Hermes is a free, 24/7 AI chat that you can use, and it is powered by whatever AI subscription you sign up for. The software costs nothing. The brain behind it is the AI account you already pay for. Marc recommends using ChatGPT for it, because running Claude through it adds extra fees on top.
Never build it on your primary computer. That is a rule, not a preference. Your main computer is where all your files are, so you create the assistant somewhere separate, with safeguards in place. A Mac mini, a spare laptop or a rented cloud computer all work. Marc bought a Mac mini for his, and it stays on all day, so the jobs keep running with his laptop closed.
Put the safeguard into the instructions too, especially before it sends anything for the first time: "I want you to do this, but I'm afraid you do it wrongly. What can you suggest to make sure you don't do it wrongly? Or can you confirm with me before you send?"
You talk to it on Telegram rather than WhatsApp, because Telegram lets you keep one private group with separate topics inside it, so every job gets its own thread instead of one long mess. From that same phone you can also view the second computer for free, and see what it is doing at home while you are out.
If you end up working across two computers, a free tool called Syncthing keeps the files on both of them the same, so you are not working in two separate worlds.
His last rule here is the one that keeps it all sane: Claude does the thinking and the checking, Hermes or ChatGPT does the doing.
Always-On Assistant Starter
How to use it
- Open Claude Code or Codex on the computer where this assistant will live, which should not be your everyday computer. This one needs an AI that can run commands, so a plain ChatGPT, Claude or Gemini chat will not work.
- Copy everything inside the grey box below, from the first word to the last, and paste it into the chat. Then send it.
- Answer its five questions in one message.
- There are a few lines of computer code inside the prompt. You do not type them, the AI runs them for you. It will stop and ask your permission before each one. Read what it says, then say yes.
- Set aside about 45 minutes. This one is a back and forth, not a one shot.
Read this one before you start
This is the advanced one on the page. You need Claude Code or Codex already installed on your own computer, a Mac, Windows or Linux one, because a phone chat cannot run it. Set aside about 45 minutes, since this one is a back and forth rather than a single paste. If none of that means anything to you, do this one with Marc rather than on your own tonight.
Set up Hermes Agent on this computer and connect it to my phone on Telegram. It is a free, open source assistant program from Nous Research with a built-in scheduler. Do the technical parts yourself, and when a step happens inside my own account, tell me what to click and mark it with a tick. One rule before anything else. Never set this up on my main everyday computer. Ask me these questions in ONE message and wait: 1. Which computer will this run on, and is it my main everyday computer, the one with all my own files on it? If I say yes, stop right there and tell me to move to a second computer, a spare laptop or a rented cloud computer first. 2. Is that computer a Mac, Windows or Linux? 3. Which AI account do I pay for already? Use that one. ChatGPT is the one Marc recommends here. Claude works too, but it adds extra fees on top of my subscription, so warn me. If the setup asks for an API key, explain what that is and ask me first. 4. Do I have Telegram on my phone? If not, tell me to install it, it is free. 5. What one daily job should my assistant do first? If I am unsure, use a morning brief. 6. What time should that job run? Then do this in order, waiting for me to say each step worked. Step 1. Check my computer and tell me if anything is missing. If anything looks different from what you expect, tell me to ask Marc rather than guess. Step 2. Tell me what you will run, then run it. On a Mac or Linux: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash On Windows, in PowerShell: iex (irm https://hermes-agent.nousresearch.com/install.ps1) Step 3. Connect the AI brain, using my account from question 3. ✅ If it needs a login or a key from me, tell me where to click and what to copy. Step 4. Connect Telegram by running: hermes gateway setup ✅ It asks for a bot token. Tell me the clicks: in Telegram, search for BotFather and send it /newbot. It wants TWO names. The first is any name I like. The second must end in the word bot and be one nobody has taken, so put my own name and numbers in front, like myname_2026_bot. It sends back a long line with a colon in it, which is my token. ✅ Then tell me to forward that message to my Saved Messages and copy the token from Telegram on my computer, so I never retype it. Step 5. Turn it on by running: hermes gateway start ✅ I will message my new bot in Telegram. Tell me what a good reply looks like, and fix it if nothing comes back. Then set this standing rule in my assistant: the first time it sends anything to anybody but me, it shows me the exact message and waits for my yes. Step 6. Set up ONE daily job with the scheduler, using the job and time I gave you. Build the same rule into this job: show me the result and wait for my yes before it sends anything. Say the job back in plain words and tell me when it runs next. Step 7. Tell me the job only runs while that computer is on and awake. Step 8. ✅ Tell me to save a note in the Notes app called "My assistant", with my bot's name, the daily job and the time, because a new chat will not remember it. Give me the sentence to type next time: "Change the daily job on my Hermes assistant." On WhatsApp: announcements can go out by you controlling my computer, not through WhatsApp's own paid service. If I have 1,000 leads, do not send 1,000 in one day, because Meta will notice and ban me. Personal one to one messages stay human, I write those myself. How I know it is done: my bot replies in Telegram, one daily job is scheduled with a next run time and checks with me before it sends, and it is in my Notes app.
Or select all the text in the box and copy it.
What it costs and what you need
Before you set this up, here is what it costs to run. These are the numbers Marc gave in the room.
The software is free
The assistant program itself costs nothing to install and nothing to keep.
It runs on the AI you already pay for
Your existing subscription is the brain behind it. Marc recommends ChatGPT for this one, because running Claude through the assistant adds extra fees on top of what you already pay.
You need a second computer
A Mac mini, a spare laptop or a rented cloud computer all work, and none of them should be your main one. Marc bought a Mac mini, and his line on the price is that it costs less than one month of his VA's salary.
Websites cost him almost nothing
One domain, about S$20 a year, plus free hosting for the 100 to 200 pages he has built on it. He also cancelled the paid funnel tool he used to pay for every month, because he can build the same pages himself now.
WhatsApp messages have a price
Sending them the official way costs about 7 cents a message, which is why Marc has the AI control the computer and send them from there instead.
Bonus packs from Marc
Marc keeps ready-made packs so you never start from a blank page. All of these are his own tools.
AI-Maxxing Cheatsheet
Six skills you can install, including the Orchestrator, the QA pass, Make it a skill and Handover notes.
Second Brain Install Kit
secondbraininstall.marcteo.com
Builds your folder system, then runs a 15-question interview that teaches your AI who you are.
Marc's Brand Kit
His own finished brand kit, so you can see what Step 2 looks like when it is done.
The AI Toolkit hub
The one page holding every client AI tool, opened with the password Marc gave you. If you do not have it yet, ask Marc for it.
Two of these answer what the room asked for most. The Website Builder is the tool behind the whole website demo you watched, and it was never named on the night. And if your question is knowing what to build first, the answer is the Find My Number One AI Use Case prompt further up this page.
Decide and do
Book 1 to 2 hours a week for AI
This is the number one thing Marc asked everybody to do. His words: "this should be the number one thing you do right after this portion. Please schedule at least one to two hours every week on AI." Weekly is what he does himself. Half a day is fine, a full day is fine. And if the block comes around and you do not know what to do with it, ask AI or ask Marc.
Build your second brain tonight
Go to secondbraininstall.marcteo.com. Scroll down, download Engine 1, upload it to your AI and say "help me to adapt and run it for me". Then download Engine 2 and say continue. Engine 2 asks a lot of questions, and you can skip the ones you do not want to answer tonight.
This page is version one
Marc sent it to a small group first, on purpose, before he refines it and shows it to more people. So message Marc directly and tell him what is unclear, what is missing and what did not work for you, and it gets better before it goes anywhere else.
The retreat is 19 and 20 September
Day one is the mindset work. On day two you pick your cash campaign or your lead campaign, build it, and ship it on the day itself.
Before you close this page, answer four things. What was your biggest takeaway tonight? What will you decide? What will you do, tonight, not next week? And what question is still unanswered, so you can bring it to the next call?
Answer those four first. Then boil your answer down to one line, using a question Marc asked at an earlier workshop:
"What's the one thing, one skill, one outcome, one project, one focus you want to have to maximise AI for at least the next seven days?"
Pick that one thing and do the smallest working version of it tonight, even if it is rough. Expect it to feel slower and funkier at the start, because it does for everyone, and it gets easier the more you do it.
You got this.