Beej's take

Beej's guides were foundational in forming my early life as a hacker (both in the common and silly valley sense), so naturally when he wrote an article on the subject I was bound to read it. It's titled On Making, and kind of summarizes where I'm at.

The fact is that I do enjoy programming and hacking computer problems; I have well-worn copies of the Architecture of Symbolic Computers, Let Over Lambda, and Hacker's Delight. Of late, however, programming is less hacking interesting actual problems and more doing science on someone else's code, figuring what poor life choices the two of us made that led us here. I have, to borrow from Alan Kay, no interest in dicking around in Linux anymore.

This coincides much with work, where by and large the calibre of my colleagues is somewhere in the gutter. Most of what I do is designing things and coaxing other engineers to build it. So already, I write specifications, documentation, and architecture documents, hand these off to engineers of dubious quality (some are great), then review plans and code generated by inscrutable minds [2].

Now, work has me fully burned out on dealing with technical problems. Any possible concern I have for tech problems is largely beat out of me on a regular basis. Occasionally, I get a fever pitch of working on some robotics or drone code, but the hardware development side has a high overhead and time cost. Lately, I've been pursuing a more contemplative, intellectual life along the Dominican tradition, and that necessarily involves less time staring at screens, leaking my soul into the algorithm.

That being said, there are tools and systems that I still want to build. Background computing, as it were. Trying to get friends involved never leads to a line of code being written, so getting humans involved in this is a dead end.

On this front, claude-core makes sense. It's the natural extension of the skills I've already been working on, and I have reliable "engineers" as it were working on it. Of course, I have to pay more for them, but the code actually gets written, unless it fell to me.

A case in point is kge. I wanted to port ke to C++ and add an ImGui frontend to it. I've had fits and starts in the last five years working on it. At some point I got a basic shell of a thing started. But as much as I hate web dev, I hate GUI dev just the same. I don't like figuring out layouts and UI quirks and all of that.

So I took my pile of notes (here's what a PieceTable looks like, or a syntax highlighting engine, or ...), and passed them piecemeal into claude-core. Some tens of thousands of lines of C++ later, I have an editor I use daily. It's not perfect, there's things I want to change. But within a weekend I had a functioning editor. Four months later, I'm pretty happy with it---and I daresay it's much further along than I would have gotten without wasting countless hours on quibbling with code.

However, I have put my time in already quibbling with code, bit banging and wrangling operating systems. I somewhat suspect that's why I am where I am, which is not an ideal place, and certainly makes me less likely to spend more time on tech-specific things. I have that going for me.

So I basically pay a couple hundred dollars a year to have a small team write code for me that I have put thought and care into designing, and the end product is tangible results. I spend the rest of the time doing things I actually care about, which is no longer the mechanics of writing code.

And, with the right nudging, just like with my work team, I get code that is arguably better quality than what I would have hacked together, only now I spend less time mired in the minutiæ of quotidian engineering drudgery. At the end of the day, I have actual work in hand that I can use daily, rather than a series of pages in my dev notebook with no code, much less deployed code, to show for it.

On new laptop

Anyways, I recently bought a new Framework 12 to replace the X230; 1T SSD with 16G of RAM. I like the X230, but it's a Nitropad with heads, which means I can only realistically use Ubuntu with it. I would like to have Nix, I'd like a no-frills machine to run emacs + org-mode on, and now with the agentic coding stuff (e.g. Jetbrains' Air, and the claude/junie CLIs), I don't need to worry so much about the IDEs I had been using and can get by with less. I'd say claude-core development is as much a reason why I picked this laptop with its specs as anything else. The X230 has 8G of RAM, which is a little light for testing containers and running some other stuff, but it's less than the 64G of memory my other dev laptops have used (or the 128G of my MBP).

I haven't found a good Thinkpad that was more compelling than the Framework. I've tried, buying an X1 on eBay and the Nitropad X230, for example.

On hype

I worked in San Francisco for a number of years at An Internet Company™. I am aware and intensely wary of The Hype Machine™. I am constantly assaulted by this or that crypto thing, formal verification, BEAM, blockchain, etc. AI is hyped. No doubt. I am sick of hearing about it [3]. I was a hater; when I finally started poking at it at the recommendation of colleagues in March 2024, I wasn't impressed.

However, as of some point late last year [4], it started to become unreasonably effective. Despite a fog of hype (c.f. the moltbook incident), it is still an unreasonably useful tool all things considered.

Now there's a strain of people who would never pay for software and insist on open source. I've been using *nix systems since the age of 13 or 14 or so (over two decades), an emacser, etc. However, when I find software useful, I am happy to pay [5]; I pay for the Jetbrains suite now, as it is.

I know there are complaints about sustainability and environmental friendliness. My stance on this remains that electric cars and China are orders of magnitude worse, and until anyone starts putting pressure on them to change, nothing I do matters and that is being charitable assuming they're right. There are certainly sociopolitical questions about the data centers, which I don't have strong feelings on.

All this to say: I don't mind paying for the AI subscription.

More on the subject

• AI is great at writing code. It's terrible at making decisions.
• On Making
• The AI Vampire