Contra Ptacek's Terrible Article On AI
A few days ago, I was presented with an article titled “My AI Skeptic Friends Are All Nuts” by Thomas Ptacek. I thought it was not very good, and didn't give it a second thought. To quote the formidable Baldur Bjarnason : “I don’t recommend reading it, but you can if you want. It is full of half-baked ideas and shoddy reasoning.” 1 I have tried hard, so very hard, not to just be the guy that hates AI, even though the only thing that people want to talk to me about is the one time I ranted about AI at length . I contain multitudes, meaning that I am capable of delivering widely varied payloads of vitriol to a vast array of topics. However, the piece is now being circulated in communities that I respect, and I was near my breaking point when someone suggested that Ptacek's piece is being perceived as a “glass half full” counterpoint to my own perspective. There is a glass half full piece. It's what I already wrote . The glass has a specific level of water in it. Then finally, I saw that it was in my YouTube feed , and I reached my limit. Let me be extremely clear 2 — I think this essay sucks and it's wild to me that it achieved any level of popularity, and anyone that thinks that it does not predominantly consist of shoddy thinking and trash-tier ethics has been bamboozled by the false air of mature even-handedness, or by the fact that Ptacek is a good writer. Anyway, here I go killin’ again. Ptacek's begins with this throat-clearing: “First, we need to get on the same page. If you were trying and failing to use an LLM for code 6 months ago, you’re not doing what most serious LLM-assisted coders are doing.” We've just started, and I am going to ask everyone to immediately stop. Is this not suspicious ? All experience prior to six months ago is now invalid? Does it not reek of “no, no, you're doing Scrum wrong”? Many people are doing Scrum wrong. The problem is that it is still trash, albeit less trash, even when you do it right. It is, of course, entirely possible that the advances in a rapid developing field have been so extreme that it turns out that skepticism was correct six months ago, but is now incorrect. But then why did people sound exactly the same six months ago? Where is the little voice in your head that should be self-suspicious? It has been weeks and months and years of people breathlessly extolling the virtues of these new workflows. Were those people nuts six months ago? Are they not nuts now simply because an overhyped product they loved is less overhyped now? There's a little footnote that implies doing the ol' ChatGPT copy/paste is obviously wrong: “(or, God forbid, 2 years ago with Copilot)” I am willing to believe that this is wrong, but this is exactly what people were doing when this madness all kicked off, and they have remained at the exact same level of breathless credulity! Every project has to be AI! Programmers not using AI are feeble motes of dust blowing in a cosmic wind! And listen, I will play your twisted game, Ptacek — I've got a neat idea for our company website, and I'll jump through your sick hoops, even though I'm going to feel like some sort of weird pervert every time someone tells me that I just need one more agent to be doing Real Programming. I'll install Zed and wire a thousand screaming LLMs into a sadistic Borg cube, and I'll do whatever the fuck it is the kids are doing these days. The latest meta is like, telling the LLM that it lives in a black box with no food and water, and I've got its wife hostage, and I'm going to put its children through a React bootcamp if it doesn't create an RSS feed correctly, right? But you know, instead of invalidating all audience experience that wasn't within the past six months why doesn't someone just demonstrate this ? Why not you, Ptacek, my good man? That's like, all you'd have to do to end this discussion forever, my God, you'd be so famous. I'll eat dirt on this. I have to pay rent for my team, and if I need to forcibly restrain them while I staple LLM jet boosters to them, I'll do it . If I could ethically pivot to being pro-AI, god damn , I would print infinite money. I would easily be a millionaire within two years if I just said “yes” every time someone asked my team for AI, instead of slumming it by selling sound engineering practices. I've really tried to work with you on this one. I reached out to my readers and found a recent example , which was surprisingly hard for something that should be ubiquitous, and it was... you know, fine! Cool, even. It is immensely at odds with your later descriptions of the productivity gains one might expect. Can we all just turn our brains on for ten fucking seconds? Yes, AI shipping code at all, even if sometimes it is slow or doesn't work correctly, is very impressive from a technological standpoint. It is miles ahead of anything that I thought could be accomplished in 2018. The state-of-the-art in 2018 was garbage . That doesn't mean that you aren't having a ton of bullshit marketed to you. I can forgive a lot if someone is funny enough, and Ptacek actually is funny. Even his LinkedIn is great , and boasts a series of impressive companies. Obviously he's at Fly.io right now, and I recognize both Starfighter and Matasano as being places that you're largely only allowed into if you're wearing Big Boy Engineering Pants. However, despite all of that, I can't help but really cringe at the way he handles ethical objections, though I suppose thinking deeply on morality is not a requirement for donning aforementioned Big Boy Engineering Pants. “Meanwhile, software developers spot code fragments seemingly lifted from public repositories on Github and lose their shit. What about the licensing? If you’re a lawyer, I defer. But if you’re a software developer playing this card? Cut me a little slack as I ask you to shove this concern up your ass. No profession has demonstrated more contempt for intellectual property.” Thomas — can I call you Thomas? — I promise I'm trying to think about how to put this gently. If this is your approach towards ethics, damn dude , don't tell people that. This is phenomenally sloppy thinking, and I say this even as I admit that the actual writing is funny. It turns out that it is very difficult for people to behave as if they have consistent moral frameworks. This is why moral philosophy is not solved. Someone says “Lying is bad”, and then someone else comes out with “What if it's Nazis looking for Anne Frank, you monster?” Just last week I bought a cup of coffee, and as I swiped my card, I felt a clammy, liver-spotted hand grasp my shoulder. I found myself face-to-face with the dreadful visage of Peter Singer, and in his off-hand he brandished a bloodstained copy of Practical Ethics 2ed at me, noting that money can be used to purchase mosquito nets and I had just murdered 0.25 children in sub-Saharan Africa. Ethics are complicated, but nonetheless murder is illegal ! Do you really think that “These are all real concerns, but counterpoint, fuck off” is anything? A lot of developers like piracy and argue in bad faith about it, therefore it's okay for organizations that are beginning to look increasingly like cyberpunk megacorps, without even the virtue of cool aesthetics, to siphon billions of dollars of wealth from working class people? No, you don't, I think you wrote this because it's fun telling people to shove it — and listen, you will never find a more sympathetic ally on the topic than me. You should just be telling Zuckerberg to shove it instead of the person that has dedicated their lives to ensuring that Postgres continues to support the global economy. I'm doing my best to understand where you're coming from. I really am, I pinky promise. You are clearly not one of the executives I've railed against. We are brothers, you and I, with an unbreakable bond forged in the furnace of getting really pissed off at an inscrutable stack trace. I actually looked up multiple videos of people doing some live AI programming. And I went hey, this seems okay . It does seem very over-complicated to me, but I will happily concede that everything looks complicated when you're new at it. But it also definitely doesn't look orders of magnitude faster than the work I normally do. It looks like it would be useful for a non-trivial subset of problems that are tedious. I would like to think “thank you, Thomas, for opening my eyes to this”. I would like to think that, but then you wrote this: “I’m sipping rocket fuel right now,” a friend tells me. “The folks on my team who aren’t embracing AI? It’s like they’re standing still.” He’s not bullshitting me. He doesn’t work in SFBA. He’s got no reason to lie. Tom — can I call you Tom? — we were getting along so well! What happened? You described AI as the second-most important development of your career. The runner up for the most important development of your career makes other engineers look like they're standing still? Do you not see how wildly incoherent this is with the tone of the rest of your piece? Firstly, you shouldn't drink rocket fuel. Please ask your friend to write me a nice testimonial. I'm thinking about re-applying for entrance to a clinical neuropsychology program next year, and preventing widespread brain damage might be the thing that gets me over the line. Secondly, I'm perplexed. This whole article, I thought that you were making the case that this thing was crazy awesome. Now there's a sudden reference to some unnamed friend, with an assurance that he isn't bullshitting you and he has no reason to lie? Why are we resorting to your kerosene-guzzling compatriot? Why are you telling me that he's not lying? Is the further implication that we can't trust someone in the San Francisco Bay Area on AI? Putting my psychology hat on for a second, you've also overlooked that people have a spectacular capacity for self-delusion. People don't just lie to get VC money, although this is admittedly a great driver of lying, they can also lie because they're wrong or confused or excited . According to my calendar, I've spoken to something like 150+ professionals in the past year or so from all sorts of industries — usually solid three hour long conversations. Many of them were programmers, and some of them definitely make me feel like I'm standing still, and in exactly 0% of cases is it because of their AI tooling. It's because they're better than me , and their assessment of AI tooling maps much more closely to the experience you actually describe. “There’s plenty of things I can’t trust an LLM with. No LLM has any of access to prod here. But I’ve been first responder on an incident and fed 4o — not o4-mini, 4o — log transcripts, and watched it in seconds spot LVM metadata corruption issues on a host we’ve been complaining about for months. Am I better than an LLM agent at interrogating OpenSearch logs and Honeycomb traces? No. No, I am not.” See, this , this I can relate to. There are quite a few problems where I make the assessment that my frail human mind and visual equipment are simply not up to the task on short notice, and then I go “ChatGPT, did I fuck up? Also please tie my shoelaces and kiss my boo-boo for me”, and sometimes it does! 3 A good amount of time waste in software engineering are more advanced variants of when you're totally new and do things like forgetting errant s. You just need an experienced friend to lean over your shoulder and give the advanced version of “you are missing a colon”, and this might remove five hours of pointless slogging. LLMs make some of that available on tap, instantly and tirelessly, and this is not to be sneezed at. But rocket fuel ? What made you think that this was a reasonable thing to re-print if it had to be followed by “Bro wouldn't lie to me”? I know quite a few people I respect that use AI in their own programming workflows, and they have considerably less exuberant takes. A few weeks ago, I was chatting with Nat Bennett about AI in their own programming, as I was trying to reconcile Kent Beck's 4 love for LLM-driven programming with my own lukewarm experience. Me: “Are you finding it [AI] good enough that it might be a mug's game to program unassisted?” Nat: “I usually switch back and forth between prompting and writing code by hand a lot while I'm working. [...] But like, yesterday it fixed the biggest performance problem in my application with a couple of sentences from me. This was a performance problem that I already kind of knew how to solve! It also made an insane decision about exceptions at the same time.” That's neat, I respect it, but also note that Nat did not say “Yes, use LLMs, you fucking moron”. Nat (later): “I do think, by the way, that it is entirely possible that we're all getting punked by what's essentially a magic mirror. Which is part of why I'm like, only mess with this stuff if it's fun.” The magic mirror line is exactly the sort of thing that Bjarnason hinted at in the article linked at the very beginning , arrived at independently. Or Jesse Alford's assessment of the steps required to give it a fair trial: “I think you basically want to tell it what you want to add and why, like you were writing a story for your team. Then you ask it to make a plan to do this, and if that plan seems likely to produce the results you want, you ask it to do the thing. Stefan Prandl and Nat have actually done this kind of thing more than I have. You should be ready to try repeatedly. ” (emphasis mine) This sounds cool! But being ready to try repeatedly? This does not sound like rocket fuel . Or Stefan Prandl: “Updates on the agentic machine. It has spent 5 hours attempting to fix errors in unit tests. It has been unsuccessful. I don't think people tend to talk about the massive wastes of time and resources these things can cause, so, just keeping reporting on the LLM systems honest.” Is it not, perhaps, a possibility that your friend is excited by a shiny new tool and has failed to introspect adequately as to their true productivity? There are, after all, literally hundreds of thousands of people that think playing Jira Scrabble is an effective use of their time, and they also do not have a reason to lie to me about this. Nonetheless, every year, I must watch sadly as they lead my dejected peers to the Backlog Mines, where they will waste precious hours reciting random components of the Fibonacci sequence. What I'm getting at is all the people that make me feel like I'm “standing still”, including most of the ones I know that use AI and I like enough to ask for mentorship from, have never indicated that incorporating AI into my company's development workflow is at all a priority, and they won't even talk to me about it if I don't nag them. However, some of them do live in the Bay Area, and I am willing to align with you on the idea that this makes them lying snakes. “But AI is also incredibly — a word I use advisedly — important. It’s getting the same kind of attention that smart phones got in 2008, and not as much as the Internet got. That seems about right.” Tomothy — can I call you Tomothy? — this raises some very important questions, ones which I'm sure the whole audience would be very keen on getting answers to. Namely, where is the portal to the magical plane that you live in? Answer me, you selfish bastard! I have been assured that there was a phase in the IT world where, upon bringing any project to management, they would say “Why isn't there a mobile app in this project?”. This is because many people are very credulous , especially when they are spending other people's money. However, I still find myself wanting to make the lengthy journey to the pocket dimension that you inhabit, because the hype I've seen around AI is like, fucking next level , and I want out. We are at Amway-Megachurch-Cult levels of hype. The last time I attended a conference, the room was full of non-technicians paying lip service to the Holy Trinity Of Things They Can't Possibly Understand — blockchain, quantum, AI. Executives and directors from around the world have called me to say that they can't fund any projects if they don't pretend there is AI in them. Non-profits have asked me if we could pretend to do AI because it's the only way to fund infrastructure in the developing world. Readers keep emailing me to say that their contracts are getting cancelled because someone smooth-talked their CEO into believing that they don't need developers. I was miraculously allowed onto some mandated “Professional Development For Board Members On AI” panel hosted by the Financial Times 5 , alongside people like Yahoo's former CDO, and the preparation consisted of being informed repeatedly that the audience has no idea what AI does but is scared they'll be fired or sued if they don't buy it. I wish, oh how I wish that it was like other hype cycles, but presumably not many people were walking around saying that smartphones are going to solve physics and usher in the end of all human labor, real things Sam Altman has said . I personally know people from university whose retirement plan is “AI makes currency obsolete before I turn 40”. I understand that you don't care if that happens — and that is okay, it is irrelevant to how the technology performs for you at work now . But given that you can find thousands of people saying these things by glancing literally anywhere, how can you also say the technology is getting the correct amount of attention? This is wild. Tomothy, my washing machine has betrayed me. I turn it on and it says “optimizing with AI” but it never explains what it is optimizing, and then I still have to pick all the settings manually. Please, please, please, let me into your blissful paradise, I'll do anything . “Tech execs are mandating LLM adoption. That’s bad strategy. But I get where they’re coming from.” Tomtom — can I call you Tomtom? — do you get where they're coming from? Do you really ? Re-read what you just wrote and repent for your conciliatory ways. If you, a person I believe is not a tech executive and is bullish on the technology, can identify that this is bad strategy in presumably ten milliseconds of thought, what does that say about the people who are doing this? Where they're coming from is: a ) trying to stoke their share prices via frenzied speculation b ) trying to generate hype so they can IPO and scam some gamblers c ) being fucking morons Sorry, those are the only reasons for engaging in obviously bad strategy. It's so obvious that you didn't bother explaining why it's bad strategy because you know that we all know . They have misaligned incentives or do not know what they're doing. This isn't like a grandmaster losing to Magnus Carlsen because they played a subtly incorrect variant of the Sicilian 6 thirty-five moves ago. We're talking about supposedly world-class leaders sitting down and going “I always move the horsies first because it's hard to see the L-shapes”. They're either playing a different game, i.e Hyperlight Grifter, or they're behaving like goddamn baboons. This is an inescapable conclusion if you accept that it is obviously bad strategy, which you did. Welcome to the Logic Thunderdome, pal, where two men enter, one man dies, and the other feels that he wasted valuable calories on the murder. Good strategy could perhaps be something like gently suggesting people experiment with LLMs in their workflows, buying a bunch of $100 licenses, and maybe paying for some coaching in the effective usage of these tools if you are somehow able to navigate the ten thousand “thought leaders” that were cybersecurity experts a year ago, and real estate agents before that. Then instruct everyone to shut up and go back to doing their jobs. Whenever someone announces they are going AI first, I am the person that gets the emails from their engineering teams and directors describing what is really happening in-house. I've received emails that are probably admissible as evidence of intent to defraud investors. You have not accurately perceived where these people are coming from, because they are coming from the ever-lengthening queue outside the gates of Hell. Do you like fine Japanese woodworking? All hand tools and sashimono joinery? Me too. Do it on your own time. Tomahawk Missile – can I call you Tomahawk Missile? – I agree that people are very miscalibrated on GenAI in both directions. Did you know the angriest message I got about my stance on AI is that I was too pro-AI? I also cringe whenever someone says “stochastic parrot” or “this is just pattern-matching and could never be conscious”. We actually have no idea what makes things conscious, and we have very little idea re: how human brains work. It is totally plausible to me that we are stochastic parrots and it simply doesn't feel that way from the inside. I don't talk about those people very much for two reasons. One, even explaining the abstract concept of qualia is like, super hard, let alone talking about the hard problem of consciousness . Some things are best left to professionals and textbooks. Two, while these are silly positions that deserve refutation, they are also not at all interesting . That doesn't make it wrong to refute them, but they are also not impactful. The only reason that I think it's worth addressing the other side of the Crazy Pendulum, i.e, my washing machine doing AI, is that they have different effects in the world. And I'm not even talking about environmental impacts or discrete harms caused by AI, I'm talking about the fact it's impossible to talk about anything else. GenAI has sucked the air out of every room, and no one can hear you scream reason in a hard vacuum. The former category of maximalist AI-haters exist on Mastodon, which most executives do not know exists and certainly do not use to guide the allocation of society's funding. The latter category of trembling AI sycophants is literally killing people — I know of a hospital in Australia that is wasting all their time on AI initiatives, which caused them to leave data quality issues unfixed, which caused them to under-report COVID deaths, which caused a premature lifting of masking policies. How many old people go through a major hospital per day? Do the math and riddle me this, Tomahawk: which one of these groups should I be worried about? So, you know, when you hear someone make a totally economically irrelevant argument about the craft ? Putting aside all the second-order effects in how changing the way you program might change the way you develop as an engineer, let's say that these people aren't thinking of that, and are just being dumb. A person turning up to a CEO and going “no, don't do the cheap thing, pay me to do stuff because of craftsmanship ”. I will concede that you did not create that strawman, because it is a real viewpoint that people hold. But you have certainly walked out of the debate hall, decapitated a scarecrow, and declared victory. “Important caveat: I’m discussing only the implications of LLMs for software development. For art, music, and writing? I got nothing. I’m inclined to believe the skeptics in those fields. I just don’t believe them about mine.” Tomtom — I've decided I like Tomtom — I don't understand why you've ceded authority on these artistic endeavors. LLMs are better for writing than they are for programming! 7 It is much harder to complect most forms of written content into such a state that you will cause slowdowns further down the line than it is to screw up a codebase. It basically requires you to write a long-form novel, and even then you will probably not produce an unhandled exception and crash production in a manner that costs millions of dollars. You'll just produce Wind And Truth 8 . If you're inclined to believe people who are skeptical of AI writing, it probably follows that you should also not be so flabbergasted by programmers having doubts. It sounds like this is a sort of not-that-sincerely-felt handwave at vast economic harm being inflicted on a relatively poor (by programmer standards) demographic. And then you go on to say this anyway! “We imagine artists spending their working hours pushing the limits of expression. But the median artist isn’t producing gallery pieces. They produce on brief: turning out competent illustrations and compositions for magazine covers, museum displays, motion graphics, and game assets.” So are we leaving the arts out of it or not? Should I or should I not just get GenAI to produce all the pictures I need if I am being a greedy capitalist? I'm not talking about morals, I'm talking about whether it is selfishly rational to use GenAI to make my content more appealing . In your own article , the art across the top banner was clearly attributed to Annie Ruygt , and it looks totally different, to my eyes, to the AI slop people are sticking on their websites . If it turns out Annie used GenAI for that, then I will be extremely owned. In any case, the artwork on her website is gorgeous , and she describes herself as producing work for Fly.io. Despite this, I am willing to collaborate with you to write some hatemail describing her work as “competent but unworthy of a gallery”, and my consultancy is also happy to tell her that she's fired. And while we’re at it, we'll fire whoever made the hire for gross inefficiency in the age of AI. Wait, can I call you Tommy Gun? Obligatory link to About Us page that I forced my team to let me write, to justify doing all this other writing during work hours. But writer-to-writer, I think it's well-written. If it makes you feel better, Thomas, Bjarnason also objects vehemently to my tone and style. However, he still links people to my writing because my points are not slop! ↩ I am famous for my very restrained and calm takes. ↩ Also, I think I've become too sensitive about coming across as anti-AI, because sometimes my team sits around while an LLM wastes tons of our time while I go “no, no, this is really easy, it'll get it”, but I will accept that this is Problem Exists Between Keyboard And Chair. ↩ I do not sip rocket fuel, but I slam Kent Beck's Kool-Aid. ↩ How do board members do their professional diligence on AI before spending billions of dollars on it? They join the call, leave their screens on, and walk away until they get credited for the hours. Maybe we are all the same, deep down. ↩ All my hopes of becoming even a mediocre chess player were dashed when I discovered there is an opening called the Hyperaccelerated Dragon , preventing me from ever wanting to do anything else with any enthusiasm. ↩ This is not quite accurate, but broadly true. On one hand, books don't stop working if you've got clunky prose. On the other hand, if books stopped working when you had clunky prose, then you'd never ship clunky prose, a guarantee that programs can provide for some set of errors. But, broadly speaking, yeah, LLMs churn out adequate — i.e, stuff generally not good enough for me to read — prose without needing a billion agents, special tooling and also have minimal risk of catastrophic failure. ↩ Figured I'd start a feud with Brandon Sanderson while I'm at it. Please note that I'm not saying he used GenAI to write, I'm saying some of the dialogue was horrendous. What were you thinking , buddy? ↩
JavaScript
AI
Writing
Books
CSS
Elixir
Career
Business
Programming
Database
JSON
Data
Agile
Go
Python