I found this thought funny. A few years ago everyone was all learn to code so you don’t lose your job! Now there wont be any programming jobs in 10 years. But we will need a lot of manual labor still.
LLMs can recite code when asked properly, with a lot of errors. Trying to put code together with it without understanding how said code works is a greater insanity, than making random numbers with mathematics.
The real reason why there’s a downtick in coding jobs is due to Xitter not imploding immediately after the mass firings. Now coders are working overtime with skeleton teams on the same problems, while being overburdened and making more mistakes.
I think AI is a component of the decline.
For decades, companies have operated under the misunderstanding that more software developers equals more success, despite countless works explaining that’s not how it works. As a result many of these companies have employed an order of magnitude more than they probably should have and got worse results than they would have. However the fact they got subpar results with 10x a good number just convinced them that they didn’t hire enough. Smaller team produce better results made zero sense.
So now the AI companies come along and give a plausible rationalization to decrease team size. Even if the LLM hypothetically does zero to provide direct value, the reduced teams start yielding better results, because of mitigating the problems of “make sure everyone is utilized, make sure these cheap unqualified offshored programmers are giving you value, communicate and plan, reach consensus along a set up people who might all have viable approaches, but devolved into arguments over which way to go”.
AI gives then a rationalization to do what they should have done from the onset.
I think the post is about the state of agentic coders in 10 years, not the present.
The reactionary “learn to code” nonsense started a lot further back than a few years! Also, who told you there won’t be any software development positions in 10 years?
Anthropic for starters. What they don’t realize is everyone will be switching to cyber security jobs in 10 years when every vibe coded piece of software is riddle with security vulnerabilities.
Anthropic and others are over hyping their product so they can sell it. They are likely wrong.
And yeah your point stands.
Wait AI is overhyped? /s
¯\_(ツ)_/¯
¯\ /¯ \_(ツ)_/
If quantum computers become a standard thing in 10 years it could be even worse.
God, imagine some two-year-old iPad kid with a quantum tablet accidentally trashing a vibe coded DOGE system by happenstance.
More like there’ll be more coding jobs because of how fucked up LLM-coding can be
Now there wont be any programming jobs in 10 years.
Bullshit.
While “any” is a bit much, I do anticipate a rather dramatic decline.
One is that there are a large chunk of programming jobs that I do think LLM can displace. Think of those dumb unimaginative mobile games that bleed out a few dollars a week from folks. I think LLM has a good chance at cranking those out. If you’ve seen companies that have utterly trivial yet somehow subtly unique internal applications, LLMs can probably crank out a lot of those to. There’s a lot of stupid trivial stuff that has been done a million times before that still gets done by people.
Another is that a lot of software teams have overhired anyway. Business folk think more developers mean better results, so they want to hire up to success, as long as their funding permits. This isn’t how programming really works, but explanations that fewer people can do more than more people in some cases can’t crack through how counter-intuitive that is. AI offers a rationalization for a lot of those folks to finally arrive at the efficient conclusion.
Finally, the software industry has significantly converted transactional purchases to subscription. With perpetual license, you needed to provide some value to drive that customer who bought from you 5 years ago a reason to upgrade. Now with subscription models, you just have to coast and keep the lights on for those customers. Often with effective lock-in of the customers data to make it extra hard or impossible for them to jump to a competitor, even if competitors could reverse-engineer your proprietary formats, the customer might not even be able to download their actual data files. So a company that acheived “good enough” with subscription might severely curtail investment because it makes no difference to their bottom line if they are delivering awesome new capability or just same old same old. Anticipate a log of stagnation as they shuffle around things like design language to give a feeling of progress while things just kinda plateau out.
With perpetual license, you needed to provide some value to drive that customer who bought from you 5 years ago a reason to upgrade. Now with subscription models, you just have to coast and keep the lights on for those customers.
True, but it’s that market preference is a pendulum. It swings back and forth. It’s funny how hard companies are pushing today to (fail to) keep it from swimming swinging back towards owning things.
Companies that try to charge monthly for service that isn’t improving eventually lose their customers, except in the rare cases where stability is the only customer motivation.
Being able to just cut off access to the application means a customer has little choice.
For a competitor to pass them, they first have to catch up. To catch up, the customer needs to be able to extract the data from the application to give competition a chance. If they get closer to catching up, they tend to be bought out. Lot of speedbumps to discourage competition. Also, to get funding those competitors have to pretty much promise investors they will also do “as a service”.
For assets versus expense, I see a pendulum, largely based on how appreciation/depreciation pans out versus acquisition cost and loan interest rates, as well as uncertain start up versus steady business. I’m not sure software is giving enough choice in the matter the let that swing.
Hes not wrong. Amazon for example is getting rid of SDE roles for AI as per leaked conversation from the head of AWS.
Its coming anyone who thinks other wise will have a surprise pikachu face.
The real answer is somewhere in between.
There’s going to be less programming jobs, but there’s still always going to be some demand for them, there’s always going to be some technical knowledge required, even if just “prompt engineers” or similar concepts. Things still need to be built and fixed, and if you’ve worked for enough project managers/product managers, you know their lack of technical knowledge would not be enough to even prompt an LLM much less do anything else.
Yea pretty much you will still need a handful of SDEs either way to validate and work on the AI models but those big teams where they hire shit ton of devs those will soon be a thing of the past. Even IT outside of break fix hardware support will be replaced by AI at a help desk/IT Support level. (Already seeing that at my job as an IT analyst)
For the most part I’ve only ever been on smaller teams anyway, my largest team has been my current job with like 15 developers but there’s so much work to go around, so many projects constantly being worked on it’s kind of expected to have this many (and still be hiring more every year) lol
I can think of no better way to train an AI to hate humanity enough to invent Skynet and kill us all, than to introduce them to MS Teams meetings with managers who all want things that are completely incompatible with what they asked for the last time, and require you to throw away about 40% of what you already wrote.
Remember when Biden told coal miners to learn to code
“My liberal friends were saying, ‘You can’t expect them to be able to do that,’” Biden told his New Hampshire audience. “Gimme a break! Anybody who can throw coal into a furnace can learn how to program for God’s sake.”
These politicians and policy makers don’t know what they talk about when it comes to tech. Any one who tells you that programming jobs will be gone because of AI has never written a complex piece of software before. Also the trades pay well because there is a shortage of workers. If everyone starts going into the trades wages will crater. It’s just cycles. I remember when nobody wanted to go into the trades because it didn’t pay well. This created the shortage of workers. And since salaries are better now because of the shortage lots of people want to go into the trades This will create an oversupply of tradespeople and the cycle will repeat.
Building trades are hell on your body and there is no goddamn way that any construction worker (Electrician, carpenter, plumber, pipefitter, mason etc) can last until SS age-- esp. as they are planning to raise it to 70.
It’s two quotes. Miners don’t throw coal into furnaces.
My liberal friends were saying, ‘You can’t expect them to be able to do that,’” Biden told his New Hampshire audience. “Anybody who can go down 300 to 3,000 feet in a mine, sure in hell can learn to program as well, but we don’t think of it that way,” he said.
“Gimme a break! Anybody who can throw coal into a furnace can learn how to program for God’s sake.”
Part of me wants to argue this isn’t a cycle of demand and is instead capitalism. The trades didn’t pay poorly because there were too many people as much as people willing to work for less and the employer will pocket the difference. I admit this is extremely pedantic of me to split hairs here but people have an effective floor for how much they can work for. Coal miners weren’t being told to code because there were too many coal miners but that they could never work for as little as the machines that were replacing them.
To be clear I’m not saying AI is a replacement for programmers, I’m not able to see the future here, but capitalists will attempt to to replace any labor with machines if possible.
Lol anyone who thinks you don’t need any programmer in 10 years of time will burn and crash in the next few years when finally realizing that AI isnt as intelligent as we’re being sold.
Good luck trying to troubleshoot the code AI wrote tho.
the only job that can’t be replaced is… venture capital guy
The ai that’s been trained specifically to predict trends: ;3
The Learn To Code hype was being driven by employers to create a work surplus to drive wages down. Now those same employers think they can use AI instead.
then it’s going to inadvertently happen to the trades too 😬
Any new construction job is going to crash because no one will have any money to build new anymore. I’m already seeing stalled projects near me. Not that I have a big problem with that. They like to cut down and cleared trees to build a warehouse instead of tearing down old buildings.
Large corporations will always have the money. I wouldn’t be surprised if some of them get so large they start forming in-house construction companies, initially offering above market pay and benefits, to attract large teams of workers and undercut existing independent (often unionized) construction services. The competition forces the indie union shops to shutter or sell, and now, in control of the entire workforce, the corporations slash the wages and benefits as the workers no longer have other places to apply to.
Race to the bottom, baybeeee
Yes. The billionaire goal is to get back to building company towns.
They’re not going to be building malls and warehouses if no one is buying shit anymore.
can’t wait for all of us to become so poor that they re-legalize company stores and scrip
I think that’s their plan. The billionaires seem inconvenienced by the rest of us living where-ever we want, and buying we want from who we want to buy from. Company towns would solve that, for them.
If machine intelligence is indeed a different form of intelligence, then it can be observed and judged on the basis of its own merits, as opposed to a messianic waiting for a moment where it might equal or eclipse (weakly defined) human intelligence. This would even render obsolete the question as to whether or not machines can think—which in itself willfully glosses over the corresponding opposite question, “Can humans think?” posed by the former Fluxus artist (and Emmett Williams collaborator) Tomas Schmit in the year 2000 (Schmit et al. 2007, 18–19). — Crapularity Hermeneutics: Interpretation as the Blind Spot of Analytics, Artificial Intelligence, and Other Algorithmic Producers of the Postapocalyptic Present. Florian Cramer.
Trades were being pushed 20 years ago
But programming is a good supplemental skill in every field
Yes! As Heinlein wrote - “Specialization is for insects.” Any human can benefit from any skill in any of the trade skills and programming, too.
Problem is, people want a silver bullet and there just isn’t one.
You need to create an economy that works for everyone where skilled workers from all professions can be successful. You can’t cram everybody into one job and expect everything to just work out.
Just about all jobs are important, and all workers deserve a living wage and fair compensation.
No amount of Band-Aid job stuffing is going to make up for a leadership that doesn’t believe that everyone ought to be able to live a good life.
Demand for human workforce is primarily caused by the economy’s growth. It requires significantly less human workers to maintain an economy.
That really can’t be correct. Growth adds some jobs for sure, but not as many as you’re implying. While “maintaining” an economy, you still need just as much health care, food production, retail stores, education, road maintenance… I mean just about everything I can think of with the one exception of construction won’t be significantly different in a growth economy.
I remain deeply skeptical that AI can solve the types of complex problems that require human thought. AIs will never be able to abstract away details correctly or design sensible workflows for boutique problems.
They can’t, this is the same shit that happened when the dipshit ceos sent dev jobs over seas to code farms. Devs lost their jobs, and the code went to shit. Then when shit started breaking, they magically rehired everyone again to spend years cleaning up the shit code. LLMs are this all over again, just quicker this time.
The problems start if it can take on a lot of the junior work. If nobody can enter the industry, nobody can get the experience required to do the real engineering.
Open-source and personal work may be the only way to enter the programming field in the next decade.
Now is the worst time to try to enter the field. We need to see the AI bubble burst much more spectacularly, and only then might it be more reasonable. You certainly don’t want to try to get into a field when you have a lot of other choices when that field is already flooded with all of these people who have been laid off, combined with the increased availability of programmers in other countries, knowing that at the moment many domestic programmers are not smart enough to form strong unions to protect their own jobs.
It was really hard in the mid 1980s to find a job as a new grad as all the Boomers who had been laid off during the recession were hired first as they had experience. It was McJobs or nothing unless you were a computer science/programming grad. Things have changed dramatically since then. It is a different world.
AIs will never be able to abstract away details correctly or design sensible workflows for boutique problems.
Not the current direction of AI, no. But the field is ever advancing. I won’t be shocked if we see AI capable of these things within my lifetime.
A lot of the things that current “AI” is doing exist since the 90s or even earlier. It is just that now the computational capacity is big enough to make much more complex looking inputs and results.
Claiming it will never be able to do something is silly. We have no idea what advances will come in the future.
I never said that.
The key point is that we are still limited by what LLMs can and can’t do and fundamentally this is no new technology, just refined technology.
Think of it like cars. Cars exist since more than a hundred years. A modern car looks much fancier than a car a hundred years ago. But when it comes to the core aspect -moving passengers and cargo around on the ground- modern cars can’t do more than cars from a hundred years ago. They are restricted by the same restriction (usually requiring some sort of road, requiring refueling points…)
We are pushing the boundaries of what LLMs can do, but there seems no indication, that it actually is a suitable tool for automated programming. LLMs are most likely just cars, where you need something that can fly.
I never said that.
No, but the person I had replied to, hence the context for my post, did:
AIs will never be able to abstract away details correctly or design sensible workflows for boutique problems.
I’m not saying LLMs can, or will be able to, do these things. LLMs are likely a dead-end on the road to AGI. Dead-ends are part of progress. The crossbow eventually hits a dead-end in terms of propelling projectiles with ease faster and harder, but that isn’t the end of projectiles. We got cannons, then hand cannons, and then guns.
I’m saying if LLMs are cars, AI is “vehicles.” LLM is a subset of the broader category. We have helicopters and planes. They came later than horse carts and cars, but they’re still vehicles. And used some of what we learned building carts and cars, but also with new ideas and concepts.
And for all we know, someday someone will figure out how to harness the power of gravity like we did with electromagnetism, and we’ll have flying cars. We can’t know, but just because we don’t have the technology now doesn’t mean we never will.
deleted by creator
As a software engineer who uses AI agents daily, let me tell you: now is as good a time as any to learn to code. LLMs won’t replace any developers.
As a graduate from good university in computer science who is struggling to find a job. Go learn something that can be aided by code, but don’t make code the center of your career…
Well the job market for developers is still pretty tight at the moment. I don’t have the insight to say for sure why (though I have some guesses), but I know that for me and every junior developer I know it’s rough out there.
As a junior dev with prior working experience, currently not working as a programmer, yeah. I can only agree.
We might understand AI won’t actually solve the same problems we are able to solve, but the people deciding budgets dont understand that.
They don’t understand a lot. For whatever reason, higher management still thinks things are like a factory, and you just build your software like building a car.
Why? Because that’s the only way they know in the world of MBAs. They can’t speak any other language than “product.”
Having been around for a few decades now I can tell you that the job market comes and goes. Things have been tight before, and there has been more openings than people to work them many times in the past. I can’t tell you when things will turn around, but odds are they will. (this is sadly not helpful if you are one of those currently needing a job)
Well it’s good to know my schooling might not be a complete waste at least
I’ve known more than one person who found a completely different career and never went back. You might take a job in Real Estate as one person I know did and discover you like it better and so all that time in school was a waste now that you know you don’t want to do that. Or maybe not - you might take that job to make ends meet (as I once had to take a non-tech job) and decide you hate enough that you don’t want to go back.
I don’t have the insight to say for sure why (though I have some guesses),
In the USA, there’s a tax break for research teams expiring this year. Supposedly it made software develoent team salaries fully tax deductable.
In the USA, I suspect this is the real motive for using the AI hype train to justify layoffs.
I’m willing to admit “Most CEOs are stupid” also has merit, of course.
“any developers?” bad choice of words. I can promise you with absolute certainty that SOME developers WILL be made redundant because of AI.
not all, not lots, not the majority, but some
LLMs are going to replace some developers, the companies that do that will fold because their product doesn’t work, the developers will get jobs elsewhere.
The market can stay irrational for longer than you can afford not to eat
Learn code anyway. LLMs can’t code worth a shit, so there will be plenty of jobs available to clean up their mess.
LLMs can’t code worth a shit yet. But techbros are determined to change that. The sad reality is that code is just a form of language, and LLMs are good at learning languages. They can’t code worth shit right now, but the progress likely will improve them.
We’ll still need experienced debuggers who can actually code. But in a decade, the broad strokes will likely be done by LLMs, which will vastly shrink the demand for experienced coders.
The sad reality is that code is just a form of language, and LLMs are good at learning languages.
This is debatable. LLMs are prediction machines.
What use is prediction when you are trying to code something new?
The vast majority of coding isn’t making something new, it’s using existing patterns and tools and arranging them to fit a specific use case.
Llms may not be able to create a new framework or design pattern, but neither will most coders in there day to day.
The vast majority of coding isn’t making something new, it’s using existing patterns and tools and arranging them to fit a specific use case.
I would argue that arranging something to fit a specific use case is making something new.
Ask any designer how difficult it is to get a spec sheet from a client and meet their expectations. We’re expecting LLMs to suddenly solve this problem.
Llms may not be able to create a new framework or design pattern
Until they can do this, there is little threat to designers. There will be less grunt work, of course.
Tbh this whole thing made me realize what we really need is a modular automated code bank. There’s so much duplication of effort it’s honestly absurd.
Right we’ve got this scattershot network of libraries but no one’s really been up to the task of taking the next logical steps.
Open source, libraries, frameworks and language development is how this is tackled.
Making software is implementing business logic. It’s the specific nature of whatever problem you are solving which means you can’t use some existing off-the-shelf product.
There are dozens (if not hundreds) of no-code/low-code app builders out there. Things like n8n or ndoe-red.
They get very difficult to maintain at scale.
Right now they are. Who knows what tomorrow will bring.
Compared to just 20 years ago we’re living in the future. You may not have noticed the progress because you’d expect the future to includes hoverboards.
Right now they are. Who knows what tomorrow will bring?
We do. Experienced programmers who have been promised we’re about to be obsolete several times, now. For many of us, this isn’t our first rodeo.
As an expert in computers, there’s two things I can guarantee about the future of computers:
- Computers will just keep getting smarter.
- After decades of getting smarter, computers remain deeply stupid in ways that non-experts cannot imagine. However dumb you think your computer might be, I promise it’s somehow significantly dumber than that.
Lmfao the hardest part of building a product is understanding customer wants and needs. LLMs are incapable of understanding
You described project management
No, it’s the difference between software engineering and software development. If your project manager is handling that, your org is wack
If you’re not understanding why the spec is the way it is, you’re just creating job security for your replacement lol
As far as I’m concerned, my PM represents the customer(s). They spend time with customer feedback, the sales and executive teams to strategize with the company first. I ain’t got time for all that bs.
If that’s not how you work, it’s probably just a smaller org where people have to wear more hats. Nothin’ wrong with that.
I thought that was just the job we give people who are trying their best but can’t really anything.
Learning a language and forming and expressing complex thoughts in an efficient way are three different things.
Learning the syntax of a programming language doesn’t make you a programmer.
Being able to solve complex problems with the programming language makes you a programmer.
Being able to solve complex problems with the programming language in an efficient way makes you a good programmer.
