I've also seen non-technical folks in the corporate and government worlds glom onto vibe coding, only to start flailing like they were caught in a rip current when the underlying technical debt is too strong, even with strong and capable models.
For a certain type of engineer who enjoys really arcane technical problems, yeah, the gravy train may be over, or at least will involve a lot more machines than it used to, as Alan Turing predicted about "mechanical work" becoming a "drudge."[0]
And I think Big Tech is for the foreseeable future a sink for capital in the form of specialized hardware rather than human capital.
I'm old enough to remember when compiled languages like C were bemoaned by people because their skill at crafting assembly by hand was suddenly less important. Or similarly with people who were brilliant with arithmetic, or symbolic manipulation, or manual drafting/lettering. These skills were all paramount at one time, and it's natural that the people who excelled at them will feel a sense of loss.
The real goal, however, will always be to solve real-world problems. Engineering consists of whatever gets you there most effectively.
The best you can find is people claiming that it would reduce the amount of assembly people write, which was true. They also predicted programmers would lose their low level understanding of the machine. That was also true. They didn't predict anything about making human thought non-viable economically. And not one C compiler vendor said their product could cause human extinction.
If AI predictions have the same level of truth as C compiler predictions, then the doomers are right.
Let us now assume, for the sake of argument, that these machines are a genuine possibility, and look at the consequences of constructing them.
To do so would of course meet with great opposition, unless we have advanced greatly in religious toleration from the days of Galileo.
There would be great opposition from the intellectuals who were afraid of being put out of a job.
It is probable though that the intellectuals would be mistaken about this.
There would be plenty to do in trying, say, to keep one's intelligence up to the standard set by the machines, for it seems probable that once the machine thinking method had started, it would not take long to outstrip our feeble powers.
There would be no question of the machines dying, and they would be able to converse with each other to sharpen their wits.
At some stage therefore we should have to expect the machines to take control, in the way that is mentioned in Samuel Butler's *Erewhon*.
In case it's not obvious: things do not work out in Erewhon :(Most companies that are running the “how many SWEs can we fire before it all breaks” experiment have learned it’s a failed experiment - this validated by every single major company/product outage where their uptime went below 90% this year.
I can’t help but think this anxiety and fear is just marketing. chill out everyone
A lot of people seem to think they can pick up an LLM and suddenly write software solutions as good as a seasoned SWE. But it turns into this terrible mess that's thrown out all engineering fundamentals, solving the symptoms instead of the cause.
If I were just writing boilerplate code with an LLM all day I'd be very scared for my job too.
This is a big problem. When this kind of thing gets launched, its all fanfare, this is the future, get on the boat or die.
But then it fails miserably and its like it never existed
Here are few anecdotes:
1. I've a friend who owns a PT (physical therapy) business. He's not even a tech guy, but uses claude to manage most of the business aspects, and he has told me repeatedly that he would've to spend thousands of dollars to get things done, and instead he pays 250 per year.
2. Another friend is smart, and is a jack-of-all-trades type in the software/hardware/firmware business. He is solo bootstrapping by doing all the things from pcb to webapp. He doesn't see hiring anyone for the niche business.
3. A 65 year old friend of mine who never in any tech career was able to build website for his side business (that website looks really good btw) and he now talks about json and ocr.
You are the one who is out of touch with reality.
That seems like it would be such a paradigm shift that the second and third order effects are so crazy and nobody can predict what that world will actually look like
This is engineering thinking, not business thinking. The question is, did the uptime drops cost the company more money than the firings saved? If the answer is no (which for megacorps like Microsoft it is) then the experiment was a success.
He wrote a series of blog posts in which he talked about how Gas Town failed and blamed it on Opus 4.7 changes causing all his workflows to fail. Then he announced Gas City as a rewrite.
At some point after that he seemed to give up on the whole idea and instead just started working on reviving an old graphical MUD style game he made where I guess now he spends $10,000 a month on agents to build a game that currently has 0 concurrent players (4 concurrent 24 hour peak) on Steam:
https://steamdb.info/app/1541710/charts/
The game is available from other platforms than Steam so maybe he's got a bit more users than that across the platforms, and his stated plan is to bilk "whales", so maybe that's working out for him? But it still seems like he's lighting massive amounts of money on fire in what appears to be a prolonged state of ai psychosis.
Must be nice to be that financially secure.
if the hype was all true, i could just prompt the llm to make me money and it would be able to do it, but it's not possible
but at the same time I think the companies are pumping this idea that SWE is over and stoking anxiety and we have yet to see how bad things get when everyone can write software without thinking about it, we will see what insecure slop gets produced and it's going to be huge.
The issues with AI security are quite big as well, people who will rely on just an LLM to be the intelligence will get burnt I am sure especially in operations, prompt injection is still an issue, would someone running LLM's notice an provider endpoint getting hijacked and acting maliciously and inserting malware into production servers.
The future is going to be a strange one for sure but being able to turn your brain of and get your LLM to run you servers write you servers is not going to work alone.
I have also yet it write a good fairly novel program without babysitting (quite trivial babysitting sometimes) but stuff that is bread and butter for us, for those who don't have those skills, will they really know how to explain the issues they may run into that requires creativity, for example telling it to hack something together with curl and jq, knowing why a certain thing with the sandbox is not working.
I think these issues are not as easy to resolve with just better tooling and they will soon get annoyed at prompting.
Additionally I think Gas Town shows the sort of almost mentally ill mentality that is baked into some of these tools and ideas, Everything I read of Gas Town was so strange and the amount of tokens used to anthropomorphize agents.
My friends at FAANG companies would disagree here. If anything, it seems to be the exact opposite.
Sure, it's not a end-to-end 0 touch factory, but a majority of folks I talk to these days that I consider good engineers aren't writing _any_ code anymore. They do that as a hobby outside of work.
Now it seems like it's the sunset of this way of making a living. I think people who manage to move up the food chain a bit will succeed but its hard to advise any junior in the industry exactly what that looks like.
What is the common thread?
These jobs depend on some combination of physical skill, human connection, trust, judgment and accountability, and such jobs are out of reach of AI.
Then they will be able to do more and more, and 1 nurse will be able to do the job of 4-5 2020 nurses.
Until that nurse is a dude from Mexico [where im from/based] giving instructions to all the robots.
If predictions about robotics are accurate, things will get even more difficult in the next ten years as humanoid robots mature.
Arguably, software has times that require physical presence, human connection, trust, judgment, and accountability.
None of these end up really being out of reach of AI if Anthropic can justify this valuation. They're just slower to roll out.
Of course the elderly and frail and people too afraid to touch something lest they break it will still call an expert, and original build and installations will need those trained, but I could see vast swaths of less technically skilled people opting for a go at routine repairs which could really hollow out the demand.
LLMs are hugely suited to code because it's token heavy and the compiler is the RLHF verifier.
The LLMs are still very bad at concepts.
What we see in code does not translate to other, arbitrary domains.
Legal - yes, due to text heavy content with tons of great examples as well.
Everything else ... not really. At least not LLMs as we understand them today.
I'm not the only one that had that idea, they're advertising these things as features now.
I don't know what's next, but keep in mind that today's LLM's are the worst they'll ever be.
Nearly everything will (eventually) be impacted.
Therapists? People are using LLMs for “free” help.
Physical trainers? Just upload a video of your workout to an LLM for tuning advice and other exercises you can focus on.
Appliance/vehicle repair? Just DIY it with an LLM.
Teachers? Just teach yourself with an LLM.
I could go on.
To be clear, I’m not saying LLMs are better at this. But just saying LLMs are impacting nearly every field.
EDIT: just to clarify some of the responses talking about full replacement -- that's not my point here. I'm talking about impacting, not replacing. Software engineering likely won't get fully replaced (anytime soon), either -- but it's definitely being impacted.
I think this is a very good template for what sort of skills will survive the LLM-apocalypse and what won't. LLMs are the great homogenizer. Any type of knowledge work where there is a broadly-accepted set of best practices for how to do it is gonna get outsourced to an AI.
But AI is terrible at doing anything unique, personalized, and based on your own desires, because it doesn't have your desires. It's building on a corpus of shared knowledge, but there is a large set of knowledge that is unshared, and it can't do anything for that. A large portion of business, for example, is secrets and lies, and an LLM isn't going to be able to help you much with that. At its best it can set up the frameworks and tools to help you keep secrets and tell lies.
And then a large portion of life is personal preferences, the things that make you unique and distinct from every other human, and an LLM can't help you with that either. There is a reason I don't have a nanny raise my kids, even though we could afford it. If I did that, it would be the nanny raising my kids, which defeats a lot of the point of having kids in the first place. LLMs have the same problem.
I keep trying to nudge them to be relevant with specific examples of a few shows I had watched that were very popular at the time but now considered "old/boring/not remastered" so don't make it into the Best X Shows of XXs lists that emphasize modern tastes. But the recommendations were either the same things I already knew about, or were just wrong/irrelevant for various reasons.
Recently I stumbled onto a Reddit post from years ago with a similar request to mine, and both the question itself and top answers were a goldmine of suggestions that were never recommended and I hadn't heard of. I knew within seconds of looking at the IMDB that they would be exactly what I wanted, so it's not like the AI recs are failing just because of idiosyncratic taste. They just don't seem to search from a deep enough info source, have been RLed into conformity, or some other disconnect between intent and execution.
Or — more disturbingly — what if LLMs lock us into our current state of development forever? This is all we will ever be; nothing will fundamentally change from here on out.
It feels a bit like cryptocurrency and LLMs were dropshipped alien technologies specifically designed to bind us and drain our resources by leveraging our own greed.
Done right, that accreditation could be objectively superior to brand names like Harvard, etc.
The real danger for teachers is society deciding people don’t need to be educated. Although that would be a problem for all of us actually.
More workers competing means lower pay. Less good paying jobs means less customers. Less customers means less jobs. This is an over simplification, but is one possible future.
I'd be happier if my prediction turns out wrong.