The AI journey.
In the last six months my robots built me a software platform, a brain that can find anything I have saved in forty years, and a morning brief that replaced pretty much two hours of reading every morning. I am writing this on Sunday, November 1, 2026, and posting it on the 7th. Keep that date in mind, because at the speed it is going, everything I am about to describe will look primitive in weeks, or months. The tools I use today will soon be the stone age of AI, the way the first cell phones look to us now, bricks with an antenna. So this is not a piece about where AI is going. Plenty of people write those, and most of them will age badly. This is the story of how one person got from a computer plugged into a television to a team of robots that run his day, and what changed along the way. The technology will date. The journey will not.
The machines that raised me.
I was a kid in Lebanon in the mid eighties when the first personal computers reached the stores, and my parents bought one for the three of us: me, my brother and my sister. A Commodore 64. Two joysticks, a cassette reader for the software, and the machine itself, which was the computer and the keyboard in one piece. You plugged it into the TV. We were the coolest.
The games came on cassettes, the same kind you used for music, and we would copy them from one to another with a radio recorder to exchange between friends. I still remember the robotic sound a program made when we tried to play one on a regular cassette player.
After playing around with the games that came with it, I found a program in the manual that drew a hot air balloon on the screen and made it float across the TV on its own. You typed the code in line by line. That was the first programming I ever did. A balloon, in big pixels, drifting across a television in Lebanon. I have never stopped since.
I was always intrigued by technology, and I always had the newest gadgets. The calculator watch my uncle gifted me when I was nine or ten. The first pager, in 1995, while I was still in high school, before all my friends. Then the first cell phone, and I ended up starting a business selling them. Most of my friends bought their first pagers and cell phones from me. Then I built a website for the business, with a phone and plan comparison sheet, one of the first in my line of business, when very few businesses had one. Then the Palm Pilot. The laptop. The first BlackBerry when it came out. The first iPad. If it was new and it did something cool, I had to have it.
That is the whole pattern: I always followed the tech. And it is the reason this piece exists. I came to AI the way I came to everything else: early, curious, and with my hands on it.
Studying it, not waiting.
For three or four years I followed everything about AI, including reading a bunch of daily AI newsletters every morning, and every time a new tool came out, I installed it to try it out: ChatGPT, Claude, Gemini, Perplexity, DeepSeek, Kimi, Grok, and a dozen more you probably never heard of. A few days ago, I added Grok Bot, xAI's new agent, to the team. Still to be tested. Most people met AI through the answer that now sits at the top of a Google search. I was also using ChatGPT for small research, as a lab sidekick, a researcher, Google on steroids: you ask, it answers. Ask it to dig deeper and it comes back with a full breakdown.
But I did not go all in yet. I was studying it, and I knew the tools were not ready. The problem was memory. You could have the smartest conversation of your life with it, close the window, and it would forget you existed. An assistant that forgets you every morning is not an assistant. So I waited, and I watched, and every morning I read.
Going all in.
When ChatGPT came out and people started using it, I would always interview my friends on how they used it in their day-to-day and in their business, and every one of them was already using it in a completely different way: one drafted emails with it, one analyzed reports, one planned trips. I would also ask whether they paid for it. At the beginning, paying did not get you much more, and I kept asking myself whether it would make me use it more.
What tipped me was ChatGPT shipping memory, then projects, then agent mode. By the summer of 2025 it had all three. Memory meant it would remember me tomorrow. A project meant I could keep everything about one piece of work in one place. Agent mode meant it could go and do things instead of only answering. The thing I had been waiting for, a tool that remembers you and acts for you, had arrived. Agent mode shipped on July 17, 2025. I upgraded on July 31.
That was it. Time to go all in. I set up my first project that same day, and then a bunch more, and I started doing real work inside it.
The $100,000 app.
I had an idea for a platform, a tool, and I wanted a portal and an app for it. A couple of friends who have experience building apps told me what I wanted would cost at least a hundred thousand dollars for a basic version. My answer was no, there must be a way, using AI.
ChatGPT also pushed back at the beginning and suggested a working prototype, a version without the portal and the app: Google Workspace, with Drive documents and a shared spreadsheet. I kept pushing for the app, and it kept telling me it was very complicated. So I asked it to show me how complicated. It explained the steps and the tools I would need: Vercel, Supabase, GitHub, VS Code. It was all Chinese to me. I did not know what any of these were.
At this time I still had the free version of Claude, which I barely used, and I was getting more and more curious, because of all the hype I kept reading about it. In early April 2026 I gave ChatGPT's plan to Claude and asked: can you help me build this? It said yes. I said how. It said: I will guide you step by step. And it did. I got locked out of the free plan pretty fast. The next day I upgraded my plan, hit the limit again, kept having to wait for the resets, lost patience, and started topping up, five dollars at a time, then ten, then twenty. It felt like online gambling. I spent over a hundred and fifty dollars in tokens in one week. The week after I moved to the Max plan. I still get locked out from time to time, when I have too many things going at once. I realized fast that Claude is the Apple of AI: you pay a premium and get the edge.
And this is when I understood what vibe coding is: you describe what you want, the machine writes the code, you test it, you describe what is wrong, and you keep going. I built myself a workflow for it. I use Cowork, Claude's desktop agent, as my sidekick, writing the prompts for Claude Code, the coding tool, and the two of them run in a loop. I stopped using the regular chat after less than a week, and I have not used it since. I came up with my own word for all this, Clauding, like coding, but with Claude. I still only half understand how a lot of the pieces fit together. Supabase is the database, Vercel is where it runs, GitHub is where the code lives, and that is about as deep as my knowledge goes. It did not matter.
What came out is not merely an app. It is a full-on operating system: a backend, a frontend, a portal, and an agent that manages it, the whole thing. A mastery-through-discipline platform, with an interactive phone app. I am using it myself and testing it with a couple of friends. I will talk more about it, and show more of it, at some point. For now, the point is simpler. A hundred thousand dollars was what I was told it would cost, for a basic one, and mine did not end up being a basic one. It cost me about three hundred dollars, and some time learning and vibe coding.
From doer to director.
I call them robots. I have a bunch of agents doing things for me, and that is what it feels like: a small staff, none of whom need a desk, a break or a dental plan. Twenty-four hours, no rest, no sick days.
Here is a day from last week. I asked one agent to sweep every screenshot on my phone from the last six months and pull out everything about AI for this piece, then search the web for sources behind a dozen claims. At the same time, another one was building me a website I had been meaning to make: it dug through my old files and folders, some of them dating from the end of the nineties, pulled out the text, the pictures and the documents, and I talked it through the rest. In about three hours it was pretty much built. Another was going through my email for one of my businesses, analyzing the bookkeeping folders to pull out numbers and drafting a reply to a supplier, and reporting back. I said yes, no, add this, remove that, send. All of it, plus a bunch of other things, ran at the same time. I was just switching tabs to reply and direct.
The same thing happens with this publication. I give it ideas for subjects I would want to write about, and it files them into a master document. I forward it the articles I find interesting, so it can go deeper on them. When I am ready to start the next piece, it shows me the list and I choose. It pulls everything I archived, scans my folders and interviews me. I talk, it researches and fact-checks, it builds a draft, and I edit. The image, the layout, the tags: it builds all of it, and I direct and approve.
My mornings changed the same way. I used to read ten to twenty newsletters every day. Clean up, delete the irrelevant, scan everything, read the interesting ones, screenshot the important things, flag the emails to come back to. Two hours, every morning. So I created a new email address and forwarded all the newsletters to it, and an agent reads them there. Now my inbox gets three newsletters a morning: the two I chose to keep because I enjoy reading the full articles, The New York Times and Monocle, plus the brief the agent writes for me. It reads every newsletter, sorts them, shows me what matters, sends it to a portal I built where I read and sort, and gives me buttons: put it in my loops, come back later, go deeper, act on it, file it. Two hours became about twenty minutes. And I keep optimizing it, and it keeps getting better and smarter.
It was not always like this. I was always tech-savvy: I played around with code, found ways to solve things myself, and learned fast, but never anything close to this. When I built the online store for my flip flop brand a few years back, I used freelancers to customize my Shopify store, and the back-and-forth with designers was a full-time job and a headache. So I slowly learned to fix things myself. Then, when Shopify 2.0 came out and some apps stopped working without it, I had to rebuild the whole site from scratch. After getting a few quotes I decided to try to do it myself. I estimated it would take me a couple of weeks. I sat down one afternoon and built it in less than 24 hours: hours and hours of coding, trials and tests, in flow mode, the way my brain works best. That was the old way. Now I just ask, with clear instructions, and my robots build, fix, adjust and write while I move on to the next thing.
That shift has a name. I took it from Dan Martell, who wrote Buy Back Your Time and teaches business owners how to do exactly that: you go from being the doer to being the director. In his telling, only about eight percent of your work is something only you can do; the other ninety-two is what you hand over. He describes three levels of using AI: the assistant, where it helps you do the work faster; the agent operator, where it does the work and you manage it; and the organization, where it runs whole workflows and you set the direction. Martell says only 0.3 percent of AI users have reached the second level, and 0.05 percent the third. I am probably somewhere in the second level, working my way through the third.
My eight percent is the big decisions, creativity, structure, direction and copywriting. And checking everything, because the machines still get things wrong, and the person whose name is on the work is still me.
I talk and it writes.
If you take one thing from this piece, take this one. You have probably heard that the most important thing in AI is the prompt. Everybody says it, and they are right: the prompt is everything. The more detail you give the machine, the better the work comes back. But typing a detailed prompt is slow and painful, and going into detail is where most people stop. The biggest secret of how I got so deep into AI so fast is speech-to-text. I talk, fast.
Our brain moves faster than our typing. When I speak, I do not lose my thought. I can give the machine three paragraphs of context, what I want, what I do not want, and what good looks like, in the time it would take me to type a title. The tool I use is Wispr Flow. It sits on top of every app, I hold a key and talk, and polished text appears wherever my cursor is. Since I started I have dictated over a million words, and in October the app put me in the top 0.1 percent of its users, at 145 words a minute. That is how much prompting it takes to run a life through machines. Andrej Karpathy, one of the most respected minds in the field, a founding member of OpenAI who went on to run AI at Tesla, put it in one line back in 2023: the hottest new programming language is English.
This publication exists because of it, and so does everything else I do with AI. I talk, and my AI writes. I would never have dreamed of writing and publishing anything without a tool like that. It is like working with a ghostwriter, and ghostwriters have been around for centuries. You have probably read books written that way: Donald Trump's The Art of the Deal, Prince Harry's Spare, Andre Agassi's Open. The ghostwriter does not invent the story. It comes out of hours of interviews, and the writing goes back and forth until it sounds like the person on the cover. That is exactly how this works, except my ghostwriter never sleeps.
Even the voice you hear on the audio version is AI, cloned from my own recorded voice. That one was not easy. It took two months and several tools to get it this far, every improvement seemed to make it worse before it got better, and it is still not perfect. But it is my voice, reading my words, without me spending an hour in front of a microphone every Sunday.
AI is a race, so no single tool.
I am a one brand person. When I find a brand I like, clothing, tools, whatever it is, I stick to it and stay loyal, and I like to stay in one ecosystem. For example, I use Apple for everything, because I want one system that just works and keeps it all synced in one place. With AI I had to give that up. AI is a race. Every few weeks one company jumps ahead, then another. You cannot rely on one tool, and you cannot count on one tool to hold all your work. The system and the brain I built, which I will get to in a moment, are what made that possible.
Before I get to it, it helps to know what the tools are today, because most people stop at the first one. There is the chat, where you ask and it answers, and where almost everyone lives. There is the agent: you give it access to your files, folders and tools, and it goes and does the work and comes back, a bit like a personal assistant. I call mine my chief of staff, because I can have him spin up a bunch of sub-agents that run different errands and report back. And there is the coding tool, which builds software from a description. The main players have most or all three of these, and depending on the latest release one is better than the other, and the lead changes hands all the time. If you had asked me last month, I would have said Claude, no question. Today I think I am in love with ChatGPT, amazed at how powerful it suddenly became. Thankfully they do not have emotions, yet. Claude is writing this piece. The next milestone is moving between them without losing your context and what you already built. I think I am there. That bridge is what I built next: my second brain.
I named it Noa. One folder that every AI can read, with a few files at the center: one that describes who I am and how I like things done, one that holds every decision I have made, one that holds the skills, the ways of doing things I taught it. Any model can join, including the ones that run locally on my laptop, not on the cloud. It took more attempts than I want to count. I started in the terminal, typing commands I did not understand, then pasting commands the assistant wrote for me. Now I give it a one line pointer to a file, it does the work, and it writes what it did back into the same file for the next one to read. Along the way I tried a dozen tools you have probably never heard of, mind maps, note vaults, local servers and .md files. I kept what worked and replaced what needed replacing as tools evolved. The brain is also fed by local agents that scan my hard drives, my photos and my screenshots, over 200,000 images on my phone alone, plus the albums on my hard drives going back to the start of digital photography, and folders of folders of files and archives, so anything I saved over a lifetime can be found with one command.
Memory, the problem that kept me from going all in, is what the brain is meant to help solve, and it is still not fully solved. But it is better and better, and with the system and the flow I have, I can see when it breaks and fix it. It is not perfect yet. It still forgets things and it still makes things up now and then. It is usable, and it is far from a waste of time. As the tool evolves on a daily basis, I keep optimizing the memory and the brain, and that is part of the experiment.
Ten times more things, and my time is still not saved.
Here is the honest part. My days are ten times more productive, and my time is still not being saved. The day is as full as it ever was. It is just full of bigger things. I end the day with much more accomplished: things that used to take weeks now take hours, and things that took hours sometimes take minutes.
I wrote about this in Experiment 004, About Time, and it was true before AI. In 1930 John Maynard Keynes predicted our grandchildren would work fifteen hours a week. We got the productivity. We did not get the free time. A 2026 ActivTrak study that tracked more than ten thousand workers before and after they got AI found that no category of work went down, and time spent in email more than doubled.
I think that is what we are going to do with AI too. We will have more time, and we will find other things to spend it on. Hopefully to learn more, explore more, experience life more. Now I get much more done, and things keep getting built and tested in the background while I keep on learning and optimizing.
A match made in heaven.
AI and the ADHD brain were made for each other. The ADHD brain thinks outside the box, jumps between ideas, and loses the thread while it is writing the first one down. A machine that catches every thread, holds it, and brings it back when you ask is the thing that brain always needed. And going from doer to director is a dopamine machine: you say it, it happens, you say the next thing, and the reward arrives before the attention runs out. Randy Kulman, a child clinical psychologist who writes about ADHD, put it this way in Psychology Today: AI "might be best viewed as an agent that supports ineffective executive functioning skills." Executive function is the part of the brain that starts a task, holds the plan and finishes. It is the part ADHD makes unreliable, and it is the part my robots now hold for me.
I gave my AI a standing instruction early on, to ask me and answer me in a way that fits how my brain works: short answers, the point first, numbered lists I can answer by number, and a recap at the end of what it needs from me, so I keep going instead of stalling. My own ADHD story is its own Experiment, and it is coming.
No two people use it the same way.
After a few years of going deep into it, most of my admin is now automated. And a lot of the people I still ask are using it in a basic way: questions and answers. They never went deep enough, past the complexity, to experiment. I understand. It is complicated. Every day there are new tools, skills, connectors, updates and new terms, and it takes some time to really understand each one. If I skip a day or two of AI news, I feel behind, like I missed something important, given how fast it is all moving.
It is a tool, like a car or a phone. AI is not one solution. Every person uses it differently, depending on what they do, what they have going on in their life, how their brain works, and where the pain points are. Medical advice. A business plan. A poster for an event. A birthday plan. A recipe from what is in the fridge. Planning holidays. Researching something and turning it into a PDF. It can help you with your day-to-day business, your legal questions, your accounting, your taxes. And all of that is with the chat alone. Imagine what you can do with the agent.
Everything I am not writing about today.
This piece is about the tool and my journey mastering it. There is a lot about AI I am not covering here, and you have most probably been reading about it: the jobs it will take and the ones it will not, whether the machine is honest, AGI and the race with China, the renaming to superintelligence, quantum computing, the electricity bill behind every answer, and what it does to your brain. Each of those is bigger than one essay, and I will come back to some of them in upcoming Lab Notes and Notes, without going into all of it here.
And one more thing, the one I tell my friends about around dinner tables whenever the conversation turns to AI. Everybody is obsessed with AI, and nobody is following what AI robotics is doing. It is being built in the background, and it is going to be the real game changer, the thing that touches the physical world. Do not mix it up with the robots that are currently on factory floors; those are nothing new. Folding intelligence into them is what takes it to the next level, and the training is happening right now, with cameras in factories watching every move. That is going to be the big explosion. I always joke that the coolest one I have seen so far is a laser that shoots mosquitoes out of the air. I read about it over a month ago in one of my newsletters. This week it made the front page of The New York Times, so I guess I am not the only one excited about it. Robotics gets its own Experiment.
The sky is not even the limit anymore.
And all of this is only the tip of the iceberg. Here is the one thing to do this week. Write down every task you did in the last seven days. Pick one that a machine could do, and hand it to an agent. Not a question. A job. If you want a starting point, here are two. Point it at the screenshots on your phone and ask it to pull out everything you saved to read later, sorted by subject. Or point it at your messiest folder and ask it to rename and organize every file by date. When it comes back done, you have just hired your first robot.
In Experiment 003, It Was Never the Gold, I quoted Marc Andreessen: AI is our alchemy, our Philosopher's Stone. This piece is what that alchemy looks like in my day to day. The tools will keep changing. I will keep up, because I am still the same kid who coded a flying balloon onto a TV screen in the mid eighties. Only now I have robots, building in minutes what would have taken me months a year ago.
Experiment 010 is complete. You just read it.
Georges
Usually one Experiment on Sunday. Lab Notes in between. Free.
If you enjoyed it, leave me a like or a comment.