I attended a live lecture by DeNA President Tomoko Namba—one that isn't on YouTube. The most shocking part was her current work environment. It centers around Claude, and she even said, "I can never go back to the days when I was gathering information in NotebookLM myself."
Emails, chats, schedules, notes, Drive, internal Wikis—all work information is connected to AI, with Claude at the center.
Whether it's researching someone she's meeting for the first time or remembering past meetings, the AI proactively provides information before she even asks. It's no longer a "tool-using" posture; it's a way of working where she has a command center—an alter ego.
As I listened, I felt restless. This was very similar to how I usually work.
So today, along with the stories that resonated from the lecture, I want to share how I design my environment, what instructions I use, and how to replicate this way of working.
The Meaning of "I Can't Go Back to NotebookLM"
A year ago, Ms. Namba was a "chat master." She would research first-time contacts on Perplexity, bundle them in NotebookLM, and ask questions while in a taxi. That "Namba-style AI technique" spread across Japan.
Now, she says she can't go back. After returning home, I checked public information and found she said the same thing in other lectures.
"I feel nostalgic for the days when I used to copy various things into NotebookLM myself."
This isn't just lip service for the lecture; it's her official current status. So, what changed? She summed it up in one sentence during a public lecture:
"AI agents have undergone a mega-evolution."
The evolution of agents has completely changed how AI is used. In short, it's the difference between "carrying information yourself" and "AI going to get the information."
*NotebookLM (Gemini Notebook) is also a great tool.
Before:
You research, copy, give it to the AI, and get an answer. The AI is smart, but you are the transporter.
After:
AI is connected to your work environment (email, schedule, notes, Drive, Wiki) and goes to get the necessary information itself. The transporter disappears.
What happens in the "After" world?
Imagine my typical weekday:
- In the morning, Claude looks at my calendar and prepares research and simulated Q&A for today's business meeting in advance.
- In my inbox, draft replies are automatically created based on past interactions. I just check and send.
- Meeting notes are automatically recorded, and by the time it's over, a draft thank-you email and To-Dos for next time are ready.
- Those notes are accumulated and called up as "personal knowledge" when a similar project comes up.
- It refers to the internal Wiki (Notion in my case) and creates documents following company rules and formats.
This is the actual configuration I run.
The command center (thinking side) is Claude, and the alter ego (remembering side) is Notion. I don't know Ms. Namba's exact tool configuration, but the workflow I saw in the lecture was almost exactly this.
The point is that none of these are "requested on the spot."
Once designed, it keeps running on its own. I believe this is what Ms. Namba means by a command center or alter ego.
[How to Build] Connect, Remember, Schedule
The mechanism looks massive, but it's not that difficult.
Even as a non-engineer, I did it easily. I only did these three things:
① Connect (MCP)
Make it so the AI can reach your work information on its own. Calendar, email, Drive, chat. Now, with MCP (a common standard for connecting AI and external tools), almost all major work tools can be connected. Start with just one. Every time you eliminate "copy-pasting," the chat gets closer to being a command center.
② Make it Remember (CLAUDE.md / Notion)
Place an "Instruction Manual for AI" that describes your work rules, preferences, and "don'ts" in a place the AI reads every time. For Claude Code, the "CLAUDE.md" instruction file is that place, and I also use Notion as the main body of memory (the "alter ego" in Image 2). The criterion is simple: if you write the same instruction twice, it's a sign to write it into the environment.
③ Schedule (Routines)
Finally, set up routines that run automatically at fixed times or under certain conditions. This completes the "runs without being asked" part.
For example, what I actually run:
- Weekday morning: Look at today's business schedule from the calendar, research the partner, and create a hypothesis slide.
- Upon receiving an email: Detect unreplied messages and prepare a draft reply based on past interactions.
- Friday evening: Summarize learnings from this week's sales/meeting notes into 3 lines and transcribe them into report materials.
Having a brief "ready" first thing in the morning isn't magic; it's just a schedule setting.
(I decide the safety boundaries first, such as "do not perform external sending," etc.)
As for the prompts to build this mechanism...
Actually, this way of working hardly uses "clever prompts." This is because it's a system where you don't have to write instructions every time. You build skills (packages of frequently used procedures) and routines in natural language. It's not difficult at all. What I always keep is the "Instruction Manual for AI," and since it would be too long, I'll post a mini version here as a sample.
1AI Instruction Manual (Mini Sample)23## About Me4- Role: IT Sales (Major clients are large corporations, manufacturing industry)5- Values: Quality and speed. Reply within the same day.67## Writing Style8- Email = Brief. Conclusion first. No emojis, "!" is OK.9- Internal Chat = Casual. Use bullet points.10- Materials = One message per slide.1112## Regular Work (Automatic routines)13- Pre-meeting brief: 1. Partner news 2. Past interaction summary 3. Hypothesis slide.14- Post-meeting: 1. Format notes 2. Draft thank-you email 3. Extract To-Dos.15- Weekly review: Summarize learnings into 3 lines for the weekly report.1617## If in Doubt18- If info is missing, ask all questions together before starting.19- Do not fabricate numbers. Mark unknown parts as "*To be confirmed."2021## Prohibitions22- Do not output customer names or amounts to external tools.23- Do not perform email sending.
⚠️ This is just a sample mini-version to convey the image.
If you want to actually try it, I've summarized the full version (design procedures for 5 scenes, how to set up routines, and instruction texts) in a Note, so please check that.
There is a link to the Note at the end of the article.
Planning Documents Disappear: "Show me the Claude"
At DeNA, the concept of a planning document is changing, and people are reportedly saying, "Show me the Claude."
"Paper planning documents often lead to discrepancies in mutual images, but having a prototype is completely different; images can be shared instantly."
From an era of persuading with words to an era of showing something that moves. This is something individuals can imitate starting today. For proposals or consultations, before PowerPoint, first a moving prototype.
When I bring a new proposal to a client, I create a prototype myself and go show it. The reaction changes.
The Trap of Free Time... "Move Ahead Decisively"
Another thing that resonated was the talk about time.
What do people do when they have free time due to AI efficiency?
According to Ms. Namba, the more serious a person is, the more they fill that free time with "work they couldn't reach before (which, in a sense, they didn't have to do)."
That's why they stay busy forever. In the lecture, she said:
"Japanese people are serious... oh, I shouldn't say things like that, haha,"
while making the audience laugh, she talked about asking employees, "Please don't do things you don't have to do." In public lectures, she says the same thing with even stronger words:
"No one will ever say on their own, 'People have become available.' So, just move them ahead decisively."
Don't use the time saved for more of the same work. Move to new work first. For individuals, the practical form is to reserve the time saved by AI for "new things you've always wanted to do" in advance.
Conclusion
Thank you for reading.
Ms. Namba's way of working: Connect, Remember, Schedule.
This is something individuals can build starting today. (Of course, some of it is within the realm of imagination.)
《 For those who want to actually build the mechanism 》
I've summarized the overall picture of my environment and the design procedures for reproducing it—including detailed steps for 5 scenes, how to set up routines, and initial prompts—in a Note 👇
https://note.com/uchita_success/n/n5bc96a63047e
(I'd be happy if you could 'Like' the Note too!)
I will continue to share information on work techniques using AI agents, AI x document creation, and sales and back-office automation using Claude Code. Thank you very much.
