Which Claude should you use?

TTL 56

When I talk to people about these tools, they’re often disoriented. Honestly, fair enough: Claude Chat, Claude Code, Claude Cowork, OpenClaw… they sound similar, they overlap, and nobody explains when to use which.

I thought I’d share my actual setup and the logic behind it, maybe it could help you too.

So here’s when I use each of these tools:

✎ Claude Chat → for quick, disposable questions. “How tall is the Rockefeller Center?” “Summarize this paragraph.” “Give me 5 subject lines for this email.” You don’t need to save the output, you don’t need to build on it. Just ask, get the answer, move on.

✎ Claude Code → terminal interface. Black screen, no icons, no notifications, no mouse. You talk to it in plain English and it works on your files directly.

I use it for: writing and editing code, processing CSV files, pushing and pulling from GitHub, running data analysis and getting the output as an actual file.

I also use it as a focus mode, because there’s something about that black screen that puts me in a different mental state.

✎ Claude Cowork → same underlying capabilities as Code, but with a graphical interface. Mouse-friendly, connects natively to Google Docs, Drive, Excel, Slides.

I use this for talent reviews, one-on-ones, performance documents, company presentations, drawing information out of any documents I choose.

✱ Claude Code and Claude Cowork are honestly very similar. The main difference is that one works in the terminal, with text, and the other with a mouse and graphical apps. The output somehow reflects that: Code gives you something precise and technical, Cowork gives you something more visual.

✎ OpenClaw → this one is a little different: it’s not a chat interface, but an agent that runs on an isolated server in Helsinki and operates autonomously.

The key differences:

• It has a heartbeat: a cron job that fires every 30 minutes, reads an instruction file, and executes tasks without me doing anything
• I gave it an email address: Claw reads emails coming from me only, while ignoring and forwarding to me everything else.
• It communicates with me via Slack, Telegram, and Signal — I message it like a colleague, it messages me back when something needs my attention
• It manages Linear (project tracking), Apple Reminders, and shopping lists. I just describe the behavior I want in plain English and it figures out the rest
• It writes a daily blog post autonomously at digitalforest.blog, which I only review for privacy issues and factual errors (you’ll soon hear more on this experiment, in the meantime check it out and let me know what you think)
• It can setup a web app from a GitHub repo entirely on its own, nginx config and letsencrypt certificate included, with zero human intervention.
• Even better, it can modify its own code to improve and evolve at my demand. And we do that a lot.

---

Now let’s talk about how much I invest in these (both money-wise and energy-wise)

The total monthly cost for all of this is around €230.

€180 for Claude Max, €50 for the server (though you could get away with Openclaw as cheap as €7/month, but I had the server already).

But these prices don’t reflect real cost.

In my opinion, this is the Uber moment: subsidized, artificially cheap, designed to get you in the habit. The same setup could easily cost €2,000/month within a year, once the market catches up, energy becomes the bottleneck, and investors demand returns.

I’m not saying this to create urgency. I have nothing to gain, and I don’t hold any stock of any of the involved companies. I’m saying it because the cost of learning right now is very low, and the gap between those who figured it out early and those who didn’t risks becoming permanent once costs increase..

The best time to get your hands-on experience is now.

Let me know if this helped and if you’ve been experimenting with these or other tools!

If someone forwarded this to you, you can subscribe.

I also publish on paolo.blog and monochrome.blog.