Out of the GUI and into the CLI

Like a lot of people in early 2026, I’ve done some serious soul searching when it comes to how much I’m spending on the myriad services that make life worth living.

The biggest experiment already is underway — we’ve “paused” our $80-a-month subscription to YouTube TV. It now costs more than twice as much as it did when we first subscribed years ago, and the number of hours we’re actually around to watch it is down to maybe 4 a day. That math just isn’t mathing anymore. More on that at a later date.

But the biggest change over the past six months or so has been my shift from using the Claude app (desktop or web, really) and relying heavily on Claude Code in a command line interface, or CLI.

I don’t mind admitting that it took a little time for my brain to grok the difference. But so many more doors have been opened. I’m still on the $20-a-month Pro plan, which usually is plenty. I only occasionally run into rate limits, depending on what I’m working on. But that’s just a reminder to take a break and go do something else for a while.

On my Mac, I run Claude Code inside an OrbStack container — for safety reasons. This is my main machine, and like too many folks who are out a little far over their skis, I have a habit of hitting that “Proceed” command a little too quickly sometimes — though I have forced myself to at least glance at what's about to be run. So I fire up OrbStack, and then VS Code, and Claude Code is constrained to a single directory on this computer, with my various projects in there. There are a few edge cases when it'd be nice to have it less constrained, but nothing that's a deal-breaker. In those instances I just hop into a traditional terminal outside of the container.

I also installed Claude Code on my main Raspberry Pi. This one’s working without a net — it has access to the entire file structure. But there’s nothing on that Pi that can’t be recreated with a few installation commands, so I’m less uncomfortable with that. I can have Claude Code on my Mac control the Pi directly if I want, or I can just SSH in myself and run Claude Code directly on the Pi, which is a little less clunky.

It’s been fascinating to learn all these different interactions, especially when you through the “claude.md” instructions and auto-memory into the mix. Think of the former like a contextual document for each project, and the latter . I also have a claude.md file that lives over the entire container-constrained workspace so Claude knows the basics of how I’m working, where various credentials live, my basic workflow, etc.

I'm still not a coder or a developer, though. Don't even think about calling me that.

And I haven’t gotten anywhere close to scratching the surface of MCP servers. Baby steps.