🎉 Giảm 20% cho đơn hàng đầu tiên!✦🚚 Miễn phí vận chuyển cho đơn trên 500.000₫✦⚡ Flash Sale — Số lượng có hạn!✦🎁 Quà tặng kèm cho đơn từ 1.000.000₫✦🎉 Giảm 20% cho đơn hàng đầu tiên!✦🚚 Miễn phí vận chuyển cho đơn trên 500.000₫✦⚡ Flash Sale — Số lượng có hạn!✦🎁 Quà tặng kèm cho đơn từ 1.000.000₫✦
5 Steps to Bring AI Into All of Your Work — and Stay in Control: Second Brain, Company Brain, and One-Job Agents
Quay lại Blog

5 Steps to Bring AI Into All of Your Work — and Stay in Control: Second Brain, Company Brain, and One-Job Agents

17 tháng 11, 202610 phút đọcVũ Khôi
English
#English
Vũ Khôi

Vũ Khôi

Tác giả tại Wotech Electronics Blog

Illustration: the Obsidian interface with graph view and the file list of "Ngoc's brain", a screenshot from the original post

5 Steps to Bring AI Into All of Your Work — and Stay in Control: Second Brain, Company Brain, and One-Job Agents

Nguyen Anh Ngoc argues that merely opening ChatGPT, Claude or Gemini to ask questions — or dropping in a file and asking for analysis — creates no real difference. The difference comes when you reorganise how AI reads your data and works for you. In a post dated 17/11/2026, he shares the 5 steps he applies to his own work and to his company, Cham Station, and claims anyone can apply them to see a completely different level of effectiveness. All processes, figures and judgements in this article follow the author's own shared experience and have not been independently verified.

Quick answer: According to Nguyen Anh Ngoc, bringing AI into all of your work while staying in control takes 5 steps: build a Second Brain — a second brain in Obsidian for AI to read in every conversation; separate a Company Brain — an enterprise brain holding verified knowledge, kept apart from the personal brain; build a multi-channel AI assistant that reads only the company brain and hands important customers over to you; let an agent sit in on every meeting, turning transcripts into structured minutes and feeding the knowledge back into both brains; and give each agent exactly one job, with a separate budget and a full log. The author claims this model saves significant cost and time compared with merely asking a chatbot isolated questions.

A Second Brain is a personal data store that AI can read in every conversation: how you write, how you decide, lessons learned, products, customers and project context. A Company Brain is the shared knowledge store of a company or project — pricing, decision history, customer records, product specifications, hard-won lessons and business results — kept separate from the personal brain so that agents can share it.

What is a Second Brain, and how do you build one?

The author uses Obsidian to organise his Second Brain. In essence, he says, it is simply a folder holding hundreds of .md files, recording everything: how he writes, how he decides, lessons, products, customers, and project and company context.

The Second Brain now holds over 100 pages, according to the author, and is connected straight into Claude through a folder called "Ngoc's brain", so that AI can use all of that content every time he opens any conversation. With a Second Brain in place, he argues, you no longer need isolated projects in chat or cowork modes: each time you talk to AI, it reads the whole Second Brain, and you never have to explain everything from scratch.

This is the author's first key point: AI is only truly useful when it already has your context, instead of every conversation starting from zero.

How is a Company Brain different from a Second Brain?

A Second Brain is a personal brain — how you think. A Company Brain is an enterprise brain: it holds pricing, decision history, customer records, product specifications, hard-won lessons and business results. The author keeps the two brains separate, and argues that a company or project without a Company Brain has no shared memory built for its agents.

At the root of the project, the author keeps 2 important files:

  • One file recording the operating role — he feeds it to the coordinating agents every morning when opening the "office".
  • One file recording all of the company's verified knowledge — every agent reads it before working.

Alongside them are 2 folders:

  • Staging: where agents write down raw lessons as they work.
  • Canon: where verified knowledge lives, reviewed by the author every week.

This boundary, according to the author, is what keeps company knowledge trustworthy: raw lessons must pass review before becoming official knowledge that all agents may use.

How does the multi-channel AI assistant CHAM work?

The author's assistant is called CHAM, trained on the company brain. The author stresses one principle and asks readers to remember it: the assistant should only read the company or project brain — never allow it to read the Second Brain.

Put the assistant on the channel your customers prefer, the author says: that could be Zalo, Facebook or Telegram. Grant it permissions and limit what the assistant may say, set classification and customer-evaluation principles, and require it to hand over to you immediately when it recognises an important or potential customer — based on the consultation chain.

In other words, the assistant is not a chatbot answering freely, but a consulting staff member with a clear scope of work and a hotline to a real person when a valuable customer appears.

What can an agent do by sitting in on every meeting?

Every Zoom meeting, Google Meet, customer interview and important call the author holds is recorded. Then an agent turns the transcript into structured minutes covering: lessons learned, what I owe whom, who owes me, memorable quotes, and what this month's or quarter's results are.

The author notes that an upcoming Claude meeting feature will make this simpler. But the most important point, he says, comes in the next step: all the results the agent hears and writes down should be fed back into the Second Brain or the Company Brain, thickening the brains over time. The agents that execute work then read that same knowledge — creating a loop in which the company's experience and lessons accumulate, and new employees keep working on top of what has already been accumulated.

Why should each agent do only one job?

The author currently creates agents running through Hermes and Claude, each with one clear job:

  • One agent only reads email each morning, reporting which messages matter and what to do with them.
  • One agent only works with Google Calendar, scheduling, reading the calendar from his system and reminding him one day and 15 minutes before each event.
  • One agent writes content in the author's voice (brand voice).
  • One agent filters the knowledge of the project and the company.
  • One agent packages content for the "office" briefing each morning.

Every run, every agent has a log file. Because of that, the author knows which agent did what, how many tokens it used and what errors occurred. He says he likes running agents through Hermes because it can fix itself and write its own skills.

On budget, the author runs agents through separate API keys, in order to compare and find ways to optimise cost over time. Splitting jobs, separating budgets and logging — these three, he says, are how you stay in control as the number of agents grows.

Apply it now: how do you start the 5 steps?

Following the author's own sequence, you can start as follows:

  1. Gather context into one folder. Record everything you keep having to re-explain — how you write, how you decide, lessons, products, customers, project context — as .md files. That is the Second Brain. Connect the folder to the AI tool you use so it reads it in every conversation.
  2. Separate the Company Brain. Create 2 root files — the operating role and the verified knowledge — plus 2 folders, Staging (raw lessons) and Canon (reviewed knowledge). Review Canon every week.
  3. Place the assistant on the one channel customers use most. Grant permission to read only the company brain, state clearly the limits on what the assistant may say, and set a rule to hand important customers over to you immediately.
  4. Record every important meeting. Let an agent turn the recording into structured minutes, then feed the results back into both brains so knowledge accumulates over time.
  5. Give one agent one job. Each agent gets its own budget and its own log file; track each agent's tokens and errors to optimise cost gradually.

How do isolated questions differ from the 5-step model?

Aspect Only asking a chatbot isolated questions The author's 5-step model Why the author chooses this way
Memory Every conversation starts from scratch AI reads the whole Second Brain in every conversation Never explain the context again
Reliability AI answers from whatever you just gave it Agents read verified knowledge in Canon before working Knowledge must pass weekly review before shared use
Data boundaries Personal context easily leaks into shared work The assistant reads only the company brain, never the Second Brain Personal context stays protected from customer channels
Accumulation Experience is lost with each conversation Meeting minutes feed back into both brains, creating an accumulation loop Newcomers work on top of accumulated experience
Cost control No way to measure which agent costs what Each agent has its own budget, its own API key and one log file Compare and optimise cost over time

The author's consistent view is: chatting with a chatbot is not a system. A real system consists of two brains with clear boundaries, one-job agents with their own logs and budgets, and a loop feeding knowledge from daily work back into the brains. He claims applying these 5 steps saves a great deal of cost and time — a judgement based on his own experience at Cham Station, not independently verified.

Frequently asked questions

1. How are a Second Brain and a Company Brain different? According to the author, a Second Brain is a personal brain — how you think, write and decide. A Company Brain is an enterprise brain — pricing, decision history, customer records, product specifications, hard-won lessons and business results. The two brains stay separate so agents share company memory without touching personal context.

2. Why may the AI assistant read only the company brain, never the Second Brain? This is the principle the author stresses most. The assistant lives on customer channels, so it may only access company knowledge; the Second Brain holds personal thinking and context, and must never be opened to the assistant.

3. If I work alone, which step do I start with? Following the author's sequence, start with step 1: gather context into a Second Brain using Obsidian. Once the personal brain exists, separate a Company Brain for the project and build the agents on top.

4. With one job per agent, how do you manage them and control cost? The author gives each agent its own log file and runs it through a separate API key. That way he knows which agent did what, how many tokens it used and what errors occurred — and can compare to optimise cost over time.

5. Does this model require expensive tools? The post gives no total cost. The author says he uses Obsidian for the Second Brain, Claude as the main conversational model and Hermes to run agents, while separating API keys per agent to track and optimise cost.

Source: a Facebook post by Nguyen Anh Ngoc (Cham Station), dated 17/11/2026. All processes, figures and judgements in this article follow the author's shared experience and have not been independently verified.