Now back to a Second Brain Operating System.
Once upon a time, humans used a computer. Now they ask AI to use a computer. The end.
But wait. The AI uses the computer? Ah, there’s a catch.
It sucks at doing things without being told what to do. And I know you wish it could just guess 100% of what’s on your mind and just do it.
And when I say somewhere between Tiago Forte and Tony Stark, we are closer to Tony Stark now. Which is exactly why you need multiple AI Agents (each equipped with knowledge, tools & skills) to shape your Second Brain Operating System.
By having these agents, you cut down the load of instruction to just a few commands. Sometimes (like me) a single one.
If I need to recall what I did in Q4 2024, I don’t need to tell Claude Code:
‘Please use listTodoistTasks to find my tasks between September to December 2024, write to a file and discuss with me how this Q4 needs to be different’.
Instead I just go like this ‘/productivity-coach-agent: I’d like to recap Q4 2024 and plan Q4 2025 differently’.
Now scale this: if you had multiple AI Agents from being able to cut through your knowledge, to figure out what content needs to be created to updating your website to reading your analytics to making decisions on Ad Spend to tallying payments & filing your accounts… well you guessed it. You have a Second Brain Operating System.
My goal is to take you to 2030, sooner rather than later. Speaking of, what will the world look like in 2030?
Well I am glad you (didn’t or did) ask my friend. It’s a world where the Big Giants have their hands tied. Their biggest fear where people stop paying for SAAS subscriptions, pay only for heavy compute and do almost everything locally on their own device becomes a reality.
Where the sheer amount of data they have on you stops accumulating the way it used to.
Which is all of what this newsletter is about. Spoiler: I intend to sell you things that free you from (stupid) monthly subscriptions and make you independent.
But the information is always free. You pay for the AI Agents and not the courses. All of my courses are public, so you can steal them with a simple google search. You're welcome!
Fun fact: did you know your MacBook Air is idle 95% of the time you're using it?
Not exaggerating. Open Activity Monitor right now, mid-work, twenty tabs deep. The chip is asleep between your keystrokes. It has been all day.
So when your Air lags, it isn't short of power. It's short of somewhere to PUT things, and the moment RAM fills up, your Mac starts misbehaving. That's the problem. Same chip, same speed, nowhere to work.
Which means the fix is memory, not a second computer. And you can stop convincing your wife "Honey I NEED a Mac Mini!" (She's right Kevin, you F***n don't!!).
Fun(ner) fact: Here’s why you don't need to buy a Mac Mini, Mac Studio or Mac Ultra. Running a harness (OpenClaw, Claude Code, Cursor, Codex) is light work, and the MacBook Air has plenty of room for it. What actually costs you is running an AI model on your own machine, and on Apple Silicon the thing that decides whether a model runs at all is RAM. Not the chip. Which only makes sense once you have good enough local models, and right now even the best of them are miles & miles away from Opus, Fable, Kimi K3 (which won't fit on a Mac Ultra, Mac Studio and definitely not on a Mac Mini).
Quick laptop buying guide:
- Browser + Terminal (Claude Code) use -> Air
- ++ Video Exports & Edits at the same time -> Pro
- ++ Run small to medium model (LM Studio + Open Code) at the same time -> Max
All the people on YouTube who make reviews, benchmarks and do "real world tests" have one conclusion: hardly anyone manages to use a MacBook Air to its fullest.
All your problems will go away the day you have a bigger screen and more RAM.
The chip (M1 or M5) is rarely what's holding you back, because it still has a ton of spare capacity. RAM is. Stay portable, please.
P.S. If you can get an M2, M3, M4 Air LAPTOP with a 15.6 inch display and 24Gb+ RAM at the same price, buy that over an M5.
2030 is a world where you have a Second Brain Operating System via your terminal
And the harness we use is Claude Code.
(Though it should work out of the box with OpenCode, OpenClaw, Codex, Cursor, Hermes or whatever. Fingers crossed!)
Not because it’s the prettiest. Because it’s the one harness where things just work.
Where a hook, a memory file and a slash command already exist, so an agent can be handed to you fully wired. Don’t get distracted by OpenClaw, Hermes, or whatever the Buzz is about.
So on day one, you don’t just get a course. You get an employee. The Claude Code Knowledge Work Coach.
It's purposely $36 so you make an easy commitment to 2030.
But first: what is a terminal actually FOR?
Fair question. Nobody ever tells you. They just say "learn the terminal" like it's kale.
Here's the honest answer: in a terminal, every program installed on your machine becomes a tool your AI can use. That's it. That's the whole thing. And it lands differently once you see what it means for a normal week.
The errands. Things you did last month by hand, in a browser tab:
- A draft into a client-ready PDF.
md-to-pdf proposal.md - An iPhone photo your website will actually accept.
magick photo.heic -resize 1600x -strip web.jpg - Two contracts joined into one file.
qpdf --empty --pages a.pdf b.pdf -- merged.pdf - A 4GB video sent to a client, encrypted, no 2GB cap.
croc send film.mp4
Now notice what every one of those has in common: nothing got uploaded.
Because every free web tool that does these things is, underneath, an upload-your-confidential-document service. iLovePDF wants your contracts. TinyPNG wants your images. You've been handing them client files for years, because the alternative looked like programming.
Then the bigger category: the jobs you quietly gave up on.
Not hard jobs. Jobs that were two hundred of something. Pulling the text out of forty PDFs so you could search them. Stripping GPS data from photos before publishing (your home address is in the metadata of that photo of your desk, by the way). Finding which of four thousand documents mentions a client. Opening a CSV that Excel simply refuses to open because it has more than a million rows.
Each of those is the same one command with a * in it. Two hundred files cost the same effort as one.
Read that again, because it's not a time-saving claim. It changes which jobs are worth doing at all.
And the surprising category: things you assumed you had to buy.
Transcribing a client call on your own machine, where the recording of a privileged conversation never leaves it. Cutting 90 seconds out of the middle of a recording, losslessly, in about a second. Making a scanned contract searchable. Capturing a web page as clean markdown that lands in your notes and is findable by meaning before you've closed the tab.
If you're a consultant or a coach, that first one isn't convenience. It's the difference between being allowed to record the call and not.
Now — none of that works well until the plumbing is right. Which is the actual job of the $36.
And out of the box, it sets up Claude Code with the right settings
Without you having to figure it out or learn the terminal on day one.
Its first job, Hooks & Utilities Setup, is five skills that do the boring plumbing you would otherwise copy off a YouTube video wrong. Four words first, because you’ll meet them everywhere from here on.
1. A hook is a tiny automation that triggers at a specific moment.
Some of those moments are when you start a Claude Code session or end one, before you use an external tool, after you use an external tool, when Claude finishes its response, etc.
Think of the light that comes on when you open the fridge.
That's a hook firing on 'fridge door opened'.
A defined, inevitable moment that happens as part of your use. A trigger.
Why would you need a hook? For one, an AI model doesn't have a reliable clock.
Every time you open Claude Code, it wakes up with no idea what time it is, what timezone you’re in, or what you were doing when you closed it last night. Ask it to ‘log this for today’ and it will confidently write a time from its training data.
Ask it to ‘continue where we left off’ and it will ask you where that was. Not because it’s dumb. Because nobody told it. A hook is how you tell it, every single time, without typing it.
The Coach wires two hooks that solve both these problems, once and for all.
The first is a session-start hook. The moment you start Claude Code in your terminal (from where you operate your Second Brain OS), it runs a script that tells the AI what day it is, what timezone you’re in and the current time, and that this session has begun. Before you say a word, it already knows where it is in your week.
[Screenshot: a fresh Claude Code session, the session-start hook output showing the date and timezone already loaded]
The second is a pre-tool-use hook. And this means 'before it uses a tool'.
The two cloud tools the Knowledge Work Coach uses are updateActivityState and addActivityEvent.
The point? You can log into a dashboard and have a clear visual of the work your Second Brain OS has been up to.
This is purely analytical, purely for you and does not put any of your knowledge/notes on the cloud.
It lets you see when a large task was completed, a big goal was met, a difficult research question was tackled, and more. And you can ask it to pull those logs into the chat to help you backtrack. For me personally, it's a way to walk backwards across my Second Brain Operating System.
Now this little hook just has one purpose: it fires right before each of those two tools is called, and hands it a fresh timestamp taken from your machine at that exact moment. The session-start hook tells the AI what time it is when you sit down; this one makes sure that an event logged four hours into a session is stamped four hours later, and not with the time you started. Which means every event, from a task to a goal, lands with the correct date & time every single time (in your time).
And that accuracy makes the dashboard worth looking at.
[Screenshot: the pre-tool-use hook firing, the injected timestamp visible right before a tool call]
2. Next, Git for tracking changes line by line.
If you've ever wondered 'why everyone keeps saying git commit, git commit, git commit', then get ready for your life to change.
Unlike a normal save, a git commit treats the ‘history of changes’ like business class passengers.
Now you might recall this way of saving files "-final-v1", "final-final-v2", "ultra-final-final-v3"...
And that's a problem developers solved 20 years ago; you’ve been living without it in Google Docs and Notion, where ‘version history’ is a menu you never open.
But why does a folder of notes need what developers use?
Because of how AI likes to read, write and track changes (as you will too!).
In 2030, your docs & notes are markdown. Plain text with a little punctuation for headings and bold, so light that a year of daily notes weighs less than one photo on your iPhone.
The Coach turns your knowledge folder into a git repo on day one (if it isn’t one already). You don’t need to learn git. The Claude Code Knowledge Work coach just does it.
Now plain text has lines. Line 1, line 2, line 214. Which means Claude Code doesn’t have to swallow a 3,000-word document to change one sentence in it. It reads lines 40 to 60, edits line 52, done. That’s how it avoids context overload, and it’s how you avoid it too. You can point at a line number the way you’d point at a page number, and both of you are looking at the same thing.
[Screenshot: Claude Code reading a range of lines from a note, then editing a single line]
And then comes the biggest upgrade, and it’s one word: diff.
When you edit a Google Doc, the old sentence is gone. You trust that whatever replaced it is better. But when Claude Code edits a line in a git folder, you see the diff.
The old line in red, the new line in green, side by side, nothing else on the screen.
You see exactly what changed, and so does the AI, because it reads the same diff before it moves on. There is no ‘I think it rewrote my whole intro’. You look at 4 red lines and 5 green ones and you know.
Heck, you can even just search the red (the discard) and the green (the addition/change) across time. Let that sink in: your Second Brain OS can search through just the changes, without loading the full file into its context/burning tokens and help you figure out when & why you (or it) made a change.
[Screenshot: a diff in Claude Code, old line in red, new line in green]
[Screenshot: git history of a note, one line per day with the change message]
And when you want to read a note like a human instead of a robot, CMD + SHIFT + V in VS Code and the markdown turns into the pretty version: headings, bold, tables, diagrams.
Write in the raw, read in the pretty. The file never changes. No Obsidian, Notion, Evernote needed.
Common confusion: Git itself is a certain technology for managing files line by line. Github is a provider famous for making that technology cloud first. You always have the option to create a personal GitHub account & backup your knowledge. It's an option, not a necessity. Other options like iCloud also & always exist for cloud backups.
[Screenshot: the same note raw on the left, CMD + SHIFT + V preview on the right]
3. Memory is where the AI keeps how you like things done.
Not what you know. How you work. There’s a difference and it matters.
Every time you tell Claude Code ‘put the date at the top, not the bottom’, or ‘my weekly plans live in the progress folder’, or ‘never use em dashes’, you are stating a preference.
Memory is often a tiny file with each line stating a preference that Claude Code reads at start of every session.
Without memory, you state it again tomorrow. And the day after. That’s the part of Claude Code that makes people give up after 2 days.
It’s not that Claude Code can’t do the thing, it’s that people forget to set their preferences.
[Screenshot: the memory folder, a handful of one-preference files, one opened]
4. QMD is how the AI finds what you know.
Memory is preferences. Knowledge is everything else. Every note, every reflection, every client call summary, every idea you wrote at 2am and forgot.
But the problem with knowledge isn’t writing it. It’s finding it again in a year.
QMD stands for ‘query markdown’. It’s a search engine that lives on your machine, points at your folders, and never touches the internet.
And that isn’t just great for your privacy, it’s also 0 cost and lightning fast. Rarely does something tick all the boxes, and QMD does.
The Coach installs it and registers each subfolder as a collection: progress, general, clients, ideas, whatever your folders are called. From then on, every note you save is indexed within seconds.
[Screenshot: QMD status showing the registered collections and document counts]
Now the part that beats every ‘Second Brain’ tool you’ve ever used (and dropped). QMD searches three ways, and the Claude Code Knowledge Work coach picks the right one:
- By keyword. You know the word. ‘Stripe fee’. It finds every note with those words, fastest.
- By meaning. You don’t know the word. ‘That thing about payment costs eating margin.’ It finds the Stripe note anyway, because it relates notes by what they mean, not what they say.
- By hypothesis. You know what the answer would look like but not where it is. The AI writes a fake paragraph of the answer and goes hunting for the note that looks most like it.
Combine keyword and meaning, and your 2024 note surfaces next to your 2026 one because they were about the same thing, even if you never used the same words twice.
[Screenshot: a QMD query in Claude Code, keyword and meaning results side by side, a 2024 note surfacing next to a 2026 one]
And because it’s local, the AI can read a result by line number, exactly like git taught it. Line 40 to 60 of the note that matched. Not the whole file, just that one relevant paragraph.
Recap: hooks are for automation, git is for version history, memory is for preferences, QMD is for searching your knowledge by meaning. Four words. Set up once. The Claude Code Knowledge Coach does all four in the first hour.
Then it writes the way a second brain should be written. Every piece of knowledge work becomes a markdown file with a title, tags and a date at the top, in the collection it belongs to, on your local disk — never uploaded to anyone's vector store, never indexed for anyone else to search. All searchable by meaning, with results ranked & returned in < 1 second, by a search engine running on your own disk. And it logs its activity & tasks so you can see EXACTLY what it did from Monday to Sunday. You get a knowledge-work employee you can monitor, not a chat window you have to babysit.
Thing is all the problems you experience from day to day have been solved already. You just need that setup (a Second Brain OS) to slowly (or quickly) but surely come to your machine sooner rather than later.
All the Claude Code Knowledge Work coach needs is 1 hour of your uninterrupted attention after purchase.
This is why everything that comes from me comes with $0/month bill. I aim to give you whatever can work on your device, you pay for the Agent once, the agent sets it up and it runs. Whatever can’t run practically on your device (e.g cloud tools), you’re on a pay as you go plan with the Second Brain OS and it’s purely credit based billing. You only pay for what you use.
P.P.S. If you're down to get started with a Second Brain Operating System, here's how
