Claude Code Lives in Five Places

I like AI chats. They are convenient and simple. They live in one place, syncs across every device, and works the same everywhere. I learned Claude Code isn’t like that. It can live in five different places, each with its own powers and trade-offs, and you might have to pick depending on what you’re doing.

That was hard to comprehend. Today I’d like to share what I’ve learned about this beast.

Same engine, five hats: the CLI (your terminal), an IDE extension, the Desktop app, the Web, and Mobile. Each one is tuned for a different way of working.

Here’s the one-glance version. The detail is below it.

Installing anything into the terminal is the most confusing thing for a non-tech person. Long, weird commands. You feel like something is hacking your laptop. Definitely doesn’t feel simple.

One or two weeks later, it flipped. Now every other surface feels a little powerless next to it.

I think it’s because the terminal is the rawest, most native way to talk to a computer. Put your AI agent there and it gets direct access to pretty much everything - any tool, any file, any function on your machine (if you let it read of course). Just two basic commands (cd to move, ls to look around) and it stops being scary. Leran more and it starts being your edge over people who only ever touch the friendly buttons.

Ease: mid. Most developers would say easy. For the rest of us it feels low on day one but improves once you get used to it.

Power: GOAT. If I had to keep one version of Claude Code and delete the rest, it’s this. Fewest limitations, first to get every new feature.

What’s unique here:

Cons: The one real downside is there’s no visual map - you won’t see a tidy panel of “here are your connectors, pick what to add.” It’s mostly text and commands. No icons ☹ You have to be curious and patient enough to go exploring.

Best for: honestly, everything. Coding, productivity, building little skills, scheduled automations. The most versatile hat by far.

This is the version most of you have already, or have already tried. It’s just another tab in the Claude app next to Chat and Cowork.

Ease: high. Everything is structured and visible.

Power: high. It lives on your laptop, so it reaches both your local files and the cloud, and it has every connector - which you add visually, the exact thing the CLI makes you work for. It also has computer use: it can open apps, take live screenshots, and click buttons on your behalf (the CLI can do this too on a Mac, but here you watch it happen).

What’s unique here: routines. Setting up scheduled tasks is native and easy in the Desktop app. It’s basically a prompt that is being lauched on your laptop or in the cloud on schedule. My two recent examples:

Cons: it’s heavier. I can feel my laptop slow down and heating up during active sessions. And it’s a lot in one place - chat and Cowork and Claude Code means I sometimes lose track of where a thing actually lives. A Cowork scheduled task is stored somewhere completely different on a laptop from a Claude Code one.

Best for: your very first try. If you already use the chat app, this has the smallest friction to dip a toe in.

The web version is pure cloud. Open it in a browser and it has no access to your local files or your machine - which sounds like a downside until you realise that’s the whole point.

It keeps running after you close the laptop. Anthropic’s cloud does the work, and it reaches cloud things through connectors - a GitHub repo, your Google Drive documents - reads them, structures them, acts on them.

I use it if I need to test if some process can work entirely without my laptop in the loop (e.g., leveraging only GitHub, Supabase or other connectors)

Ease: high. It’s a familiar website.

Power: mid. The cloud-only ceiling by design.

Cons: no local files. It can’t touch the documents or the project sitting on your laptop.

Best for: cloud routines that should fire whether your laptop is on or off, triggered by Anthropic’s cloud. Drafting replies in your Gmail, sorting email, checking whether a new lead landed in a spreadsheet or someone answered your Google form. The daily and hourly babysitting jobs.

If you start poking at code, you’ll download an editor - VS Code, Cursor, maybe a JetBrains one. Claude Code can live inside it as an extension.

In my experience it’s the closest thing to the CLI, because it basically spawns a terminal session but wraps it in a friendly interface. Best of both worlds, and much easier to set up.

My wife is about as far from code as a person can be. She tried Claude Code, loved exploring different ways of working, and the Desktop app was her most convenient entry point. But the moment she started her own vibe-code project, the IDE extension became her go-to - comfortable UI to talk to the agent, and she could still see all the files and how they connect.

Ease: high.

Power: high. It carries most of the CLI’s muscle

What’s unique here: you’re inside the codebase. You watch how files interconnect while a convenient agent panel sits right beside them. That’s the learning, not just the doing.

Cons: Not convenient for juggling lots of cross-project windows from one place.

Best for: your first focused vibe-coding session, when you want to stay in control of the codebase and actually learn how the pieces fit - without committing fully to the terminal yet.

A couple of months ago I’d have called mobile barely usable. Small tasks only, a permission prompt at every step. Powerless.

Recently I learned about Remote Control, and it changed the scene completely. You type a prompt on your phone, and it runs on a session living somewhere else - CLI, desktop app, the web. Your phone is just the keyboard; the work happens on the powerful machine. That removes the phone as the bottleneck.

Ease: high. There are so few buttons it physically can’t be anything else.

Power: mid – requires you to set up remote-control to boost it up to high.

What’s unique here:

Cons: if you forget to set Remote Control up, mobile slides right back to way less useful.

Best for: when you leave your desktop at home on strong Wi-Fi and travel somewhere with thin edge signal. You send a tiny text prompt - light enough to slip through a weak connection - it tunnels to your laptop, and the laptop starts cooking. Sending a small signal beats sending none.

I dare you to install Claude Code in your terminal and run one session. Try the developer hat for one afternoon and it might rewire how you see all of this.

Start here

Fun fact. I watched the Claude Code sessions Anthropic ran in San Francisco and London. The rooms were full of hardcore developers. And the staff engineer presenting the new Desktop app features had to almost beg them:

For developers, you have to sell the friendly app. For the rest of us, the terminal is the scary ask. Same tool, opposite fears.

A pivotal week for Twitter Screenshot - my first vibe-coded project. We’ve reached 400 users and the growth rate is insane compared to 9 months ago.

It’s the clearest proof yet that the slow, boring SEO investment compounds over a long distance.

Seeing hundreds of people use a free tool every single day gives me a very specific kind of joy - the “this actually helped someone” feeling. That niche little value is the whole point I enjoy building things.