If you write in Notion, track leads in HubSpot, record payments in a Google Sheet and edit images in Canva... then you need a Second Brain Operating System.

Open the old chat, ask it to summarise itself, copy the summary, paste it into a new chat.

Repeat 30 times per day, 30 days a month.

And then, you finally decide to give it everything. The project feature lets you upload 25 files on the $20 plan and 40 on the $100+ one.

The issue? You and I were promised an assistant. BUT what we got instead is a second job as a middleman. Constant back and forth between an AI and our own work.

Thing is, your Documents folder has 4,000 files in it. Eight years of context: documents, spreadsheets, proposals, reports, calls and drafts. All the context is right there, if only AI could somehow search through it and get to it without you doing the labour.

None of that is a prompting problem.

I know you have used AI every day for two years. You are good at prompting it. But somehow you still feel there's something lacking. The thing you are prompting cannot see your files, cannot use your programs, and cannot remember a enough about your preferences.

Picture a brilliant chef who is not allowed inside your kitchen. They stand outside and shout recipes through the door. They cannot see what is in your fridge, they cannot touch the stove, and tomorrow they will not remember your name.

So you run back and forth between the door and the stove, carrying ingredients out and instructions in. That is the paste. That is the scramblingg. That is the middleman you are today. Constantly struggling between context and prompts.

Every scene above is one of those three things missing:

• It cannot see your files, so you paste, you upload, you hit the wall at 25, you add by hand.
• It cannot use your programs, so it describes the job and you do it.
• It cannot remember, so you repeat yourself, and the good work sinks into a chat history.

Better prompts do not fix a chef who is locked outside. Opening the door does.

So here is the same week, with the chef inside the kitchen.

The door is called Claude Code. It is the same kind of AI model you already use, except it does not live in a browser tab. It uses a folder on your own computer, and it can use the computer.

Let me answer it the way people ask me.

"Do I still paste my context in every morning?" No. Your notes, drafts and client files are in the folder it is working in, so it reads what it needs by itself. And it searches everything you have ever written by meaning, so the note from two years ago turns up because it is relevant today, without you fetching it.

"Do I still repeat my preferences?" No. Your preferences live in a small text file, one preference per line, which it reads at the start of every session. You can open that file, read it top to bottom, change a line, delete a line. Nothing about you is kept in a box you cannot look inside.

"Will it do the thing, or describe the thing?" It does the thing, because it can run the programs on your machine. Convert the file, rename the 200 invoices, pull the figures out of the spreadsheet. You watch it happen and the result is sitting in your folder when it finishes. No dead links. No "I'll get back to you".

"What about the 25 file wall?" A folder has no wall. It does not try to swallow 4,000 files. It searches them, opens the three that matter, and reads the twenty lines it needs. That is why it stays sharp where a long chat goes soggy.

"Can I find what it wrote three weeks ago?" It is a file. Titled, dated, in your folder, next to your other notes on the same theme. And every change to it is tracked, the old line in red and the new line in green, so "I think it rewrote my whole intro" becomes four red lines and five green ones.

"And my clients' confidential files?" Here is the honest version, in two halves. The converting, merging, scanning, transcribing and searching all happen on your machine, so nothing gets uploaded to a free website ever again. What the AI model itself reads still goes to that model's provider, exactly as in a chat tab, so the terminal does not turn a cloud model into a privileged one. What it gives you is the choice. You decide which folder the AI may read, and which model reads it, including a small one that runs on your laptop with the wifi off. A browser tab never offers you that choice.

That is the transition. Same intelligence, but now it can see, act and remember, which means you stop being the courier.

Now, the door happens to be a black window with a blinking cursor.

I know. Stay with me.

Everything you have learned about AI happened inside a browser tab. And a tab is slow for one reason: somebody has to navigate. Click here, upload there, wait, download, drag.

Have you tried the AI tools that click around your browser for you? Yup. You are still faster than they are.

But no matter how good you are at clicking buttons, filling forms and browsing websites, there is a way that will always be faster.

It is called commanding. And for the longest time, it has been in the back seat for humans.

It happens in the terminal, the oldest way of using a computer there is, and in 2026 the fastest growing one. OpenAI's Codex went from three million people a week to five million in eight weeks. Claude Code gets installed 69 million times a month.

And one in five of those people is not a developer. It is someone like you. A writer. An operator. A mum organising voice notes on a stroller walk. People who had never touched a terminal and now will not put it down.

These are the people of 2030 and I want you to be one of them.

Here is what a terminal can actually do.

People see a blinking cursor on a black screen, assume the green lines are code they will never understand, and close the window.

Truth is you do not need to be a hacker, a prodigy or even a coder. Commands are close to plain English.

Take ocrmypdf scan.pdf searchable.pdf. Three words. It reads the text inside a scanned PDF (that is what OCR, optical character recognition, means) and makes the whole thing searchable.

Or ffmpeg -i call.mp4 -vn audio.mp3, which hands you the audio from a video. No website, no upload, no waiting for it to process, no download at the end. One step, without the frustration.

And because the terminal can be commanded, every program on your machine becomes something AI can use. Every serious AI lab, from Anthropic and OpenAI to Moonshot and Alibaba, trains its models hard on exactly this, to 80%+ accuracy on terminal commands.

To make it less abstract, here are everyday jobs you already do, and what each one becomes:

Now notice what every single row has in common. Nothing was uploaded to a strange website.

Every free site that does these jobs (iLovePDF, TinyPNG, CloudConvert) is, underneath, an upload-your-confidential-document service. You have been handing them client files, passports and identity cards for years.

All because the alternative looked like programming.

It was never programming. It was a one line command, and it is a shame nobody told you.

And you do not have to memorise a single one. If a command is easy enough for you to read, it is easy enough for the AI to write and run for you. You say "merge these two and send me pages 3 to 7", it picks the row.

Remember the marketer with five exports and the creator with five apps? Those are rows in that table now.

Which is what makes skills worth having.

You have heard the word. Sometimes it is called a prompt, recently it got rebranded as a skill.

A skill is a prompt, written and stored in advance. In Claude Code a skill becomes a slash command, a shorthand like /compress-image, which tells Claude Code (you guessed it) which terminal commands to run to shrink an image.

That is all there is to it: a set of instructions telling the AI which commands to use and in what order. It is also why useful skills need a terminal. In a chat window they have nowhere to run, so they sit there like a recipe card taped to the outside of the front door.

In Claude Code the chef is inside, so the recipe card becomes dinner.

Because the commands above flow. They hand files to each other without touching the internet. pdftotext writes the text, rg searches it, duckdb tabulates it, md-to-pdf renders the result.

Every app you pay for is a handful of workflows like that with buttons painted on top. And an app needs new buttons built for every new workflow, which is why you keep hitting the edge of what it can do.

So a skill you set up, find or buy can do this from a single instruction: read every PDF in a folder, find the clause you care about, tabulate what it finds, write the summary, render it as a PDF, email it to the person who asked.

No app will ever sell you that, because every app has an edge. The terminal does not.

But Claude Code out of the box is a new kitchen with nothing plugged in.

Look back at that table. Almost every row starts with a file. A draft, a report, a recording, a spreadsheet, a folder of PDFs.

That is not a coincidence. Commands operate on files. So the first move is not learning commands. It is getting your files into a place where you and the AI can both reach them.

And there is a step before even that, which decides how fast you get results.

Claude Code out of the box does not know what time it is, does not remember how you like things done, cannot search what you have written, and keeps no record of what it changed.

None of that is a flaw. It is just unconfigured, the way a new laptop is not already on your home wifi.

The trouble is that this is exactly where people quit.

They install Claude Code on a Saturday and copy a setup off a YouTube video slightly wrong. By Wednesday it has written the wrong date into three files, forgotten a preference they stated twice, and overwritten a paragraph they wanted. So they decide the terminal is not for them, when the foundation was wrong.

THEY WERE ONE HOUR AWAY.

Which is what the Claude Code Knowledge Work Coach is for. $36, once, one hour.

It is not a course. It is an employee that arrives already wired, and its first hour goes on the four things that Saturday install got wrong.

• Hooks, tiny automations that fire at a set moment, like the fridge light coming on when you open the door. Every session starts with the AI already knowing today's date, your timezone and the time.
• Git, which tracks every change line by line. The red and the green. Nothing it edits is ever silently gone.
• Memory, the readable file of how you like things done. State a preference once.
• QMD, a search engine that lives on your machine, never touches the internet, and finds your notes by meaning in under a second.

Four words: hooks are for automation, git is for version history, memory is for preferences, QMD is for searching. You do not learn any of them. You watch them get set up.

Then it brings your knowledge in, because everything else stands on it.

Good news here. Everything you have written is already text.

The Google Doc is text, the Notion page is text, the Apple Note is text. Google Docs exports to markdown from the File menu, Apple Notes exports to markdown, Notion hands you markdown.

So you make a folder on your machine and put your notes in it. That is the whole migration. Nothing is lost, and the three-way search across three apps becomes one.

The moment they are in that folder, two things become true at once.

They are searchable by meaning, which is the obvious half.

They are also reachable by every tool on your machine, which is the half nobody mentions. The same folder that answers "what did I decide about this client last quarter" is the folder a command can read, convert, search, tabulate and render from.

Your knowledge stops being something you look things up in and becomes material your work gets done on. The consultant's old proposals stop being an archive and become the first draft of the next one.

That is why this is the first product and not a later one. A terminal full of tools without your knowledge in it is a kitchen with an empty fridge. The chef is finally inside, and there is nothing to cook.

And it is $36 on purpose, so the commitment is an easy one.

Now the part you should be sceptical about.

Your apps already do some of this. Notion has search by meaning. Google Drive will find a file from a plain English description. Both genuinely work.

So here is the fair question. If Notion can already search my notes by meaning, why would I move them?

Because searching is the only thing it will ever do with them.

Say you have a Notion page of meeting notes from a client project, with a recording and a few PDFs attached. Here is what you would actually want to do with it on a Thursday afternoon.

"Send the client a branded PDF of the decisions." Notion exports a PDF that looks like a Notion page. You want your letterhead and your layout, so you copy it into Google Docs and reformat it by hand.

"Transcribe the call so it is searchable." Download the recording, upload it to a transcription service, pay per minute, wait, copy the text back in.

"Check the payment clause is the same across their last forty invoices." Download all forty. Open each one. Or give up, which is what actually happens.

"Publish the whiteboard photo without the GPS coordinates of my kitchen in it." Notion will not remove that. There is no menu for it.

"Pull the figures from the attached spreadsheet into the summary." Copy, paste, reformat, check you did not drop a row.

Every one of those is a single command once the same material is a folder on your disk. And you do not run them one by one. You use a slash command, because that is what skills are for.

This is not Notion being badly built. Your notes live inside their product, and a product can only offer the operations it decided to build, for everybody at once and nobody in particular.

A folder on your own disk has no such edge. Every program you install can reach it. The search index is a file sitting next to your notes. And the model reading them is a decision you keep making: the best one today, a better one in six months, a small local one on a flight with no signal, three of them on the same question when the question deserves it.

Models get replaced constantly. Your folder does not get migrated, re-indexed or re-bought when they do. It was always just a folder.

Then it stops being a search box.

Everything so far concerns the notes you already had. Here is the half that changes your week.

You are working something out. Researching a decision, shaping an offer, reading three things and forming a view. The Coach is in that session with you, and as the thinking happens it writes it down: a markdown file in the folder you are already in, titled, dated, sitting with your other notes on that theme.

So next month, when the same question turns up wearing different clothes, the answer is already on your disk and the search finds it by meaning.

Remember the positioning you rewrote from memory, slightly worse? That stops happening, because nothing good lives in a chat history any more. It lives in a file.

That is a knowledge-work partner, where before you had a note-taking app. The app waits for you to file something. The partner is in the work, writes down what came out of it, and can read everything you have ever written while it helps with the next thing.

It also logs its own work, so you can walk backwards across a week and see what your Second Brain OS has been up to. That log is for you and it is analytical: what leaves your machine is the record that work happened, never the content of the work.

P.S. Warning: once you have watched it find a note from 2023 that you had forgotten writing, you will start writing a lot more down.

Where this is going, and where it stops.

A Second Brain ends at "I have notes, I want them related by meaning, and I want an AI to search them and talk to me about them".

That is one layer of a Second Brain Operating System. It is the right layer to start on, because it gets you comfortable in the terminal, where the payoff is immediate and the stakes are your own notes.

No subscription. No seat. No monthly fee for the privilege of searching your own writing.

Walk away in two years and you still have a folder of markdown files on your disk, which every editor and every model can open, plus an index you could rebuild in minutes. Compare that with exporting a workspace and hoping.

What it does not do: write your daily recaps, your weekly strategy or your monthly retros. That is its sibling, the Productivity Coach, which reads everything this one wrote and hands your week back.

This one handles knowledge. Writing it down, keeping the history, finding it by meaning.

One hour of uninterrupted attention, and the hour is mostly you watching.

Get the Knowledge Work Coach →

Stop shouting through the letterbox. Let the chef in.

Your writing, your leads, your money and your visuals were never four problems. They were one business, kept in four places.

Umair

P.S. The Knowledge Work Coach is the $36 employee that sets up Claude Code properly on day one: hooks, git, memory and QMD, in one hour. It starts with your knowledge, because everything else stands on it.