Making

TLDR: I gain a lot of fulfillment by making things. I don't consider things built by others at my request to be made by me, and are therefore much less fulfilling. And then I feel sad. This article starts strong and then heads off into the weeds.

There have been a lot of pieces written about what I'll call "the AI dev schism" And I think there's a lot of truth to those:

We'll just grant those as being correct for various developers. But there's something else that troubles me.

Backstory before we get going, so you can get a better idea of my perspective:

I'm going to use "AI" to mean "Generative AI and LLMs" in this essay. Sorry, veterans of so many AI winters.

Before we begin, I'd like to share with you a bit of my latest sci-fi novel. Some of you might unaware that, in addition to Beej's Guides , I also write science fiction.

Kael pressed his back against the shattered bulkhead, plasma scoring the air centimeters from his face. The Vorrkai assault drones had anticipated their route through the lower decks and now Rin was bleeding through her jacket sleeve and old Maret couldn't stop coughing from the vented coolant still hazing the corridor. Kael counted the pulse-intervals between shots. Three seconds. Maybe four. That was all the universe was offering him.

Then he saw it: the maintenance shaft behind the collapsed generator housing, its grate blown half-open by the same explosion that had caved in their original exit. It was tight. It was ugly. It ran directly over the Vorrkai's forward position, which was either the most dangerous path imaginable or the last one they'd ever think to watch. Kael grabbed Maret's collar and pointed without a word. The old man's eyes went wide, then hard. He nodded. Rin was already moving.

Kael came last, returning fire blind around the bulkhead corner, not to hit anything, just to make noise, and to give the drones something thermal to track while his people scrambled into the dark. A bolt caught the generator housing and the whole structure groaned, raining sparks down into the shaft on top of them. He hauled himself in, knees burning on the torn metal, and pulled the grate closed behind him with a sound he was certain every Vorrkai unit on the deck had heard. In the black ahead, Rin's hand found his wrist. Move, her grip said. Now. And so they did.

—Excerpt from The Vorrkai Interval , by Brian "Beej Jorgensen" Hall

And, in my now-copious spare time I make art! This is a woodcut, painted in pastels, showing some of my favorite subjects.

Mirrors of the Machine by Brian "Beej Jorgensen" Hall, $1300.

Carpentry? You bet I dabble! I rebuilt my front deck recently. The old one was rotting out, so I grabbed a bunch of cedar and put it together. I'd been meaning to do it for a while, but couldn't find the time.

And, finally, here's some of the code I wrote for a TUI adventure roguelike:

fn try_move(&mut self, dx: i32, dy: i32) { let nx = self.player.x + dx; let ny = self.player.y + dy; // Check for monster combat if let Some(idx) = self.world.monster_at(nx, ny) { let result = { let monster = &mut self.world.monsters[idx]; resolve_combat(&mut self.player, monster, &mut self.rng) }; self.messages.push_many(result.messages); if result.monster_defeated { let monster = &self.world.monsters[idx]; let xp = monster.xp_reward; let gold = monster.gold_reward; self.player.xp += xp; self.player.gold += gold; if gold > 0 { self.messages.push(format!("You find {} gold!", gold)); } if self.player.try_level_up() { self.messages.push(format!( "Level up! You are now level {}!", self.player.level )); } } if !self.player.is_alive() { self.messages.push("You have been slain! Rest in peace..."); } self.advance_turn(); return; } // Check terrain passability if self.world.is_passable(nx, ny) { self.player.x = nx; self.player.y = ny; self.advance_turn(); } else { let terrain = self.world.terrain_at(nx, ny); self.messages.push(format!("The {} blocks your path.", terrain.name())); } } I'm an extremely prolific polymath, I'm sure you'd agree!

I don't like lying. And yet I feel, dear reader, I have misled you. Yes, all that has been created (including my deck) and I was the initiator of all that creation. But I don't really feel like I made any of it. I'm uncomfortable claiming that I did so.

Since you are certainly aware by now that all of the above is AI-generated (except my deck, which was created by skilled, paid craftsmen), perhaps you feel a little bit of discomfort with me claiming credit for doing those things, too.

However, I don't think everyone feels this way. I know many people who ask contractors to build things and they phrase it like they built it.

"I put in a new front deck," they'd say, even though other people did all the work. Personally, I feel that's misleading. I'm more of a "I had a new front deck put in" kind of person.

And when I do have Claude create something for me, I just can't say that I made it. Other people can, but I just can't. Again, I'm more prone to say, "I had this code built for me." I don't even feel comfortable MIT-licensing that (not-for-hire) work, if that's even legally possible. I just Unlicense it all.

As a manager, I'd never say that I built a product. "My team built this," I'd say. And as a manager of LLMs: "My Agents built this."

And that, for me, has very little weight in terms of making .

I don't feel like I did anything. And I like doing things. I find pride in doing things.

Completing projects is great. I love completing projects. Capitalists love completing projects. Real artists ship.

But having others complete projects I initiated is entirely less fulfilling to me.

It's not just the loss of the craft and the problem-solving challenge and whatever else. It's the loss of making .

My wife wanted a no-frills flash card system for learning Spanish. "I just want a thing where I can put the words I want in a spreadsheet and then see it on flash cards." A prompt!

So I wrote it. By hand. I did use Claude to learn some basics, like the easiest way to get the data out of a Google Sheet (spoiler: it's the CSV endpoint), but I told it to generate no code.

-------------------------------------------- Language files code -------------------------------------------- JavaScript 2 112 CSS 1 33 HTML 1 32 -------------------------------------------- SUM: 4 177 -------------------------------------------- Didn't take long. Only about 50x longer than it would have taken Claude to do it.

But I can put my name on that code and say that I made it. Was it a lot of code? No. Was it groundbreaking and amazing? Certainly not. But I'm infinitely more proud of that code than anything I've had Claude write, because I'm not capable of being proud of the latter.

And my wife wouldn't go to her book club and say, "I wrote a flash card system to study Spanish." Admittedly, part of this would be because she didn't want to appear a geek, but mostly it's because it's untrue, even though she initiated the process.

After all, you create the prompts, don't you? You said you were proud of doing things. Isn't that doing a thing? And since so much got done, isn't it even more of doing a thing?

I don't disagree. And I do agree that there is skill here in some important ways.

Not all prompts are equally effective. Not all users of AI are as effective as one another. There's a very human contribution to be made here.

But the skill is in effectively asking someone to make something for you.

"Leadership is the art of getting someone else to do something you want done because he wants to do it."

And I'm the kind of person who really misses the making of software. And prompting for software, to me, isn't the same as making the software. It's the same as asking someone else to make it.

For any given C or Rust program, did I write that program? I wrote it in C. I didn't write any machine code. The computer wrote that.

There's something different in my mind about that. I'm not sure if it's the mathematically-precise relationship between the C and the machine code...? And yet that's not spelled out. I'm relying on some hacker's deep knowledge of the CPU to generate that magnificent set of optimized instructions when I stupidly divide something by seven . And it's not deterministic, strictly, in that Clang and GCC emit different instructions. What about different platforms? The C is portable and the machine code is not. Did I write that program that's running on Windows even though I personally only ever built it on Linux?

Going further in the murky gray, what if I instruct an LLM to be a C compiler and emit assembly code for some platform?

Convert this C program into x86_64 assembly for Linux. #include <stdio.h> int fib(int n) { if (n <= 1) return n; return fib(n-1) + fib(n-2); } int main(void) { for (int i = 0; i < 10; i++) printf("%d: %d\n", i, fib(i)); } Claude says:

# fib.s - Recursive Fibonacci, x86_64 Linux, AT&T syntax # Build: gcc fib.s -o fib -no-pie .section .rodata fmt: .string "%d: %d\n" .section .text .globl main # ┌─────────────────────────────────────────────────────┐ # │ int fib(int n) n in: %edi / return: %eax │ # │ Stack frame: │ # │ %rbp-8 = saved %rbx (via push) │ # │ %rbp-12 = n (spilled so it survives both calls) │ # └─────────────────────────────────────────────────────┘ fib: pushq %rbp movq %rsp, %rbp pushq %rbx # callee-saved scratch subq $8, %rsp # 1 local (int n); keeps stack 16-byte aligned movl %edi, -12(%rbp) # spill n to stack cmpl $1, %edi jle .Lbase # n <= 1 → return n # --- fib(n-1) --- movl -12(%rbp), %edi subl $1, %edi call fib movl %eax, %ebx # %ebx = fib(n-1) # --- fib(n-2) --- movl -12(%rbp), %edi # reload n (call clobbered %edi) subl $2, %edi call fib # %eax = fib(n-2) addl %ebx, %eax # return fib(n-1) + fib(n-2) jmp .Lret .Lbase: movl %edi, %eax # return n .Lret: addq $8, %rsp popq %rbx popq %rbp ret # ┌─────────────────────────────────────────────────────┐ # │ int main(void) │ # │ Stack frame: │ # │ %rbp-4 = int i (loop counter) │ # └─────────────────────────────────────────────────────┘ main: pushq %rbp movq %rsp, %rbp subq $16, %rsp # locals + alignment movl $0, -4(%rbp) # i = 0 .Lloop: cmpl $10, -4(%rbp) jge .Ldone # i >= 10 → exit # fib(i) movl -4(%rbp), %edi call fib # printf("%d: %d\n", i, fib(i)) movl -4(%rbp), %esi # arg2 = i movl %eax, %edx # arg3 = fib(i) leaq fmt(%rip), %rdi # arg1 = fmt (RIP-relative) xorl %eax, %eax # AL=0: no XMM args to printf call printf addl $1, -4(%rbp) # i++ jmp .Lloop .Ldone: xorl %eax, %eax # return 0 leave ret OK, clanker, but does it work?

% gcc fib.s -o fib -no-pie % ./fib 0: 0 1: 1 2: 1 3: 2 4: 3 5: 5 6: 8 7: 13 8: 21 9: 34 Well, holy golden ratios! It even used XOR to set exit status 0 , the cheeky devil!

So did I write that program? Yes. I feel I did. (I actually did hand-code the Fibonacci thing just now—I've done it a million times for classes.) I feel I wrote the program even though I didn't write the assembly, which is the program that actually runs! The LLM wrote that!

Technically, it's the machine code that runs, but going from assembly to machine code now we're really getting into the land of strict, brainless, mathematically-precise translation, which I feel doesn't take away from the act of making any more than using a hammer to drive nails does.

What's the real difference in that gray area? I'm not sure. I think it hinges on something that has to do with making versus asking to be made , even if that line isn't particularly clear.

—US Supreme Court Justice Potter Stewart

Things that I ask others to do at my command (inspired by a nice capitalistic investment):

When I have ChatGPT create a painting, I don't feel like I painted it.

When I use the hammer, I don't feel like I'm asking the hammer to do something for me. Same with a C compiler or assembler.

And yet I just had Claude compile some C code right up there. Was that me writing C code and running it? Or was it asking something to write software for me? I have no doubt I cou