Field Note · Your AI System

Your AI Forgets You Every Morning. That Is Not a Prompt Problem.

你的 AI 每天早上都忘了你。缺的不是提示词,是记性。

In plain English Every morning you open a new chat and explain your business again. Who you are, what you sell, how you write. You get a decent answer. Tomorrow, you do it all over. The answer was fine. Nothing stayed. This piece is about the fix: make every session leave one thing behind, so your effort starts earning interest instead of resetting. The Chinese AI world calls it 复利工程. You do not need a new tool for it. You need one question at the end of every session. 每天早上,你开一个新对话,把生意重新解释一遍。你是谁、你卖什么、你怎么讲话。答案还行。明天,再来一遍。答案没问题,问题是什么都没留下。这篇讲怎么解决:让每一次协作都沉淀一样东西,让你的力气开始生息,而不是每天归零。中文圈叫它「复利工程」。你不需要新工具,你只需要在结束前,多问一句话。
TL;DR

Your AI does not have a memory problem. Your business does. Every session should leave one deposit behind: a memory, a skill, or a preference. Prompts are rented. Deposits are owned. And the lazy part comes later, not now. Hard first, then peace.

你的 AI 没有记忆问题,你的生意才有。每次协作都该沉淀一样东西:记忆技能偏好。提示词是租的,沉淀才是你的。至于「偷懒」,那是后面的奖励,不是现在的方法。先难,后安。

For the first two years, I ran my whole business out of my own head.

Every client. Every promise. Every follow-up I owed someone. I told myself I had a good memory. What I actually had was 压抑, that pressed-down feeling where you are never fully in the room, because some part of you is always busy holding everything in place. If I did not write the list down, I could not sleep properly. I still get that feeling on the days I skip it.

Then AI showed up, and I did what most owners do. Opened a chat. Explained my business. Got a good answer. Closed the tab.

Next morning: opened a new chat, explained my business.

I had hired the fastest employee in Malaysia. And I was firing him every night.

That is the part nobody says out loud. You are not bad at prompting. You are re-onboarding a brand new staff member every single morning, and then wondering why the business will not move any faster than you can talk.

Where this came from

I watch both AI worlds. The English one and the Chinese one. They barely talk to each other, and most people only live in one. My job is to take the useful thing from either side and turn it into something a Malaysian SME owner can actually use on a Tuesday.

This one came from the Chinese side. A short Douyin video by 晓辉博士 about 复利工程, compounding engineering. Her claim is that the gap between someone getting real leverage out of AI and someone stuck on the treadmill has nothing to do with prompts.

Can the experience of working with AI once be sedimented? Can we stop saying the same thing over and over? You have to make it remember.

一次跟 AI 协作的经验能不能沉淀下来?能不能不要反复的去说一件事儿,一定要让它长记性。 — 晓辉博士 · 复利工程

沉淀. To settle. The way silt settles out of water and slowly turns into ground you can stand on.

A session that leaves no sediment is water you poured through your hands. I did that for a long time, and I felt clever about it, because I had gotten very good at explaining myself to a blank chat window. That is not a skill. That is a tax. I had paid it so many times I mistook it for expertise.

The prompt is 10%. What it leaves behind is 90%.

You already understand compounding. You just apply it to money and never to your own effort.

Nobody gets rich from one good month. People get rich because what they earned last year is still working for them this year. Same rule here. Nobody gets leverage from one good prompt. You get leverage when the thing you explained in January is still doing its job in June, and you never had to say it again.

leverage sessions over time → explaining yourself, forever every session leaves something
The reset: a good prompt, nothing kept The compound: every session deposits

The shape is the point, not the numbers. This is a picture, not a study.

Look at the flat line. That is not a failure of effort, and that is what makes it cruel. You can work extremely hard along that flat line for a whole year and finish the year exactly where you started it.

Three things a session should leave behind

This is the whole practice. It is small enough to start on a Tuesday afternoon.

Deposit 1
Memory
记忆

The facts that do not change. What you sell, who you serve, what you charge, who your best customer actually is.

You stop re-explaining who you are.
Deposit 2
Skills
技能

A procedure, written down once, that can be run again. How a quotation gets made. How you answer a price objection.

You stop re-teaching the job.
Deposit 3
Preferences
偏好

How you like it done. Your tone, your format, and the things you will not do no matter who asks.

You stop correcting the same mistake.

So here is the question. It takes five seconds, and you ask it at the end of every session.

What did this leave behind?

If the honest answer is nothing, you rented that answer. It is gone. You will pay for it again next week.

What to write down first

The objection lands right here, and it is fair. Write down my whole business? I don't have time for that.

You don't write down everything. You sort it. 晓辉博士 uses the four quadrants of what you know and what you don't, and it turns a vague chore into an obvious first move.

已知的已知
What you know you know

Your offer, your prices, your customers, how you talk. You could recite it in your sleep. Which is exactly why you keep reciting it to a blank chat.

Start here · fastest win
已知的未知
What you know you don't know

Your learning queue. The things you already know you should understand, and don't yet. Point your AI research here.

未知的已知
What you know but never said

How you price a difficult client. How you sense the lead that will waste three months of your life. You have never written it down, so nobody can do it like you. Not a new staff member. Not an AI.

This one is the moat
未知的未知
Your blind spots

What you cannot see from inside your own business. AI is weakest here. Don't wait for it to find these for you. People will.

Start top-left. It is fast, and it stops the daily bleeding.

Then go and dig out the bottom-left, the things you know but have never said out loud. That quadrant is the moat. It is the whole difference between an AI that sounds like everybody and an AI that sounds like you. Your competitor can copy your prompt in ten seconds. They cannot copy twenty years of knowing exactly when to walk away from a customer.

Where I part ways with her

She ends on 偷懒. Be lazy. Cut the repetitive work, push the dirty jobs to the AI, keep your energy for the creative work.

You have to keep thinking about how to be lazy, how to cut repetitive work, so you can genuinely put your energy into creative work.

一定要去想着怎么能够偷懒,怎么能够减少重复性的事情,真正把精力投入在一些创造性的活动上。 — 晓辉博士

She is right about the destination. She skipped a step.

Writing down what you know is work. It is boring. Nobody claps for it. Sitting down and typing out how you actually price a job is not the fun part of owning a business, and the reason most owners never build a system is not that they lack the tools. It is that this part is dull and nobody is watching.

Lazy is the reward. It is not the method.

There are two kinds of peace of mind. There is the fake one, where you dodge the hard part and tell yourself you will get to it later. And there is the real one, earned on the far side of the boring work, where the system truly remembers your clients and you can finally sit with your family in the evening instead of running a list in your head.

I know which one I was buying for years. Hard first, then peace. Anyone offering you the peace without the price is selling a shortcut that does not exist.

And keep the door

One more thing she does not say, and I hold this line hard.

Compound the repetitive work away, yes. But anything you cannot take back, publishing, sending, spending, stays behind a human decision. Yours. Human before AI. The system carries your memory so your judgment is free to do the thing only you can do. Not so it can quietly make calls in your name while you are not looking.

Automation without a door is not leverage. It is a faster way to be wrong.

Start this week

Don't build a system. Build one deposit. Three moves, and the first takes an afternoon.

  • Write the top-left quadrant, once. Who you are, what you sell, who you serve, how you talk. Rough is fine. Written is the point.
  • Put it where the AI reads it first, not in a note you copy and paste from. That is the entire difference between a system that remembers and a chat that forgets. I have written up the two levels of doing this, from a shared project that only recalls, to a folder of files an agent reads and then acts on.
  • End every session with the question. What did this leave behind? Then actually put it back in. That one habit is the whole of 复利工程. Everything else is furniture.

You are not behind because somebody else found a better prompt. You are behind because your effort keeps resetting to zero every morning, and no amount of hard work survives that.

Stop explaining yourself to your AI. Start keeping what you teach it.

Less noise. A living map.

缺的不是提示词,是记性。

— Weiss Ang · 1,300+ business owners trained · 99.1% recommend

FAQ

Why does my AI keep forgetting my business?
Because a normal chat starts from a blank page every time. Whatever you explained yesterday lived in that one conversation, and it died when you closed the tab. You are not bad at prompting. You are re-hiring the same employee every morning. The fix is a place the AI always reads first, before it does anything.
What is compounding engineering (复利工程)?
You judge an AI session by what it leaves behind, not by how good today's answer was. Every session should deposit one reusable thing: a memory, a skill, or a preference. Then tomorrow starts further along instead of from zero. The term comes from Chinese AI commentator 晓辉博士.
What are the three things a session should leave behind?
Memory (the facts that don't change: what you sell, who you serve, what you charge). Skills (a procedure written once and re-run on demand, like how a quotation gets made). Preferences (your tone, your format, what you refuse). If a session leaves none of these, you rented the answer.
What should I write down first?
Start with what you already know you know: your offer, your customers, your prices, how you talk. Fastest win. Then dig out the harder one, the things you know how to do but have never written down, like how you price a difficult client. That tacit craft is your real moat, and almost nobody bothers to get it out of their head.
Should I automate everything?
No. Automate the repetitive work, but keep a human gate on anything you cannot take back: publishing, sending, spending. Human before AI. Automation without a door is not leverage, it is a faster way to be wrong.
Do I need to be technical?
No. The deposits are written files in plain English: who you are, how you work, who you serve. If you can explain your business to a new staff member, you can build this. The writing is the work, not the coding.
Is the point to be lazy?
Lazy is the reward, not the method. You get to be lazy later because you were disciplined now. Writing down what you know is real work, it is boring, and it is the price of the peace that comes after. Hard first, then peace.

Sources

  • 复利工程是超级个体用 AI 的秘密武器 (Compounding Engineering Is the Solopreneur's Secret Weapon), 晓辉博士, Douyin, 12 July 2026. The 复利工程 frame, the three deposits (记忆 / 技能 / 偏好), the four-quadrant sort, and the 偷懒 line all come from this talk. Quotes translated from the video. (v.douyin.com)
  • Where I disagree with her is mine, not hers. She ends on being lazy, and on automating the whole flow. I hold that lazy is the reward and not the method (hard first, then peace), and that anything irreversible stays behind a human decision.

Less noise. A living map.

The Solopreneur OS is the live Sprint where we build the system that keeps paying you back, so your effort stops resetting every morning.

Join the Sprint · RM711