Cut through AI noise with weekly insights from Japan to AI-proof your career or business. We aim to bring you insights that are: digestible, actionable, and unique (breaking through the information firewall in Japan).

In this week’s newsletter:

I’ve been using Claude Cowork for a couple of weeks now, and I think its a genuine (small) inflection point, not just hype. Claude recently basically split their desktop interface into “Chat”, “Cowork”, and “Code”. We are familiar with the regular Chat and also Code (technical people usually use it from the terminal but you can also use it in the app GUI). The middle tab, “Cowork” is what’s new. I think its the first genuinely mainstream “AI assistant” (the recent OpenClaw is of course another “real” AI assistant, rather more powerful but also more complicated and has inherent risks to setup).

Cowork is the most accessible AI tool thats crossed the line from "assistant that talks" to "assistant that acts."

For the past few years, most basic AI workflows have been: you ask, AI responds, you copy-paste to where you need the info and you do the actual work. You could also setup workflows or “agents” using builders like N8N, Gumloop, Zapier etc for fixed actions or agentic actions with a degree of autonomy. But you needed to setup separate tool accesses and later MCP server accesses, individually or using drag and drop interfaces with multiple steps. It also needed a separate service like those mentioned, away from our top 3 LLM interfaces (ChatGPT, Claude, Gemini). Otherwise you were ultimately stuck inside a chat box - and you yourself had to be the bridge between the AI's output and the real world of your files, your browser, and your apps.

Claude Cowork breaks that loop in the simplest way thus far. It's a feature inside the Claude desktop app, and unlike the chat interface you're used to, it can access a folder on your computer, operate your browser (Chrome) like a human would, and connect directly to your software stack via MCP (Model Context Protocol). It doesn't just draft the email. It can read your calendar, find the relevant thread in Gmail, draft the email, and save a copy to your project folder.

That is, its finally a fully fledged AI assistant that acts from your instructions in your familiar Claude window. Its been quite handy for me already.

Cowork operates on what you might call a "work in a folder" model. You grant it permission to a specific working folder on your local machine, and from there it can read, organize, rename, and act on your files. It's orchestration, a loop of observe, reason, act, and repeat, just like the agents we talked about in our previous Moltbook newsletter. The difference is that this one is genuinely useful for a business owner.

A few concrete examples of what this looks like in practice:

It is now a pretty practical tool. Here are some workflows that can genuinely save you hours a week, after some minimal setup:

As you can see, these aren't AI-generated text outputs. They're operations, real ones we do every day. I use the slide creation, meeting prep, invoicing, and inbox triages every other day now, and its getting easier and easier.

These are real interesting for anyone running a small operation.

Claude Skills are reusable, saved workflows. Think of them as digital SOPs - Standard Operating Procedures - that you build once and trigger whenever you need them. Unlike a Claude Project, which is primarily about maintaining context (your brand voice, your ICP, your backstory), a skill is about executing a specific repeatable task, with the same quality each time, without re-explaining yourself.

The structure is simple: a skill.md file with a name, description, and step-by-step instructions. You can attach reference folders with your templates, brand assets, and tone-of-voice examples. Claude only loads what it needs for the task at hand, which keeps responses fast and focused.

If you're not technical, there's a built-in Skill Creator. Walk through a task manually once, then tell Claude to package it into a skill on your behalf. It writes the file, structures the logic, adds it to your library. Done.

The promise here is that a single person can start to operate with the output of a small team, without the coordination overhead. And its becoming a reality.

A few things worth keeping in mind before you go all-in.

First, this is still early. It's currently an Early Research Preview, and Mac-only for now (Windows is coming). I am sure there are kinks to work out.

Secondly, calling it "agentic" can oversell how autonomous it really is. As we covered in the Moltbook edition, agents don't “own” anything. They run under your credentials, on your infrastructure, with access you grant. That's actually great for now, but don't expect Cowork to run your business while you're asleep on day one.

Third, the real leverage comes from building your Skills library thoughtfully. The first few you create will feel a bit clunky. That's normal. Think of it like training a new hire: the upfront time pays off once the workflows are more clear and precise.

The shift from Chat to Cowork is less about a new feature and more about a new mental model. You're not prompting anymore, you're delegating. The practical question to ask yourself is "what repetitive operation is eating up my work days, and can I hand it off to AI if 95% perfect is good enough?"

For solopreneurs or small business owners specifically, the value is obvious. The administrative burden is annoying as all hell: moving around manual documents/invoice, answering endless sales emails, meeting prep in your second language if you’re a foreigner in Japan and so on. Cowork isn't a silver bullet, but it's the first tool I have found that is actually delivering on the promise of solving some of those daily problems.

Start with one skill. Run it and tweak it till its the quality you expect to see. Add another skill, and another. Before you know it you’ll have a full fledged AI employee working for you.

​Qoder AI Builder Night Tokyo is a hands-on prototyping event
where AI builders and enterprise innovation leaders come together
to explore real-world AI agent use cases.

​🔹 Hackathon Credits & Incentives

​All new Qoder users will receive 300 free credits to build during the hackathon.

​🏆 Winning teams (up to 3) will receive 2000 extended credits per team to continue building after the event.

​Existing users who would like additional credits for the hackathon can submit their Qoder ID via the event form.

​😃 Participants will form teams and rapidly build AI agents
focused on solving actual enterprise workflow challenges
— especially in robotics, manufacturing, and internal automation.

​Selected teams will have the opportunity to demo
directly to enterprise guests from Japanese companies
and innovation programs.