Just to make this clear up front: I was always very skeptical of LLMs (I’ll never call them AI) and still see them as fundamentally limited and unreliable. See my post history if you don’t believe me.
Having said that: I work in business software and the models have progressed to the point where typing code and writing classes and functions by hand has become uneconomical. We review the output closely to avoid the idiocy of “vibe-coding”, but the quality has improved to the point where we rarely have to outright reject solutions. The discussions have moved to high-level architectural points, where we just request better implementations than the first-shot. We get them, review again and move on. Maybe there are deeper, latent problems beeing embedded in the code-base through this process, but we can’t spot any although we are looking intently.
And just like that, my profession has died. Nobody in my team writes code anymore. Things that would have taken days are now done in minutes. We are all just considering design and architecture. We are changing our processes since they are not fit for the new way of doing things. Scrum? What’s even the point anymore? We can work through a sprint-backlog that would’ve taken two weeks in two days now. We do reviews 90% of the time. When we request changes they come back in minutes.
I have a feeling like the ground has been taken from below me. My son wanted to become a developer, that isn’t a real job anymore. Everyone around me feels the same way, including our management. We are a small team and a small company, everyone is effective and our jobs are not in danger as long as our product is succesful. But my profession is essentially dead. It’s a craft I honed for many, many years and the craft itself is now automated. I wanted to improve my typing skills to code faster, this is now a ridiculous endeavour. We don’t know what comes next, but I now only have my analytical skills and knowledge of good architecture to fall back on.
And it all changed so fast. I don’t think this will transfer to other professions so easily, but for programming LLMs ended it all in the span of the last four months. I thought I knew what my future holds as a senior dev, now I can barely see past christmas.This is a nightmare.
This is the exact same thing that happened with luddites, which today we know as ‘idiots who oppose technological progress’ thanks to capitalist propaganda.
It was never about efficiency, quality, quantity, or accessibility. It’s almost exclusively about getting people to stop doing things for themselves, and instead rely on expensive machinery owned exclusively by the capitalist class for everything.
Most average people used to make and fix their own clothing, which meant they had little need for corporations that produce the same clothing, but lower quality. So capitalists forced everyone to use machines, until nobody remembered how to make clothes without them.
Same with software, languages, and a lot of other skills and office jobs right now. Educated people who know how to do stuff are not only unprofitable, but also dangerous. Make them all use AI until they forget everything and have to rely on megacorporations for the tiniest thing.
This is all the more reason to keep reading books, learning, DIY as much as possible, coding your own software, and even learning stuff like woodworking, sewing, farming, etc.
Being educated, informed, knowledgeable, and skilled is among the easiest, but at the same time most effective acts of resistance.
Do not let them turn you into mindless slaves.
Keep holding on. LLMs are still in the operate-at-a-loss-to-control-the-market stage. Once they start charging the real cost, human software engineers will be more in demand. That and refocus. All of human history is careers reinventing themselves as new tools become available.
In my experience that’s only been the case for functionally non-unique software.
At my job we write software that’s effectively unique. There isn’t a pattern to fall back on.
It makes a lot more mistakes very frequently, since the nature of it keeps trying to guide it to the known path. It’s great for debugging since the manner that some check failed is chasing through known logic, and it doesn’t matter that it doesn’t see the existing code as probable or fitting a known pattern if the shape of a switch statement is still the same.In my personal projects I’ve even run into issues getting it to do stuff like vulkan synchronization barriers in a way that doesn’t continuously devolve into what you see in tutorials, with an ultimately unstructured pile of adhoc resource allocation and barriers.
It will cause a massive loss of human knowledge, nobody would learn how because it’s just a prompt
And it’s not only for coding. Yesterday I was shown that because they probably stole a lot of parametric designs Claude can even do 3d files from photos and quotes, it spits out an openscad file and compiles it. I was shocked, it added chamfering and small details, from a rough sketch photo. Who’s going to learn how to use a cad program if it can be done in 3 minutes from a prompt? And the process of learning taught you all the quirks , but in the future who’s going to be a pro?
This is how I feel too. I’m in several work chats where everyone seems to be pretending that Claude doesn’t have its arm shoved up their ass, wielding them like a puppet.
Several of the projects I am working with are just AI generated. In fact, if I need to suggest any fixes, they will be denied unless it’s AI generated.
I’m tired. I’m depressed. I used to take so much pride in solving problems, but all of that is gone now.
I’m going to try to ride it out and see where it takes me, but the work is mind numbingly boring and the trades look better and better every day.
I’m finishing my thesis in software engineering and at this point it’s hard to justify it. 6 years of my life pretty much wasted. At least I don’t have any debt
here i am in the game industry watching practically the entire industry move overseas. I’ve given up.
I’m in the exact same boat. I’ve been thinking a lot about how it’s changed the landscape of tech, and ive come to an interesting conclusion so far. It feels kind of like the switch from horse and buggy to motorized vehicles.
But, what’s going to happen after these companies bankrupt? Will AWS still host old Anthropic models at an increased price? It feels like when Walmart moved into small towns and sold their products at a loss in order to kill mom and pop stores.
Massive enterprises will cut deals to keep using frontier models, everyone else will have to go back to using their atrophied brains. Those who have the compute will be running local models.
Similar to how employers today ask if you have a vehicle to get to work, they will ask if you have local inference. It hurt, but, I bought an AMD Halo the other day. They turned the information super highway into a stroad and the Halo is my vehicle.
OP, im really quit sympathetic to you here. I’m among the knowledge workers who haven’t been fully replaced yet, but I’ve done some of the training meant to replace me. It’ll be a little while, but not that long, before I’m nearly unemployable.
The only solice here is the gift of limited forewarning. We can see the writing on the wall. Cobblers, per an example in another comment, didn’t see it coming. We have precedent set, including manufacturing sector jobs outsourced to counties with cheap labor, to look at, only 30 years ago! You can fight it and hope to be within a small group of privileged few who knew the right people, or you can tell your kid to pivot now and get ahead of the decade of tragic and sad “wha-wha-whaaaat?” that will inevitably follow. It’s not an easy choice.
I’m not sure what records and stories exist from these past eras that we can read and apply to now. Maybe a book of those tales is an opportunity for catharsis and reflection in the current moment.
All I can do is wish you luck and good judgement as this brave new world forms around us.
A shovel stuck in the ground does not move earth until it is used to move ground.
We are currently stuck in being amazed that the shovel goes so quickly and deep, now we must reckon how to make it move what it wrought.
Tokens are still insanely underpriced, especially on these bigger models/projects.
Look at what you can do locally with a couple of 5090’s. That’s probably what’s reasonable at the moment.
They’re currently sending through tokens in batches to clusters running balls to the walls but the cost is WAY higher than the $20 a mon you can get for Claude or midjourney.
We may find a happy medium after a pretty sharp collapse where AI assist development is somewhere in between you writing your own loops and you just rubber-stamping something else’s code.
This year has been interesting. A couple years ago, I dabbled with using LLMs for fill-in-the-middle (FITM) completion before I deciding I was more effective without them. I relegated LLMs to one-off tasks like transforming text. This worked well for a year or so, until I was asked to evaluate agents so we could determine guardrails for other team members. I realized then the models and harnesses had matured substantially. The writing was on the wall, and the business needed us to adapt to keep up with competitors.
Now I’ve become much more accustomed to developing with agents, and I dare say I’m pretty effective at it. I have my ethical concerns like many folks here, but I can also be too idealistic given the chance.
I’ve been developing and working with technology since the 90s, and I have been fortunate to get my hands dirty in many facets. One of the reasons I enjoy technology is the constant change, even though the rate of change is much greater recently.
I haven’t experienced the despair that others have, and I wonder if it is related to the path I’ve taken. I started a computer engineering degree many years ago, but did not finish. I wanted to learn too many things at once and spent too much time working (developing internal software for the university). I’ve blazed a trail over the years instead, so I guess my profession has always been somewhat nebulous.
If your job can be taken by a chat bot, it wasn’t a real job to begin with…
Maybe there are deeper, latent problems beeing embedded in the code-base through this process
maybe as a professional and a computer scientist you should have already made sure that this is not the case or warned your stake holders that this is an inherent risk
sounds like your job has not disappeared quite yet
Here’s the thing, though. Give it a generation (and in this industry, a “generation” is basically the amount of time it takes one class of grads to get through college, or possibly at the outside maybe two) and there will be no one left with the skillset required to effectively babysit the LLMs anymore. Everyone who used to have those skills but leaned on AI and stopped coding themselves will have let them atrophy, and the upcoming crop of junior developers won’t know how to do anything but prompt the chatbot.
Maybe there are deeper, latent problems beeing embedded in the code-base through this process, but we can’t spot any although we are looking intently.
It’s this right here. Nobody will be able to look intently, because no one will have the knowledge of what to look for anymore. You are able to effectively manage the stupid machine’s output because you actually know more or less what functional code looks like and how it works. Once that skill has been squeezed out of the industry it’s going to be utter chaos.
Yes, that’s pretty obvious to anyone in the industry. But like many catastrophies mankind can see decades ahead, nothing will be done about until the consequence bite hard. I’ll be old by then.
I’m sort of surprised there’s no conscious effort to counter that. In other areas where automation has taken over they will ensure that people can still do the work manually. I visited an automated metro system a while ago, and they explained that even though certain lines are GoA 4 (in other words so highly automated they don’t need staff on board anymore), they still have people operating the trains manually every once in a while, specifically so that they don’t forget how to do it if need be, and that the new recruits can learn as well.
I’m sort of surprised there’s no conscious effort to counter that.
Nobody wants to be the company “wasting” money on training young coders. This is the inevitable decay caused by Capitalism.
That probably wasn’t a privately run Metro then. In private industry it’s usually cheaper to risk an accident than pay for all those man hours.
Yes, to be fair it was, at least practically. It was one of these government owned companies - so that they’re still expected to work kind of as a private company. So sort of an in between thing.
Are you? I’ve not seen many people ever pay much attention to potential problems until it’s way too late.
It’s like how you can walk into a restaurant and retail establishment now and it’s not uncommon to interact with an employee that doesn’t know how to do the simple math required to make change during a transaction.
I’m stuck using MS Teams at my job, and I’ve noticed changes to the UI, removing of visual cues and triggers, which I can only assume they’re doing to get people to ask their stupid chatbot how to make a change.
Eventually this is going to bite hard, but it’ll only bite the working class, so the people in power do not give a shit.
Almost a graduated here, here’s my small rant (kinda).
For context: I rarely use LLMs cuz there’s no fun on vibe coding, not to mention the shit ton of problems its usage carries (economical, environmental, social, ethical, skills decline, etc)
The other day I was in the classroom using my Actual Intelligence (AI) to do a python project, when a group of randoms came by and asked what I was doing, they were looking at me like if a was some sort of caveman, because apparently I’m one of the few guys that still codes by hand.
Sometimes I feel bad for them, it’s depressing to see everyone getting stupider so quickly, but I can relate to that feeling of wanting to do absolutely nothing from day to day.
In most of my classes, I’m the only one answering questions to the point that sometimes I get banned from answering (although I don’t know if it’s because everyone is kinda shy or if it’s truly because they don’t have an idea of what’s being talked about)
My own experience of almost 2 decades as a Freelance senior software developer, mostly coming in to work on top of and even fixing existing codebases of large custom-made applications which were for years in Production in mainly large companies, which generally had layer on top of layer of different coding styles and even design approaches because multiple people worked in them each thinking “I know best” and doing their own thing instead to trying to work within the existing style, is that getting to the point of understanding it at all levels takes many months, and that’s even after one is experienced enough at it to have discovered shortcuts like learning and identifying the coding and design style of specific engineers and even their typical mistakes (people tend to be pretty consistent in the mistakes they make) as a way to accelerate understanding, detection of problems and figuring out were and how it’s safe to change code and design without it cascading to problems elsewhere.
All this to say that picking up and fixing a real life system in Production which was mostly AI created and hence doesn’t even have the patterns for coding and design choices and mistakes that humans have, will take many months, maybe more than a year PER SYSTEM even for somebody who is senior and highly experience at it.
So fixing the AI produced stuff will have giant costs which probably more than offset the savings at creation time from using AI.
Now, some would say that you can just get the AI to regenerate the whole thing to fix it, however that has two problems:
- You’re just replacing old known bugs with new and entirelly different bugs. Also any UI will change. Good luck selling such changes to the system they use daily to the business side.
- In my 3 decades of experience in the Industry, clear, complete, detailed and unambigous Requirements and Interface Definition documents are as rare as hen’s teeth and without such things you can’t prompt the AI to make software that yields the same user functionality and talks with other systems correctly. So features will change or behave differently, if it’s good enough integration testing will fail in brand new and exciting ways, if it’s not good enough it will blow up in Production or, even more entertainingly, Production databases will be filled with broken data.
So, yeah, full regen just at best moves the problems around and pisses users off.
It’s this right here. Nobody will be able to look intently, because no one will have the knowledge of what to look for anymore. You are able to effectively manage the stupid machine’s output because you actually know more or less what functional code looks like and how it works. Once that skill has been squeezed out of the industry it’s going to be utter chaos.
I see no reason why people can’t learn to do this task.
I learned to write an operating system kernel in college. Not because I would ever need to do that in a job, but because it was considered important to learn how stuff worked, to gain a deeper understanding to make architectural decisions. I also learned to write assembly, for the same reason. Similarly, good educations in the future will likely teach people to code, even if they will never code in their jobs.
That’s like saying few are left knowing assembly. It’s just not needed anymore. Could it be needed? Sure. Are you going to get a job easily with that? No.
It’s just not needed anymore.
Someone needs to maintain compilers though right ?
One child policy has done a number on China. Even more sinister example is Ethyl gas or gasoline with lead. that went for almost a century. Was also just wondering what ancient cities were like when they dumped their waste out on the street.
It is absolutely astounding our species has survived as long as it has.
Why would new learners not learn what to check…?
When we went from machine code to assembly people were very negative about it. Same again when moving to high level languages. And of course this continued with every single development since then and now it continues with LLMs. Always negative about it, no more skill, etc. the exact same arguments really.
This argument is flawed, as I just took a break from reading disassembly to scroll a bit. Assembly is the human readable representation of machine code. It is relevant today (eg embedded development).
Also each move to higher level comes with its own set of issues and trade offs. I think you somewhat correct to correlate with the past advancements, but LLMs fundamentally different: It is not a new higher language, its normal language. Also it cannot grab an oscilloscope and debug HW issues which are wrongly documented by the supplier.
