FreeBSD Users: We Need to Talk About Claude Code

March 4, 2026 - 7 minutes read - 1370 words

Recently, Anthropic changed their Claude Code installer from using npm to a native installer. It works on Linux/WSL. It works on Windows. It works on Mac.

On FreeBSD today, you get a scary yellow deprecation warning suggesting that you and your platform might be a dead end. I think that’s a problem.

History is littered with tasteful products made by artisans who gave the high-hat to unwashed technologies and their users:

MULTICS / Unix
Apple Macintosh Look-and-Feel / Windows
Metallica / Music file-sharing
Sony / iPod
et al.

In every one of these cases, worse/grubby/philosophically-impure/commercial/lazier/legally-questionable/scrappier won.

We need to meet users where they are. As the classic demotivational poster had it for phone support desks (of which I am a proud veteran): If We Don’t Take Care of the Customer, Maybe They’ll Stop Bugging Us.

How I Got Here

Questions about FreeBSD

When I first sought to move from macOS/Linux to FreeBSD, AI web chat-bots were incredibly helpful for the exact kind of questions that don’t have clean answers in any handbook:

“What’s the FreeBSD equivalent of X on Linux?”
“Where should I install a custom X in FreeBSD, in /etc or /usr/local?”
“What’s the syntax to configure a jail with X parameters?”
“Explain what a zfs dataset is.”

That’s real value. It’s not novel generated code. It’s not a better sort than quicksort. But anyone who’s done any IT work will recognize that these questions are a particular type of time suck.

Questions about FreeBSD Hardware Support

Given FreeBSD’s hit-or-miss support for various wireless/video cards, questions like:

“Given this error message from this wireless card, can I fix it?”
“Given this error message from my video card, can I fix it?”

dogged me. Answers here allowed me to cut my losses and return 3 laptops before settling on the Thinkpad 14 I’m typing on now (and there’s still a wireless issue with it). But I wouldn’t have been able to diagnose that the effort wasn’t going to pan out without the web chat diagnostics coaches. That’s real value.

In both these cases, my chats helped me build better information to provide in forums (topic 1, topic 2). My post showed the proper tests and legwork and saved others time and helped me keep moving forward in a way that I wanted to stick with FreeBSD. That’s real value.

Once I had the OS settled, I installed NSCDE — a modern reimplementation of the classic Solaris CDE look and feel built atop FVWM. And that was a hairy bit of bastardization in the name of retrocomputing cosplay. But I could ask AI questions about dated technology: dithering, XPM files, etc. That’s real value.

Porting NEXTSPACE to FreeBSD

Then I took on a more ambitious project: porting NEXTSPACE — a recreation of the NeXTSTEP desktop — from Linux to FreeBSD. I’ve written about this in detail in A 2025 Obsession: NeXTSTEP, OpenSTEP, and NEXTSPACE and Porting NEXTSPACE to FreeBSD.

I’m reasonably comfortable in shell scripting, and I had some macOS 10.4-era experience with Cocoa and Objective-C. But I was far less confident with NEXTSPACE’s GNU Makefiles, its expert-level C and Objective-C, and GNUstep. Through the summer of 2025, I was using ChatGPT and Claude (web) to navigate the gaps. And then my productivity jumped to lightspeed with Claude Code.

The UX breakthrough that Claude Code represented accelerated my work meaningfully. I could point to files. I could pipe shell output back in. I could iterate rapidly without losing context. Through this workflow, I was able to port a NeXT-era desktop to FreeBSD successfully. Maybe the world isn’t going to beat a path to my door to use a 1989 desktop — but I love it.

Isn’t that why we got into FOSS in the first place? AI is not per se inimical to those goals. It gave me real value.

AI Extended My Zone of Proximal Development

I’ve written more extensively about this framing in The Positive Case for AI-Assisted Development, but the short version is this: AI extended my zone of proximal development — the gap between what I could do alone and what I could accomplish with a capable collaborator. I want to keep using it on FreeBSD to make more FreeBSD things.

A Warning and a Whimper

And here we come back to the rub. While I had been using Claude Code as an npm package, I now get an annoying deprecation warning about a “native installer.” Maybe this will be an annoy-o-gram forever, but maybe not. If support is going to stop, it’s being telegraphed here.

So I opened an issue to get FreeBSD supported. Three thumbs-down on the bot’s closure; three thumbs-up on my description; and then… another bot-close.

We need the posse. We need the crowd. We need a mob. But I’m afraid the skilled folks of FreeBSD land view it as beneath them to get AI tooling on the platform. And I think that’s a path unto irrelevance.

The Stakes Are Real

I hope I’ve convinced you that AI does help people create things that enrich our platform. Even if it never creates a new algorithm, the incidental toil — incidental IT plumbing — has often enough been avoided to suggest that it’s generating real value: rewrite commit history to pull out git-trapped binaries, help me author posts, check for broken URLs.

And yes, I’ll grant that AI represents a brontosaurus-burning sustainability problem. And AI represents an intellectual property immolation event. And that our civilization is unprepared for the impact of mass adoption of this technology. These are legitimate concerns and I don’t dismiss them.

But it’s incorrect to dismiss the users and their desire for this tool. Like it or no, a generation of teens, our future programmers / engineers / FreeBSD users are growing up using AI tools. When they see Linux/Win/Mac + AI tools OR FreeBSD and high-hat from its community, I think they’re going to choose a platform that enables them to say “I did it!” No doubt, FreeBSD machines will hum in server farms, but for an end user, our best hope is that they tolerate FreeBSD as a learning platform (like Minix): great example, great docs, ’til next semester.

Remember, this is the second year of the FreeBSD desktop, after all; I think we ought to be pragmatic.

And so here we are, the fact that my issue was closed with a whimper not a pots-and-pans-clattering bang says something about our community: We’re not there. We’re not fighting for the users. Six emoji clicks on GitHub is not a full-throated defense.

A Word to the FreeBSD Faithful

Many of FreeBSD’s supporters are some of the best programmers on earth. That’s great.

But when we publicly and repeatedly signal that AI is beneath us, we risk damning our own future. As Linux becomes increasingly riddled with corporate interests — which was, frankly, a major factor in my own migration — FreeBSD stands to become a more attractive platform for serious developers. That’s a real opportunity. Let’s not squander it with reflexive hostility toward tools that a whole generation of developers depends on.

I’ve been dabbling with BSD since 1998, when a BSD advocate made the pitch to me — a Slackware devotee — and I didn’t jump ship immediately but never forgot the impression, elegance, and focus on security. After years of sysadminning on Solaris, my respect for the project only deepened. I understand and share the purity of vision that makes FreeBSD what it is.

I’m not arguing against hand-crafted code. I love hand-crafted code. I’m not arguing against expertise. I love expertise. But there are ports whose Makefiles could use attention, sysadmins who want an AI tutor, and folks who got AI to build them a needed driver. This is real.

What You Can Do

If you’re a FreeBSD user and you’ve found AI tools useful — or you understand why platform support matters — please weigh in on GitHub issue #30640. Show Anthropic that FreeBSD is a platform of taste-makers and decision-makers that deserves their attention. A closed issue with community support is a much stronger signal than silence.

FreeBSD has survived and thrived by being excellent. Let’s also make sure it’s visible, audible, and welcoming.