You don't have to

I find myself wondering what to do with this article and the incredibly well-condensed collection of sentiments therein. Do I drop a furtive link in that #ai-enthusiasts Slack channel, in full view of everybody, including that CTO of mine who is REALLY excited about this stuff, and isn't quite ready to issue an AI usage mandate? Or do I keep this secreted away, like a bible in my left jacket pocket in case one of "their" sharpshooters attempts to assassinate me with an AI silver bullet, as though I were some wild and reckless heathen that seeks to abandon all good sense in the selfish pursuit of personal pride?

I don't mean to be hyperbolic, but I never envisioned this moment of feeling bizarre, feckless and iconoclast for wanting to do what feels like the reasonable thing. The circular finances propping this stuff up don't add up. Do I really want to be left dependent on a thing that only really feels like it just flashed into precarious existence, that will almost without a doubt be propped up, classed as "too big to fail" by one particular nation-state, whose actual intent and interests are to subjugate its subjects into a quivering oblivion when it's finally acknowledged the whole thing is in arrears? No, goddammit, I shouldn't have to accept this barrage against and wholesale weakening of my strongest assets: My mind and my discipline. At least that's how I feel about it anyway.

Don't skim this one.

Maybe the arguing is really over whether it's higher-status to enjoy longform content, or to criticize it for not being more efficient? By identifying the argument, I've revealed it as silly, and clearly proven myself to be higher status than either side. The arguing may stop now. You're welcome.

I do wonder the prospects of any etsy-like outcome for largely hand crafted software though. While you can personally find stylistic expression in the craft i'm not sure how apparent the nuances of crafting code is to users of the product beyond the requirements of a UX design and vision. It's hard not to imagine generation industrializing a lot of this part of the craft of making software.

For me I think the important thing to not lose sight as we use generation more and more in software is our care for the work piece. It feels like care, and deep understanding are set up to become further valuable rarities in the future as we become less and less intimately involved and we have to be intentional about in order to keep.

I feel like there is some parallels here to industrial designers and their desire to hold on to obsessing about and understanding the details in the face of using industrialized tooling and being very much removed from the intimate feeling of crafting every millimetre. Deeply caring is still meaningful and valuable even if it isn't minimally required.

There's a flaw in the Milli Vanilli argument. The band had no input into their songs. They 'performed' them by lip-syncing on stage, but all of the music and lyrics were someone elses. Milli Vanilli had no part in the creative process.

That's not technically true of AI content. There's some tiny little seed of a creative starting point in the form of a prompt needed for AI. When someone makes something with Claude or Nano Banana it's based on their idea, with their prompt, and their taste selecting whether the output is an acceptable artefact of what they wanted to make. I don't think you can just disregard that. They might not have wielded the IDE or camera or whatever, and you might believe that prompting and selecting which output you like has no value, but you can't claim there's no input or creativity required from the author. There is.

I read it all, and found myself engaged throughout. Not to say that it was all riveting, there were certainly dryer spots than others, but it felt 'real'. Maybe they did use AI (I somehow doubt that given the content), but even if they did they went over everything in a way that retained a voice that felt authentic.

I hate many of the articles I read now all feel like they have the same half hearted attempt at trying to grab your attention without every actually clearly saying what they mean.

As for the content, I had actually just been told by management this last week that I need to become AI 'fluent' as part of future performance evaluations and I have been deeply conflicted about it. I do think AI has value to add, but I don't think it's something that should be forced and so this article resonated with me.

It's a long read, and not for everyone, but I recommend it as a way of hearing another humans opinion and deciding for yourself if it has value.

> Companies value velocity and new launches and shipping first at all costs because of course they do; it’s table stakes. Speed of delivery is basically the number one corporate value of every organization whether they admit to it or not.

Yeah this one is again one of the causes of where we are today (alongside profit extraction, or perhaps because of it). It used to be the case that you would find companies that would offer quality at a slightly higher price, and people would be more than willing to pay for it. Now the feeling is that this is all marketing driven and there is no 'higher quality' because everyone gave up and went after speed of delivery. And well, as the old saying goes, that's valuable when you're catching fleas.

I'm in a technical but non-IT industry that currently rents its software at commercial scale. The software in question that I use is terrible, and it is probably the worst on the market. The industry is such that it is not an exaggeration to say that people have died on account of its flaws. All other solutions in the domain are better but not good.

I have had a slowly-congealing dream/blueprint in my head for over a decade about how the system I use should work, and in the last 6 months I have been accelerated enough by AI (significantly more so by Opus 4.5/6) that I have built a version of the software that I am now using in production at in my job, and it is the most satisfied I have been in my career in the last decade.

Point being (and it was almost made in the article) that the software doesn't actually matter (no-one's reading the assembly either way) but the function and what it enables does. If it doesn't, no end user cares whether it was 25k SLoC or 25M.

"I’m not arguing that this technology should be unilaterally destroyed; I am arguing that we are collectively using it in the dumbest possible way, causing the most self-inflicted injury, and maximizing the amount of angst and suffering we’ll all have to contend with. I am angry at generative AI because it seems to be making us think and act like complete idiots."

Made me smile.

I know this is probably a deliberate simplification as part of a rhetorical flourish but one of my favourite parts about semiconductors is the fact that we don't dig up the rocks, we grow them to order. The though fills me with childlike wonder...

The author has made the correct call. There's a pretty deep irony that all the top-level comments at the time of this writing are about how the article is too long. It's quite clearly not trying to succinctly convince you of a point, it's meant to be a piece of genuinely human writing, and enjoyed (or not!) on the basis of that.

----

I rather strongly disagree with the framing around the environmental impacts, though; the article would make a much stronger point if it resisted the urge to peddle the same “muh water and electricity” disinformation that gets parroted all over the place by people who can't be bothered to put numbers into the context of other numbers.

For example:

> A single DGX B200 AI server is rated to consume 14,300 watts of electrical power at peak. You can cram about four of these on a rack if you like to live on the edge, and these four units might draw something like 200 amps of current combined. For a point of comparison, a typical single-family home in the United States will have wires from the utility company that are thick enough to provide a 200 amp service.

Cool, and how many households' worth of AI queries would that single B200 (let alone the rack of 'em) be able to handle? Probably a lot more than any individual household could ever hope to produce per second, even assuming a household consisting entirely of hardcore AI stans (let alone someone like the author, or like myself, who uses AI sparingly). Each of those servers is handling requests from thousands upon thousands of users; those power and water requirements get amortized over such a large quantity of requests (and people making them) that if you've ever eaten a single hamburger in your entire life then you've done more harm to the environment than hundreds (if not thousands) of those AI queries.

This all comes after this quip in the margins:

> You think you’re just gonna self-host an open weight model like GLM-5 on your personal hardware and cut out the hosting costs? Well, alright, hope you have 1,727 GB of VRAM lying around.

and like… the author does understand that not everyone needs such a large model with such a large VRAM requirement, right? Or that VRAM itself ain't even strictly necessary (it just happens to make things faster — which is more important for a server handling requests from thousands of users than it is for my laptop handling requests from exactly one user: me)? That's indeed part of the issue the author correctly identifies with people using AI in seemingly the dumbest way possible: that dumbness includes the demand for instantaneous responses, and the consequent demand for throwing more and more VRAM and SSDs at the problem, when “just make a cup of coffee while the LLM ‘thinks’ about what you asked of it” is a perfectly workable approach. As I'm typing out this comment, I've got Olmo 3.1 on this same exact machine doing a bunch of thinking about how to respond to me asking it “How much wood would a woodchuck chuck if a woodchuck could chuck wood?”¹, and it's totally fine that it's taking multiple minutes because there are other things I can do while I wait.

This all ain't to say that we shouldn't care about AI's power and water usage. We should absolutely be pushing for better efficiency. That includes acknowledging that there are options besides “throw more and more VRAM at it and hope for the best”; the article instead prefers to assume that the big beefy servers are the only option, dismissing the notion of self-hosting with little thought, and that dismissal does the article's broader point a disservice.

----

The discussion around AI being considered a “tool” also rubbed me the wrong way a bit:

> This unlocks a common refrain from the booster class: “A true craftsperson uses every tool at their disposal!” Which, if you think about it for more than three seconds, is ridiculous on its face. Gotta dig some holes for fence posts? Okay! Bring along every shovel on the truck, the Ditch Witch, a box of ANFO and the Bagger 293. Have the people who echo this kind of stuff ever built anything in the physical world? Your average craftsperson has one real good compound miter saw that they use for basically every cut on the jobsite. They’ll use it until it breaks down, then they’ll replace it with a newer model of substantially the same thing. In what world is constantly switching tools for the sake of switching tools a remotely smart use of time?

That's pretty blatantly a strawman, and seemingly the exact opposite of how even the most vibe-codey of vibe-coders use AI. They're largely using AI as that miter saw; they might switch out blades/models for a given job, but at the end of the day it's the same tool. That's indeed yet another part of that “people using AI in the dumbest ways imaginable” problem that's otherwise correctly-identified: AI maximalists having a hammer called ChatGPT and seeing everything as a nail.

And also: who cares whether or not someone brings along every shovel + the Ditch Witch + the ANFO + the Bagger 293 if it's easy enough to bring them all? That's only a problem to the extent that carrying one tool comes at the expense of one's ability to carry another tool. If you've got a big enough truck to carry all that gear around, and you're okay with taking the time to load and unload it all, then fuck it, might as well full send — and then if there happens to be a boulder blocking the path of your fence, then it's a good thing you have that ANFO handy, right?

And of course, most software developers ain't doing their work in a pickup truck in the middle of nowhere (though some are, and that's fucking rad). Most are doing their work at their desks, in their offices or homes, wherein they're probably in close proximity to the entirety of their collection of tools. Hell, even if they are doing their work in a pickup truck in the middle of nowhere, the vast majority of the tools they need are probably already present (or could readily be made present) on whatever laptop they're bringing along for the job. Toby and Lyle don't need to worry about the logistics of carrying their tools (in particular Lyle's trusty lathe) because they do their jobs in a workshop wherein those tools already live; I don't need to worry about the logistics of carrying around my compilers and editors and manpages and such (or even an LLM!) because I do my job on a laptop wherein those tools already live.

----

¹ For the record, Olmo 3.1 concluded (like most models do these days) that “If a woodchuck could chuck wood, it would chuck as much as it could—but given its actual habits, it would probably just dig a very efficient burrow instead.”

Over sixteen thousand words about how the author doesn’t really use language models very much but might in the future

I remember people saying this about emails vs postal mail.

So you don't have to:

"you don’t have to embrace a trend, tool, or narrative simply because others say you should — especially if it doesn’t resonate with you or align with your values"

An important new twist to add to the great AI versus NO AI discussion.