Vol.073 — AI Native, Japan edition — DAY 1: What took them two weeks and thirty pages took two minutes

Vol.073 — AI Native, Japan edition, DAY 1

Date: 2026-10-14 | Newsletter · AI Native, Japan edition · DAY 1 of 4


We are a Singapore AI company, and the client in this story is in Japan. That order matters to me. Most of what I read about AI adoption in Singapore is about Singapore companies catching up with somewhere else; this is the other direction — a Singapore team reshaping what the staff of a Japanese manufacturer can do with their own hands, in four two-hour sessions. I am writing it up partly for the Japanese readers who follow the series in Japanese, and partly for Singaporean owners who are still deciding whether this is worth doing. The return is not abstract. This client pays an outside agency for every change to her website because nobody in-house can write HTML. If, by the end of four sessions, that invoice stops, the training has paid for itself — and Day 1 is where you can already see whether it will.

“She was already two hours a day on ChatGPT”

The client is a Japanese cosmetics manufacturer. Two people came to the session: the CEO and one member of staff. They did not arrive through our AI diagnosis service, which is where most Japanese SMEs start. On the first discovery call it turned out the CEO already spends about two hours a day with ChatGPT. She was well past the level the diagnosis is built for; it would have taught her nothing.

So I changed tack on the call and showed her how my own AI staff work — the webmaster, the PR writer, the promotion executor, the designer, the customer-service agent — why each has a defined role, and how they all operate under one shared folder. She was most drawn to the webmaster’s work, and she signed up for the custom training on the spot.

What converted her was not an assessment. It was seeing an AI team actually operating. For a buyer who already uses AI fluently, the product is the visible operating model, not the diagnosis.

The “before”: an invoice, because nobody in-house can write HTML

Before we started I wanted one concrete cost to track across the four sessions. Hers was easy to find. She outsources web design to an agency, and the only reason she still pays is that none of her staff can write HTML. Every change to the site goes out as a job and comes back as an invoice. That recurring fee is the “before”, and it is the whole business case in one line: bring the capability in-house, and the invoice stops. Four sessions against an agency fee that recurs every time the site changes — that is the arithmetic a Singaporean owner can do for their own company before reading further.

One piece of pre-work: their site is not WordPress. It is a simple web-hosting account holding the code and files directly. For our webmaster that is easier than WordPress, not harder — which is worth knowing, because it means the approach is not tied to one platform.

The first hour was plumbing, and I will not pretend otherwise

We ran Day 1 inside her ChatGPT, not ours. They already use it, the task is simple enough, and there was no reason to make them switch tools. The first snag was that she had only ever used the browser version, without the Work functions, so step one was downloading the desktop app and switching Work mode on.

Then the part I find most interesting as a builder. I gave them the latest operating file of our own webmaster — a single claude.md — and told ChatGPT to run that webmaster inside its own environment, with the WordPress instructions stripped out, since their site is not WordPress. An AI staff member turned out to be the operating file, not the model it was written for. The same webmaster that runs on Claude for us ran on ChatGPT for them that afternoon.

We connected the browser plug-in, logged into their hosting account, and the whole first hour was gone. Account access, app mode, connections — the plumbing eats the clock before anything that looks like AI happens. At the end of the hour I asked the one question that mattered: can you see the site’s HTML and contents in the browser? The answer came back: yes, very clearly.

“Is it that easy!?!?!”

I told them to change some minor text on the site, anything they liked. They chose the CEO’s title on the About Us page. About two minutes later the webmaster replied: Complete, please check here. They clicked. The text had changed exactly as instructed. They were stunned.

Then they showed me what that change normally costs them. They keep a thirty-page PowerPoint whose only purpose is to tell their human web designer where to revise, and the designer quotes about two weeks to complete such changes — plus the invoice. In two minutes, in front of them, the thirty pages, the two weeks and the invoice collapsed into one sentence typed into a chat.

What happened next is the part I had not planned. Once they understood what the webmaster could actually do, they stopped looking at me and started looking at each other — how about that one? — hunting for the next thing. They had an Obon-holiday notice to post on the top page, a shipment-disruption announcement. They asked the webmaster directly. Another two minutes. Stunned again.

They left with something they had not come for: a webmaster that works twenty-four hours, is easy to talk to, and is theirs.

All I had to do was watch

My role on Day 1 was almost nothing. I set things up for an hour, and then I watched two people enjoy themselves.

Watching them react, I saw myself the first time I used our webmaster on my own website — the same first-contact wonder, now on someone else’s face. It was very, very refreshing to witness.

What I am watching in Days 2 to 4

Day 2 is the site audit. Day 3 is memory, communication style and working rules — the part where a tool becomes a staff member. Day 4 is reporting and the enquiry form. Two questions run through all of them. Does the outsourced web-design cost actually get retired, and by when? And does a CEO who already uses ChatGPT two hours a day shift from asking an AI to running an AI staff member? That shift is the whole story, and Day 1 only showed where it starts.