08:07Brad Bonanno | AI Automation · Demo
How I Watch ANY Video with AI in Seconds
A free agent skill sends any video straight to Gemini's API, so your AI gets back timestamped answers instead of a wall of screenshots.
October 2ndThe argument in one line.
A useful AI setup is one folder you own with four parts working together: context files, skills, connectors and an AGENTS.md map, not a growing pile of tools.
Read if. Skip if.
READ IF YOU ARE…
- A founder or solo operator who keeps re-explaining their business to ChatGPT or Claude at the start of every new chat.
- Someone with scattered skills, projects and tools who feels like none of their AI setup fits together.
- A Claude Code or Codex user who wants one shared setup that survives switching to a better model.
SKIP IF…
- You want an autonomous multi-agent swarm or a Jarvis-style voice assistant; this is deliberately the opposite.
- You have no repeatable work yet to build skills and context around.
The full version, fast.
Most people get poor results from AI because they pile on tools, tutorials and setups that never fit together. Brad Bonanno replaces that with one OS folder built from four parts. Memory is four context files (about, customer, offer, voice) generated from your chat export. Methods are skills saved with Skill Creator. Means are MCP connectors to tools like Gmail and your calendar. The Map is an AGENTS.md telling the agent where everything lives. Each job gets its own chat, with handoff.md notes saved in the folder.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 02:32
01 · Why complex fails, simple scales
The Jarvis fantasy, the author's stripped-down setup, and the 'AI minimalism' premise.
02:32 – 05:59
02 · Memory: context files you own
Four files (about, customer, offer, voice) generated from a ChatGPT or Claude export, then reviewed by hand.
05:59 – 08:35
03 · Methods: skills
Skills as bookshelf procedures, built with Skill Creator and shared across agents via symlinks.
08:35 – 10:29
04 · Means: connectors
MCP connectors for Gmail, calendar, CRM and research tools, and checking each one's permissions.
10:29 – 12:00
05 · Map: AGENTS.md
The instruction file that tells the agent where things live, shared across Claude Code and Codex.
12:00 – 13:53
06 · Using your folder day to day
Project folders, one chat per job, handoff.md, and portability to new models.
Lines worth screenshotting.
- Complex AI setups fail and simple ones scale: the author stripped his setup down until it broke, then added back only what real work required.
- The four parts of a working AI setup are Memory, Methods, Means and Map, and if any one is missing the setup stalls.
- Your ChatGPT or Claude chat export already contains your business context, so let the AI mine it into context files instead of writing them from scratch.
- Build a voice profile only from what you actually published, otherwise the AI studies its own drafts and produces a bad voice file.
- Less volume and higher quality: read every context file yourself, because one wrong fact poisons every future output.
- If the task repeats, it belongs in a skill; if the information changes, it belongs in memory.
- Create a skill the moment you get a good result after back-and-forth, by asking Skill Creator to turn the conversation into a reusable skill.
- Skills act like a bookshelf: the agent sees names and descriptions and only reads full instructions when a task needs them.
- Store skills once in a .agents folder and symlink them into .claude so both Claude Code and Codex use a single copy.
- A connector is a new capability for your AI, but check what it can do, because reading an email and drafting a reply are very different permissions.
- Save the map as AGENTS.md, which Claude Code now supports directly, so every agent shares the same instructions.
- Give each job its own chat and keep the useful output in files, and a handoff.md lets the next chat pick up where you stopped.
- Because you own the folder, a better model means reconnecting accounts, not rebuilding your setup.
Four parts that make AI actually useful
WHAT TO LEARN
A simple folder with memory, methods, means and a map beats a complicated stack, because each part fixes a different reason AI gives generic results.
- Export your ChatGPT or Claude history and have AI draft about, customer, offer and voice files, then read and correct them yourself before saving.
- Build the voice file from writing you actually published, not from AI drafts, or the profile will mimic the model instead of you.
- Turn a good back-and-forth result into a skill with Skill Creator, and keep changing information like pricing in context files rather than inside skills.
- Store skills in one .agents folder and symlink them for Claude Code, so Claude Code and Codex share a single copy.
- Add connectors only for tools you already copy information out of, and check whether each one can read, draft or send.
- Write an AGENTS.md map saying where context, skills and projects live, and ask the AI to update it whenever the setup changes.
- Open one chat per job from the OS folder and end unfinished work with a handoff.md so the next session starts informed.
Terms worth knowing.
- Context files
- Markdown files such as about.md, customer.md, offer.md and voice.md that hold what your AI needs to know about you and your business.
- Skill
- A folder with a SKILL.md file holding steps, examples and a quality check for a task you want done the same way every time.
- MCP
- Model Context Protocol, a standard way for an AI agent to connect to outside tools like Gmail, calendars and CRMs.
- AGENTS.md
- An instruction file at the top of your folder telling AI agents where context, skills and projects live and when to use them.
- Symlink
- A pointer to an original folder, so two tools can use the same files while you maintain only one copy.
- handoff.md
- A short note listing what was decided and what remains, saved so the next chat on that project can continue.
Things they pointed at.
06:00toolSkill Creator
09:20toolModel Context Protocol (MCP)
10:40toolAGENTS.md
Lines you could clip.
01:45
“Complex fails, simple scales.”
01:20
“The gap in front of you is tiny compared to the gap behind you.”
05:40
“If the task repeats, the process belongs in a skill. If it's information about your business, that belongs in memory.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
Hey Jarvis, wake up. Organized your calendar. Invoicing for three and a half grand.
About 230 pounds a month. Ringing the restaurant now, sir. Understood, sir.
Do a billion dollar company's hand. And become Tony Stark. Why can't AI just be easy?
And what if the reason that you're not getting results with it isn't that you're not doing enough stuff, but the opposite. You're doing too much, watching too many tutorials, too many setups and another tool every week. You've got all of these pieces and none of them fit together.
Or while you're watching some guy build Jarvis that calls him sir in an English accent. So today I'll replace this massive AI to -do list with one simple folder that's transformed the way that I work as a founder. And the best part of this system is...
that will get rid of 95 % of the stuff that you thought you had to do. Let's get into it. When I started building my business around AI, I started with a super complicated setup.
Thousands of files, pretty dashboards, the whole thing. It looked great, but then I realized that I was spending more time actually working on it than I was doing work that made me money. And if I had a peek at your setup, I reckon I'd see some of the same stuff.
Skills you've never used, disconnected projects and hours spent explaining the same things over and over. Then a new model comes out and you start the whole setup again in that new provider. And when you go looking for help, you come back with another thing to install.
So what do you do? you throw it all out that's what i did i stripped my setup down until it broke then i only added things back when i had work that needed them and after going through all of that the biggest thing that I've learned is that complex fails, simple scales.
Everyone in the middle right now is trying to build the biggest Jarvis confusing spectacle with results. And right out here, there's some 20 year old AI genius running a thousand agent swarm. The bad news is that you're never catching up to those people.
But the good news is you don't need to because the gap in front of you is tiny compared to the gap behind you. And to get here, you need something much simpler, AI minimalism. For me, that's one folder.
Let me show you. These are the four parts of building a useful AI setup. I call them the four M's of AI minimalism.
They're building blocks. And if one of them is missing, your setup won't work. That's why you can have the best model, a hundred skills and a folder full of context and still spend your entire day going back and forth with AI and getting nothing done.
You need all four of these working together. Memory is what AI knows about you. Methods are the processes it follows.
Means are your connections and tools. And then the map tells it where to find things and how to use the folder. So when something isn't working, you can work out which part is missing and fix that instead of restarting your entire setup.
Ingredient number one is memory. And the easiest way to understand memory is to think about someone that's worked for you for five years. They know your business, your customers, and how you like things done.
That's all of the background knowledge that we're going to give to AI. And if you've been using ChachiBT or Claude for a while, then you've probably already given it all of this information before. You've definitely explained your work, worked through your offers, corrected the same phrases that you hate over and over.
And all of that context is buried in those conversations. conversations. The problem is that it's scattered across all of these chats inside of somebody else's product.
So we're going to pull all of that out and turn it into a few files that you own. These are called context files, and they're just marked down that tells your AI what it needs to know about you. For this setup, we're going to start with four.
The first is your about .md. This is who you are, what you do, and the background to your business. It's the basic information you'd otherwise have to explain and start every conversation with.
Next is Customer .md. It's who you help, what they're struggling with, and the questions or objections that they keep coming up with.
Include how they describe those problems in your own words so the AI knows who it's actually writing for. Then there's offer .md. That's what you sell, what's included, the costs, and the results that you can actually deliver.
So when you're working on a proposal or a landing page, it has the details of your offer to work from. And the fourth is voice .md. And this is the most important one.
This tells AI how to write in your own personal style. the phrases that you use, the things that you would never say, the examples of your actual writing. But the good news is that you don't need to sit down and write all of these files from scratch because you can actually export all of your chat history from your AI and then have it build it for you.
If you're in ChatGPT, go to settings, data controls, and export your data. In Claude, it's settings, privacy, export data. Once that export is ready, upload that into a new conversation with your AI and then ask it to go through every single one of your...
previous conversations to create these context files. Tell it to look at what you've explained, what you've corrected and what you actually decided. If something is missing or two conversations contradict each other, have it ask you and make sure that it's learning from your voice and the things that you've actually written and published.
Otherwise, it'll study its own drafts and you'll end up with a bad voice profile. You can add emails and call transcripts as well, which is some of the best sources because they capture how you explain things naturally. Then read the files.
because this is really important. Our whole minimalist system relies on less volume, higher quality. So you need to read your context files.
Because if something's wrong, that throws off every single future output. So fix anything that the AI might've messed up and then save them into your folder. These are now your living context files.
They sit outside of your conversations and Map tells your AI to read them at the start of each new chat and then update them as you work. If your pricing changes, it updates offer .md. If you learn something new about your customers, that goes into your customer file.
And then when you correct something about how it's writing, that saves back into your voice .md as well. So the next conversation starts with the updated information without you explaining anything ever again. Quickly, if you're a founder and you wanna build this system, for your business, then actually turn it into results with proven AI playbooks, that's what we'll do in my next Founder OS Bootcamp.
I'll hand over my exact setup, the AI playbooks that I use to run my business, then teach you how to adapt them to yours. Join the wait list below to get first access. ingredient number two is methods these are the processes your ai follows to do the work and we give it these processes through skills ai models are incredibly smart but if you don't give them a process they choose one for you and that's how you end up with really generic outputs a skill is just an operating procedure for ai it gives the agent specific steps to follow examples to work from and a way to check whether it's done a good job and at its simplest it's a folder with a file called skill .md that file holds the instruction and alongside it, you can keep examples, templates, or scripts for part of the job that you want done the same way every time.
These skills work like a bookshelf. The agent sees the name of each and a short description of when to use it. Then when you give it a task, it picks the right skill off of the shelf and reads the instructions, pulling in the examples when it needs them.
So having a library of skills doesn't mean that it's reading every instruction for every job. The detail stays on the shelf until it's needed. The time to create a skill is when you've just gotten a good result.
You've gone back and forth with the model, changed the prompt, corrected a few things, and finally, it's doing exactly what you wanted. If you're likely to do that same job again, stay in that conversation and use Skill Creator. It's a skill that builds other skills.
Just ask, use Skill Creator to turn what we just did into a reusable skill. Tell Skill Creator to save your skills in a .agents folder inside of your OS folder. Clawed code actually looks at a .clawed folder only.
So you need to tell your AI to also create that folder and then symlink each skill across. A symlink is just a pointer to the original folder. So both agents use the same files and you only have to maintain one copy.
Put that rule in your map as well. New skills go into the .skills folder. Then the agent knows where to save the next one.
If you're wondering how methods are different to memory, here's how I think about it. And it comes down to a simple question. Does the task repeat or does the information change?
If the task repeats, the process belongs in a skill. If it's information about your business, your customers, or what you're selling, that belongs in memory. Then when the agent runs a skill, the instructions tell it which context files to read for the current information.
So a writing skill can read voice .md to understand how you sound, customer .md to understand who it's writing for, and offer .md to understand what you're selling. Those details stay in the context files, so when your offer changes, you only have to update it once.
And the skills that use it can all read the updated version. Methods are how the work gets done. Memory is the information that it works with.
The agent brings those two together on the job. Ingredient number three is means. These are the connectors that give your AI access to the tools that you already use.
Things like Gmail, your calendar, your CRM, whatever you need to get work done should be connected as a tool. And the way to think about these is like you're adding a new capability to your AI. So for research, you might give it something like complexity.
now your ai can run perplexity searches in the same conversation with the context that you're already working on you don't have to open another app explain the task again or copy the results back and forth the capability comes into your workspace and the same applies for your business tools as well if you're writing a reply adding the gmail connector lets it find the original email create the draft in your inbox.
If you're preparing for a meeting, you can add your calendar connector so that it can pull your details about when you're free and then your CRM so it can find all of the customer information. Every connector uses MCP or model context protocol. It's a standard way for your AI agent to communicate with external tools.
And you don't really need to understand the plumbing to use it, but you do need to check what each connection lets it do because reading an email and drafting a reply are very different capabilities. Connectors are different to the files that we just saved. You enable them and connect them to your accounts in the AI tool that you're using.
I like to keep a list of connections that I'm using inside of my map. It starts with all of the apps that I'm already copying information out of, and then I connect the ones that I need for a job, for instance, so you can do more work with that same conversation. and before we get to the last ingredient the one that tells your ai how to use all of this we need to open the folder in cloud code or codex in the desktop app select your os folder as a project and then encodex that same folder as a local project and start a new chat now the agent can work with the files in that folder and we can have it create the directions for using them ingredient number four is the map this is the instruction file your ai reads when it starts working in your folder the simplest way to think about this is that your folder holds the information and your map tells AI where to find it and when to use it.
So if I'm writing a video, it tells my agent where my scripts live, which voice file to read and where to save the draft. Those directions are there every single time I start a new conversation. Your map should only contain information about how your OS works.
Things like where your context and projects are and any rules that might apply across everything. For me, that's where to save new skills and create that link that we just covered earlier. You should save it as an agent.
agents .md at the top of your OS folder so it works across Claude and ChachiBT because Claude code now supports agents .md directly so your agents can share the same map. If you're moving an existing setup across, just ask your AI to move the shared instructions from claw .md into an agents .md. All you have to do is rename the file.
Then check that each agent is actually reading it. And you can create this the same way that we created context files. Have AI read what you've already got and write it for you.
Just ask, read this folder and create an agents .md that explains where my context, skills, and projects live. when to use them and where to save new work. Include the skill links we set up and ask me about anything you can't work out from the files.
It creates the map inside of your OS folder and then you can read it and check that the directions work exactly how we've said. When you change the setup, have it update the map too. so this folder is now your complete ai operating system your context your skills and the work that you're doing all have a home here but you're probably wondering how to actually use this and if you've used projects in cloud or chat gpt before this should feel familiar you give the ai a place to work and the information it needs here you're working directly with files on your computer and you can put projects inside of other projects so you might have a client's folder a folder inside that for each client and inside that a folder for each job that you're doing for them you don't need to build all of those folders up front pick something that you're working on right now and ask ai to create a folder for it give it the notes and files that you've already got and have it update the map so it knows where that work belongs
When you sit down to do that work, open up your main OS folder and then start a new chat for the job. Tell it which project you're working on and what you want to do. It can use your shared context and skills, then go into that projects folder for all the details and have it save what it creates there too.
Any research documents or changes that you make together should go back into the OS. This is now your source of truth for all of your work. If you're stopping halfway through, ask it for a short handoff .md with what you've decided and what still needs doing.
next time that you open up that project your AI will automatically read the handoff .md. That's how I work out of one folder without keeping everything in one endless conversation. Each job gets its own chat and useful work stays in the files and you own this folder.
When a better model comes out you can open it in another compatible agent and keep the context skills and work you've already built. All you'll need to do is connect your accounts into the tool but you don't have to rebuild your system or explain your business from scratch. so that's ai minimalism memory methods means and map one folder that you can keep building on and if you want to build this out for your business join the founder os bootcamp the waitlist is in the description below subscribe if this was useful and check out my ai video editing workflow here
The Hook
The bait, then the rug-pull.
The video opens with a Jarvis-style fantasy, then flips it: the reason AI isn't working for you is that you're doing too much. The fix is one simple folder.
Named ideas worth stealing.
02:00list
The 4 Ms of AI Minimalism
- Memory
- Methods
- Means
- Map
Memory is what AI knows about you, Methods are the processes it follows, Means are connections and tools, and the Map tells it where things are and how to use the folder.
Steal fordiagnosing which part of an AI setup is missing
How they asked for the click.
VERBAL ASK
13:40product
“Join the Founder OS Bootcamp, the waitlist is in the description below.”
A mid-video soft pitch around the 4:00 mark plus a closing plug, kept brief so it does not interrupt the lesson.
MENTIONED ON CAMERA
Visual structure at a glance.
