I’m a backend engineer. For the last couple of years a big part of my job has been wiring LLM “agents” into production systems: the kind that answer customer calls and chats, pull up your account, and decide whether a human ever sees your case. Most AI discourse comes either from boosters or from people who have never seen how this stuff actually gets deployed, so here are some notes from the shop floor.
1. The product is not intelligence, it’s deskilling. Braverman laid it out in Labor and Monopoly Capital: management studies the labour process, extracts the knowledge from the worker, and hands it back to them as a fragment. Separate conception from execution, concentrate conception in management. That is literally the project. Transcripts from experienced workers become training and evaluation data. The judgement of someone with ten years on the phones gets turned into prompts, flows and guardrails. Marx called this the real subsumption of labour under capital: capital no longer just buys your time, it rebuilds the work itself in its own image. What’s left for the human is the edge cases, at a faster pace and with less autonomy.
2. “Replace workers” is the pitch, “discipline workers” is the reality. The humans mostly don’t disappear. There are fewer of them, they’re cheaper, and they’re monitored by the same system: handle time, sentiment scores, compliance flags, all generated automatically. The agent is also the foreman. Marx describes exactly this in Capital vol. 1, ch. 15: in the factory it’s not the worker who uses the instrument of labour, it’s the instrument that uses the worker. Dead labour, in this case the frozen knowledge of thousands of past calls, dominating living labour. And the ones who are let go don’t vanish either. They join the reserve army, which is the stick held over everyone still employed.
3. It’s relative surplus value, not magic. Nothing here escapes the old logic. Marx quotes Mill doubting that all the machinery ever invented had lightened a single worker’s day, and his answer is that it was never meant to. Under capital, machinery is a means for producing surplus value: it cheapens labour power and lengthens the part of the day you work for free. The goal is to cut the labour time needed per interaction, and the gain goes to the client’s margins, not to shorter hours for the people doing the work. The same machine under different relations of production would mean the opposite.
4. The general intellect is being enclosed. In the Fragment on Machines, Marx talks about the “general intellect”, society’s accumulated knowledge, becoming a direct force of production. LLMs are that almost literally: trained on the collective writing of humanity, then rented back to us by a handful of firms. Socialised production, private appropriation, the basic contradiction in its purest form. Open-weight models, a lot of them coming out of China, cut against that enclosure, which is part of why Washington treats them as a threat.
5. The tech itself is useful. Don’t burn the data centers. Marx already settled this one. The Luddites smashed the looms, and it took time and experience, he writes, before workers learned to tell machinery apart from its capitalist use, and to aim their attacks at the social form, not the material instruments. Lenin made the same point about Taylorism: the subtle brutality of bourgeois exploitation fused with real scientific achievements, so take the science and abolish the brutality. Automating tedious, repetitive work is good. Under planning it could mean shorter hours, better public services, translation and accessibility for everyone. The problem was never the machine. It’s who owns it and what it’s pointed at.
Question for comrades: if you work in tech, or in a job that’s being “AI-transformed”, what does it look like where you are? And is anyone seeing real organizing among the people who build these systems?
The data centers are still a problem however, the way they’re being built and in what fashion means that they’re physically the problem. For me, LLMs are a necessary evil at work and a helpful tool for personal projects (in moderation and used smartly); but I use deepseek exclusively.
For example, there are approx ~450 datacenters in all of China, whereas the US has approx ~3400 data centers with far more on the way. The USian figure is speculation built on an initially productive (for some applications) technology. So, the datacenters are very much the problem because they’re being built on speculation and a far higher exchange value than their labor value
it’s always the same shit with USian industry, everyone wants a piece of the speculated big cake and don’t care if they fuck up everything in their way
The problem was never the machine. It’s who owns it and what it’s pointed at.
Capitalism is destroying the environment, capitalists are stealing your art and labor. The entire reason people hate AI/LLMs is because of what capitalists are doing with it. And yet the capitalists have convinced you that the tool is the problem.
the tool is very bad at several things people are trying to use it for, and it opened up a firehose of shitty creations.
some of the problems go away when the capitalists do, but the assclowns shitting out garbage for clout would not go away and their fecal matter would still get all over everything else of actual value.
Capitalism fuels a culture in which people are motivated to chase clout, so that they can try to translate clout into saleable reputation. I’m sure there are some people who are narcissistic for the sake of it, but those people would also tend to be more communally-minded under a communist upbringing.
the people doing no-effort static recomps of 15 year old videogames are not in it for the money
That sounds very specific, not sure what you’re referring to.
LLMs are fast at decompiling software so there has been a deluge of people shitting out PC “ports” of old console games, and because LLMs are oversold and the people are unskilled the of these things only get to a menu or part of the first level and then break.
nobody is going to maintain them, the people shitting them out aren’t capable of it, and the minority of people who are putting in actual effort to do some QA are buried under the pile of shit and the people who want to play these games have to dig through it to maybe find the good one.
This post itself reeks of LLM generation
Nobody writes like this deadass
Removed by mod
Removed by mod
First of all, I’m a woman. Second of all, I do have the professional experience to confidently identify AI-written or assisted text.
It doesn’t to me, but it wouldn’t matter if it was. The analysis is on point, that’s the part that matters here.
As I said in the other comment this is post is almost entirely generated by AI. The layout is an ordered list for no reason. Two of the five subheadings are “it’s not x, it’s y”. The sentences are short and their flow is abrupt. The use of colons is also something I have seen chatgpt do a lot recently but I haven’t seen any public acknowledgement of it like the em dash.
… and what about the content?
The content is disappointing because the title says “notes from the shop floor” but all the arguments are presented in abstract terms detached from practical reality.
The question at the end is the nail in the coffin. Do you really genuinely think the folks integrating LLMs in the software pipeline are organising in any capacity?
The criticism about the lack of concrete examples is fair. I probably spent too much time connecting things to Braverman and Marx and not enough describing what actually happens in these projects.
But I really don’t understand when numbered arguments and carefully structured writing became uniquely associated with ChatGPT. Have you read political pamphlets? Party publications? Lenin’s polemics? People have been organising political arguments into numbered theses, headings and neat rhetorical contrasts for generations. That’s practically a genre of political writing.
Anyway, the practical point is that decisions previously made by experienced support workers are increasingly being encoded into prompts, workflows and evaluation criteria. The worker doesn’t necessarily disappear. They lose discretion over the process and increasingly handle whatever the automated system can’t resolve. That’s what I meant by deskilling, and that’s the discussion I was hoping to have.
Even this reply is LLM generated. Which of Lenin’s polemics uses a numbered list?
The content is very low quality, it’s just a wall of text trying to do a mishmash of marxian categories in an erudite way but with very low depth. It does reek of AI slopness. If you want to use AI feel free to do so, but at least review the output and adjust the content adequately.
I don’t think that’s all that matters. I know that an LLM was used to some extent to write this. The only alternative would be that OP has used LLMs for so long that they’ve adopted the same writing style, which would be concerning. In any event, because an LLM was almost certainly used here, I can’t tell whether the LLM was just used to polish things up or to generate the whole analysis, and I’m not about to start putting effort into discussing this topic with a mindless LLM. You’ll notice that after OP responded to me using their own words, I wrote a detailed response to them.
I know that an LLM was used to some extent to write this. The only alternative would be that OP has used LLMs for so long that they’ve adopted the same writing style, which would be concerning.
No you don’t unless the OP says they have used an LLM. Arranging something in a bullet-point style manner is not exclusive to AI, it’s just something that Instruct models were trained to do because it helps with brevity and is easier for the user to process mentally. Furthermore, the analysis made is not that much different from other ones that people having been trying to say on this instance for a long time. If you’re afraid of wasting time conversing with a bot, you’d have to avoid the internet to be certain of it because there’s long been the possibility of that. It still matters to discuss for the sake of real people who are watching, unless we’re going to say Dead Internet Theory is real.
It has nothing to do with bullet points. It has to do with how the text sounds and is structured. It screams LLM to me. I use bullet points myself, but no person I’ve ever seen writes sentences and formats their text the way this post is written and formatted. I’m sure such people exist, but they’re rare enough that I’ve never met one in my life.
But you don’t actually know. That’s the bottom line. I remember a time on here not too long ago where somebody who was being very anti-AI accused me of writing a response to them with AI (it was 100% from my own head). When I said that, they claimed it must be that my writing was being influenced from using AI.
People would like to think they can tell, but they can’t reliably do so. We need to be able to accept the implications of not having certainty about it, what that means for text and images on the internet. Trying to figure out how to tell without explicit information on what is and isn’t AI generated is a losing battle because if somebody really wants to pass something AI off as human written they can just put more thought into hiding it in plain sight by using more colloquial styles and casual structures, especially with a model that does not have Instruct style drummed into it.
We can best guess in some situations, but that’s all it is, a guess. And if there are concerns about verifiability of information written by an LLM vs. written by a human and not knowing whether to trust it because we don’t know if it’s LLM written or not, well we should be cautious about that regardless. The internet has long been full of real people who may or may not be bullshitting to various degrees. I remember before LLMs, sentiment on reddit about how you’d have people who would share supposed insight that was actually bullshit and it sounded sensible to people who didn’t know the field, but sounded like bullshit to people who did. Point being, that was happening long before LLMs and it wouldn’t suddenly stop being an issue if LLMs disappeared tomorrow.
There are posts where it kind of sounds like maybe there was an LLM used where I don’t say anything because I know it could have been written by a person, and then there’s this, which just screams LLM. I suppose I can never be 100% sure, but I’d say after reading this post, I’m about 95% sure it at least went through an LLM at some point, and at most was generated entirely by one.
ngl “sounds like AI” is becoming the new “ok boomer”, a way to skip the argument. braverman is from 1974, take it up with him lol. any thoughts on the actual content tho?
It just feels bad to read. Like I understand using the help of AI to get your thoughts in line, research, polishing etc. but this reads like it has been copy pasted from a chatbot’s output. You could make the same argument in half the word count and it would be twice as readable with just half an hour or so of drafting.
apart from the whole litany of benefits of writing for yourself (and the inherent issues with handing the production of nonliberal theory over to nonthinking LLMs trained on liberal texts), the problem with using LLMs for writing is that it is a black box through which no one can determine your writing process. i.e. when the output is obviously AI-generated, no one can determine the extent to which you used AI in your process, only that you used it at least some: did you use it to simply polish a few sentences? did you use it to create whole elaborative sentences? did you use it to create the entire structure of the piece? worst of all, did you use it to create entire points of your argument?
you can’t blame anyone for questioning the authenticity of your writing when it isn’t entirely your writing. if you care enough about the contents therein you should care enough to take the extra time to actually write it yourself. otherwise, yeah, i don’t think the contents deserve to be responded to. and, after all, this question is directly related to what LLMs are “actually for”
Sorry, I’m just very frustrated with all the unjustified LLM hype and I don’t react well to reading phrases that, at least to my mind, scream generative AI, because of bad experiences of both myself and my friends.
I work in tech (DevOps), my company offers paid LLM accounts on request, and I requested one just to see what’s up. I tried it once, and I haven’t used the account since.
I’ve found that LLMs actually slow my work down and decrease its quality rather than speeding it up and increasing quality. I can ask an LLM a question and give it every piece of context, and the answer it comes up with is nearly always either obviously incorrect, or it’s one I’ve already considered and ruled out because of a constraint (a constraint that I relayed to the LLM). Once this happens, you have to guide the LLM to a solution by asking further questions and repeating info. Ultimately, I’ve found it takes significantly longer to babysit an LLM to a solution than to just think of a solution and implement it by hand.
I hear from a lot of people who’ve found the opposite. I’ve found there are broadly two kinds of people like that. People who otherwise know what they’re doing but would rather prompt an LLM than think about the problem themselves, and people who don’t know enough about what they’re doing, and therefore LLMs speed them up because they don’t have to figure out what they’re doing to solve the problem. I don’t particularly like the first kind, but the second kind is especially dangerous because they won’t be able to realize when the LLM is wrong, and they will have to try the output and bump into a real issue before they figure out something isn’t right. By then, damage can and has been done.
Additionally, If I already have all the context required to solve the problem, I don’t need to outsource the thinking because my brain is much better and faster at it than an LLM. As a result, I’ve found LLMs useful for only two things. One is reading through search results so I don’t have to sift through them myself (I usually completely ignore what it says and just click on the sources it found to read them myself and see if I find them reasonable and trustworthy), and the second is as a sounding board. If I feel stuck, I can ask explain the problem to the LLM, and it almost never finds the solution, but it usually gets me unstuck by being a rubber duck.
In my opinion, from my experience and what I’ve seen, LLMs aren’t automation. They are tools, but not for what the AI companies would have you believe. They’re tools for language analysis. They’re certainly not general intelligence or anywhere remotely near it. Unlike lots of people I know, I don’t think burning all the data centers down is a good idea (though I don’t mind western corpos losing a ton of money when they are burnt down). However, I believe using LLMs to directly perform tasks, especially in tech, is not a correct use of them and will never yield good results. They are language models, after all, so that’s what they’re good at. That’s what I use them for.
My personal projects have a very strict LLM policy, but it’s less strict than most of the policies on my friends’ projects. It’s pretty simple. If you use an LLM, you must disclose it. You cannot credit LLMs in commits even if you used an LLM to make the commit. If the LLM wrote the actual contents of the change, the change may not concern more than 20 lines of code, and over 20 lines of code, all LLM-generated commits are closed without review, any issue or PR containing a body or comments obviously written by an LLM is closed without review. The policy is designed to be broken in some sense, because if I can’t tell the content is generated by an LLM, then it’s good enough for me. I feel this adequately leaves space for LLMs to be used for trivial tasks and changes, while preventing a lot of the extra workload that comes from babysitting LLMs and filtering out people who use LLMs and don’t know what they’re doing.
tl;dr LLMs are not automation, they’re not general intelligence, they’re specialized tools for language analysis, and they should only be used that way for acceptable results.
Yes to all of this. Especially the part about having to babysit the agent except for the simplest of problems. Anything that involves even the smallest of departures from the most common case become tedious slogs of cajoling the damn agent over and over.
Fair enough, my last reply was a cheap joke.
But I think we’re mixing up two different things here. I’m not arguing that LLMs are intelligent, or that every developer should be outsourcing their thinking to one. I work with these systems, and a considerable part of that work is dealing with their failuresand the ridiculous expectations management has of them.
Where I disagree is the claim that they’re not automation. I’ve seen actual human tasks broken down into workflows, with parts handed over to models and the remaining work pushed onto fewer people. Whether the model understands anything is almost beside the point. The labour process has changed, and that’s what I wanted to discuss.
As for the writing itself, I find the argument about structure a bit odd. Political pamphlets and theoretical texts have always been edited, condensed, divided into numbered points and written around fairly repetitive rhetorical constructions. Half the political literature we’ve inherited is structured that way. It didn’t suddenly become an AI characteristic in 2023.
I do take the criticism that I should’ve included more concrete examples from work rather than leaning so heavily on theory. That’s fair. But I’d rather discuss what’s actually happening to workers than spend the whole thread trying to establish whether a colon or a numbered paragraph was invented by ChatGPT.
It’s not about the numbered points. There’s not much specifically I can point out and say that’s the exact thing that makes it sound like an LLM. I just have an intuition for it from a long time reading things written by LLMs and this post triggers that to an extent that makes me seriously doubt it was written completely by a human. All I can say is that it’s in the way the sentences are written and the text is structured. I don’t have the ability to be more specific about it, but I know it’s not the numbered list.
For example, the phrase “‘Replace workers’ is the pitch, ‘discipline workers’ is the reality” screams LLM so incredibly loudly to me. If it was phrased more like “The point is to discipline workers, not replace them” (obviously doesn’t have to be exactly that but this should be enough to illustrate the point), that doesn’t sound nearly as much like an LLM to me.
That’s not the only thing that triggers my “AI detector” though. It’s sprinkled throughout the whole post. That’s just the most obvious one.
I get what you’re saying, but I think there’s a bit of a cultural difference here too. I’m Greek and that’s literally how a lot of political texts are written here, especially in communist and anarchist circles. Pamphlets, announcements, you name it. They’re full of these “it’s not X, it’s Y” constructions, sharp contrasts, numbered arguments etc. I’ve been reading and writing this stuff for years and I guess it just carries over into how I write in English.
I mean I get why you’d associate that style with LLMs nowadays, but these models learned those patterns from human writing in the first place lol. We cant suddenly treat every rhetorical device they picked up as evidence that a human didn’t write something.
Would you like me to make this reply more concise, or adjust the tone to sound more casual? Let me know!
Removed by mod
deleted by creator
Fuck, now I have to go double check all of the sources it cites for errors because I started to take the bait
“The Luddites smashed the looms, and it took time and experience, he writes, before workers learned to tell machinery apart from its capitalist use, and to aim their attacks at the social form, not the material instruments.”
I try to express this very concept when I see people acting like the AI is the problem, rather than the way capitalists abuse it for profit. It has always been true that the means of production can liberate workers from labor or enslave them, depending on who controls the means of production.
Capitalism has always been the problem, I don’t understand why this is so difficult to accept. Instead we just get the same regurgitated talking points. “Slop” has become a meaningless term at this point to refer to things people don’t like, it doesn’t even have to be anything related to AI.
This seems to be the case with language as a whole now. Anything someone doesn’t like is “communism/socialism” “woke” “slop” etc.
Welcome to Idiocracy.
ow my balls!
Right now at my work it is used by every director, PM and other useless positions to still deliver mediocre documentation, requirements, etc
But now it reads like AI slop, with the obvious ChatGPT comparisons, just like your post actually
Last week I attended an in-person meeting and the slides were riddled with marketing terms from LLMs
In the dev team we use it for code, we are encouraged to use it to “deliver faster”, they are unable to realize we can’t deliver faster because there are other issues (changing requirements, unclear instructions, missing docs)
I’m pretty sure my dept chief has AI psychosis, they feel free to use our faces and ask Gemini to mash them up in memes
But now it reads like AI slop, with the obvious ChatGPT comparisons, just like your post actually
The tech itself is useful. Don’t burn the data centers.
I can only picture the LLM asking for mercy in this line
Haha yeah I thought the same thing. “What are you doing, Dave?”
I am required to use Anthropic corporate subscription tokens where I work. It produces wrong results about half the time. Of course, it is difficult to tell if those results are incorrect without fully inspecting the generated code and reading all the explanatory output, and of course by also checking if unit tests are both passing and coverage has not altered. Half the time it runs existing unit tests to ensure change fidelity, but half the time it skips this step even though it is specifically instructed to run such tests after making changes. I also ask for the agent to generate automated smoke testing with a report of test results that I can run myself. This is a very tedious and time consuming process, and I wind up throwing away about two thirds of the generated code including smoke test automation per issue.
This process does not take too much longer than my regular workflow. But reviewing explanatory output from the language model which narrates the “reasoning” behind any change more complex than a simple field or function rename is overwhelmingly tedious because of the bizarre linguistic turns of phrase generated by the inference. So most of the time it actually winds up being more mentally exhausting than just debugging, or searching a code base in my editor and making changes myself.
Listening to myself complain about this new workflow I keep hearing the apologizers arguments in my head, like – this is just the beginning, man! We’re on the second rung of the exponential ladder!
I don’t know about that. I am skeptical. It doesn’t seem like it is getting significantly better, only the agent software has gotten pretty good at automating some of the stuff software developers do, like code searches and stack tracing.
So if the argument is that agent software will improve generated code quality exponentially, I really doubt that. Or, if the argument is that the models’ output quality will improve exponentially, that seems highly unlikely without devastating operating costs externalized to ecosystems or non-user populations.
Jevon’s paradox prevents any actual efficiency gains from outstripping usage rate increases (if the actual output quality improves sufficiently to make an expanding user base unavoidable) which would make externalized costs continue to grow at an also exponential (though derivitive) rate.
Either way I slice it, my sense is that most outcomes from large language model usage are morally suspect in outcomes. At best, they will supplant reasoning skills and coding skills that took me decades to develop, but will produce software that is worse for at least the next year or so, until output quality stabilizes around A grade level or better.
What I mean by that last part is that current models and agentic engines get graded by quality tests where problems are presented and solutions are graded. It seems that the quality grading winds up putting most systems at a place on a bell curve compared to humans just around C-grade quality. So, if this is just lower quality because the tech is still in its infancy, that’s fine I guess, but will all the results implemented and deployed into production systems using these subpar results be unwound when the systems capable of better quality results finally Moore’s Law their way onto our laptops? I also doubt that.
Meanwhile, if Capitalists wind up overcoming the psychosis bug and instead let the skeptic take over the narrative, it then seems unavoidable that costs could skyrocket for users of current best-of-class systems, and we’ll all be left using D-grade or worse models backing the agent software, or accepting incredibly large wait times for responses from slow local or overloaded/throttled datacenter-hosted inference processors.
So if current trends continue, I predict a bubble implosion typical of over-exuberant capital, and a drought of model-assisted code work for years afterward, if not a decade. And within that era, many people will be employed to correct sub-par software developed by model-backed agent systems that were basically merged and deployed into production systems that only “seemed” to work and pass gating unit tests, but wound up making maintenance by less available model-backed agent systems a serious challenge in a post-bubble world.
People forget about the S-curve inherent in all new technology adoption and growth. It only looks exponential at first. Eventually there’s always a plateau, and fickle Capitalists used to quarterly profits or liquidation of holdings get skittish until the next wave.
In our case, the next wave may never come without a fundamental set of improvements in underlying energy and cooling infrastructure technology. And if those improvements never arrive (and there’s a real chance such improvements are walled off by hard physics limits) then we’ll either see ecological collapse or, energy market collapse, or just simple linguistic inference quality limits. Any of which could disable further development of this technology tree.
Leaving us in the lurch to get ourselves out of the hole we find ourselves in.
If you want to convince people of your view, write your own words. I’m not reading the output of your favorite slopmachine.
Ok
Thank you ChatGPT!
Folks, I got late to this argument party, but since the discussion is too heated, I decided to lock the post. This post certainly seems to be AI generated. I don’t have a position against AI and its use in art or programming, however I do have a position on posting low quality content in social networks in order to create polemical discussions. So therefore I am closing this post and this discussion as it is not going anywhere.
And while I don’t have a problem with using bad words in a rhetorical way, I do when they are employed in an offensive manner, so I ask everybody to act respectfully to each other. This is one of the core community rules, more details here.
Edit: I removed offensive comments in the post. But as a warning, if I see the same behavior being repeated, I will enforce more serious punishments.
Seeing even an admin participate in an AI witchhunt here has really disappointed me. I’ve seen the people and culture of lemmygrad be better than this countless times. Disliking gen AI is not an excuse for the vitriolic witchhunting that goes on with it.
lots of people who should know better than presenting opinions and vibes as facts and getting pissed when people demand a little bit more than a personal hunch, and ask that they look past that to examine the actual facts.
“I know it was written by an LLM because… b-because I just know it is, ok?!? I know it in my heart!!”
I’m gonna go do something else lol this is hopeless
do you have the engineer’s ring
