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.
border-radius could have knocked me out of the game, but I decided to keep going. And I'm glad I did.
Love this statement, 100% agree.
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.
Part of this, I think, is because the people who had the inclination and ability to pick up programing have, for decades, been nudged into programming from whatever job they had before, by comp, co-workers, managers, and opportunity availability, so they’re already heavily concentrated in programming even if they’re perfectly capable of doing other jobs (perhaps with some brief re-training).
I think other roles need to fear programmers who studied a new role for a couple weekends more than programmers need to fear LLMs. If we’re bumped out of full-time programming… guess where we’re going, and what we’ll be way better at doing than them?
[edit] part of that I wrote unclearly: the main thing about people being siphoned out of other roles into programming for the last 3ish decades is that most of the people with the combo of willingness and ability to get decent at programming… are programmers. It’ll tend to be harder to train other people to be good enough at using their computers to maximally benefit from LLMs (they already usually aren’t very good at using their non-LLM computer tools, it’s not like this is a new problem) than to shift programmers into other roles, if programming as a dedicated job in-fact is diminished in a durable way and at a mass scale.
So the question is if demand for software will expand by more than the productivity improvement it creates, in this fiscal environment. I think it will, but it will be for Software as a (Specific) Solution rather than the classic Software as a Service (delivering measurable outcomes vs. delivering tools for people to use), and this will come with a shift in the types of roles hired for. Less "routine" software developers, more people in product design, more domain specialists, more system architects and integration engineers, QA, security, compliance & auditing, etc...
Just anecdotally: I am finishing up on a project proposal for a company that would pre-AI probably never considered this, but if it's greenlighted, I will need a team in two months and I'm already nervous that I may not find the right selection of people... Many such initiatives are started now because AI makes everyone more ambitious, wanting to automate everything but reliably, which requires skilled software engineers in conjunction with AI workflows.
Within couple years there is definitely going to be a shortage of people who understand how systems work. The only black swan possibility is the enterprise industry getting too small cause of heavy consolidation.
Are you solving a shared agreed upon real problem or just doing leetcode on arbitrary meaningless data structure?
There will still be jobs but orders of magnitude fewer jobs.
Closer to reality that existed before the 08 crash and ZIRP fueled ponzi schemes fueled the web SaaS mania.
US doesn't need 4+ million software devs
World doesn't need 50+ million software devs
...when big tech is just shipping black boxes that boot models to handle local inference
It just needs enough jobs to solve that big tech production pipeline
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 :(But the thing is that with AI the number of people willing to pay you to solve their art problem has dramatically fallen. The total number of jobs in art in general has plummeted. The opportunities to make a living or develop yourself with art have similarly plunged.
So there will always be another problem but that doesn't mean there will always be someone willing to pay you to solve it if they can get by without you.
Is it? It looks more like a forgettable moment to me. We're talking about an IPO, how many headline-making IPOs can you actually remember?
> as a SWE the gravy train is basically over in my view
That is not a result of the IPO even if you are right, and I have my doubts.
> they figured out how to make a program that does the job
No they figured out how to make software that makes the job easier. How is this any different from the introduction of optimizing compilers or type systems or any of the other software tools we all use? It makes programmers more productive, which means programmers can take on larger projects.
Coding agents have limits and the managers who forced their SWEs to use them are starting to figure that out. Using agents to both write code and review code sounds cool in theory but in practice it is not such a great idea -- it means that nobody actually understands the codebase and therefore nobody can actually debug the code when things break. No, you cannot just assume that you can also throw an agent at the problem when things break -- because the breakage might mean you lost access to the agent or the agent lost access to your system and cannot help you.
> difficult to grasp how much life will change in the coming years
Not too difficult to grasp. Shareholders will demand returns and Anthropic will have to deliver those returns. All the sky-high dreams about RSI and the next phase of human existence are going to evaporate and profit is going to be front and center.
Another possibility: Anthropic fails to survive the next recession and the companies that did not allow their employees to become dependent on Claude will be the winners.
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
So we are not waiting for the tech to get good enough, we're waiting for humans to understand and integrate it.
I don't work in tech, rather manufacturing, and our company is barely using any AI (or using it, but only like AI mode in Google) despite me knowing that we could easily use it to cut heads. The automation we are leaving on the table is huge.
But I'm not gonna be the one to go to my bosses and say "Hey, you want me to show you what we can do with SOTA models and agents?". For now I'll just stick to using my accounts to leverage my job.
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.
there is massive lose in revenue for a business and or infrastructure that people rely on goes down
if you’re working anything in this way, no one is just trusting a clanker. some thought leader might say the world is trusting an AI, but it’s not true. every company internally that is more than 2 people isn't trusting a bot, and it's well understand that it's a horrible idea (even by regular bullish AI business heads)
so yes, I have the bot help me write a basic website for me, and it can one shot the typescript express app-layer boilerplate for me, but if you’re serving even the most basic app at scale with consequences, the bot isn't mitigating the concerns and engineering decisions that go into this work.
a concrete real world example: literally msft outlook went down a week ago. it was out for god knows how many actual hours. it still might be partly down tbh. this is technology that is “solved” and is dumb simple in terms of complexity. their product was out for 48+ hours. and the status messages shared publicly by the engineering team did not only convey incompetence, but to paraphrase, the message they sent out was along the lines of, “We think we found the source of the problem, but unsure, the problem might be larger”
how much you want to bet they had a team of 30+ engs running around with claude asking the equivalent of, “please fix outage, no mistakes”
I’m still not convinced the outage is fully recovered because I still get errors.
point is, yes there isn’t a giant 100 football field wide moat between regular people and programming anymore. smart people can do basic stuff and less inclined people can build a static website now. if you’re worried about that, go study more and you'll learn the depths of this field go WAY beyond typing code
EDIT: typos
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.
AI can write syntactically perfect code but barely has a sense of architecture.
It needs tons of 'guide wires' to keep it from going off track.
Unless you're building basic SaaS, it's still a 'research and coding assistant'.
AI has never produced non-messy architecture that I did not have to go in, re-think and guide a re-write.
The business of software still exists and it needs to people to run it. It will require fewer people than before, and as a result cost of software will come down. There will be increased competition, and big SaaS companies will face fresh competition from upstarts. As a result there will be pressure to innovate.
AI can't innovate. This is one of the key limitations of AI. Not all humans can either. But those who can will be in increased demand.
As software becomes cheaper, the total consumption of software increases (See Jevons paradox.) Cheaper development makes smaller markets viable. There will be software available to meet needs that are unmet today because the market is too small. This will increase the number of software jobs. It won't all be coding jobs, it will be jobs that run the software business (see first paragraph).
2 options:
A) A real person with a job title that's not SWE. Will it be managers / product managers / business owners / something else? Whatever the job title, those people need SWE skills and experience to manage the agents. It is a very high ask for non-SWE job title to absorb both the skill and time commitment to drive AI-first software work.
B) AI Agent. But how does the AI Agent get setup with what it needs to run in a loop, who investigates and steers the agent course when it deliver something that doesn't work? It will needs someone SWE skills/experience.
In either case sounds like you need someone with SWE skillset in your company... so not sure why the job would go away.
Now, I only see how AI tools stagnated in the last few years. To me, Opus 5.x solves all the issues I've had with programming. I no longer need more. They also failed with massive adoption of non-tech people beyond using AI tools as simple guidance and search tools with some simple programming capabilities.
I don't see any way for them to grow more (maybe I am blind), and open weight models are only going to get better, which will sink them further.
Only time will tell I guess.
I've been building for the past few weeks a fancy text editor for myself to work with AI in a more sustainable and sane way. I don't have much time for this, so I've entirely vibecoded it: I don't look at the code and I've provided only some initial design and engineering decisions, debug hints, etc. on top of natural language feature requests.
For about a week using the strongest Claude models (Fable 5.2 and Opus 5.5) this went great, and I had a minimally functional tool that worked very nicely. It seems that I hit some sort of complexity ceiling for those models though. The second week, every feature I try adding creates a new type of jank that can (at least by me) be only vaguely characterized. Most recently inserting into a line (even in the middle of the viewport) causes the viewport to jitter about.
Our work will change, we will spend less time on tedium, we will become more ambitious. Barring an architectural paradigm shift we will still need people with strong engineering and design instincts, and even more importantly the ability to write reference code and formal specifications codifying desired behaviors.
At my current company - not a software company but we have a dozen plus software engineers - and while we are expecting them to get more done, and their productivity is increasing dramatically, the list of things we want them to do is growing faster than they’re finishing things.
Things that were huge projects, that would have been hugely expensive for us to do, are now affordable and can be done in a couple of months or days, it’s no longer the most important things get the developers time, but now people in HR, finance, etc are getting their projects “on the list”
Now, the real difference is whether I write the code in English, Python, Go, C++, etc. The new tooling (AI) forces me to express and crystalize my thoughts in natural language - and it ends there. Design documents, consensus, feature exploration, testing, knock on thinking.. these are actually now MORE important than before. I've outsourced the mechanics and syntax of writing code but not the organization, design, and layout for scale.
This is pretty awesome.
What Anthropic has excelled at isn’t engineering. It’s convincing your boss that you can be replaced with a Claude Code subscription. Just look at the leaked Claude Code source. Or read their “engineering,” blog. It’s all marketing glitz.
The reason for all this marketing is because they rent their compute and they need every software developer to need them. They need us to depend on them to do our jobs. Because this isn’t a profitable venture. It’s resting on a house of cards.
Anthropic has to deal with the very real possibility that their providers go under. And that other harnesses on alternative models get better, cheaper, etc. People seem to be pretty price sensitive after the token-maxing rush didn’t work out.
I think there are still plenty of reasons to believe that our profession isn’t going anywhere. Changed, sure. It always changes. But it’s not over and it’s not, “solved.”
I'm not scared for the future of software engineers and computer scientists.
Massively sad moment, a few handful of already well to do individuals got even richer, allowing business to cut wages. Sure it’ll create new jobs / industries but the short term impact is not something I look forward to.
Everybody is an architect now.
Don’t worry.
The programmers were/are simply capable and in relative short supply. Even with that they earned the companies massive profits. (I have completely different issues with my high-IQ nerd peers... more what they were working on.)
Now the parasite companies can simply harvest all of that value for nothing. Big Tech is moving from a gravy train to a free ride.
Defintely a very Capital Guy perspective top comment.
They certainly did not. LLMs are way worse than a competent human programmer.
More prosperity than you've ever had.
2/5
The gravy train is just starting
I have seen coding agents get better as both the models and harnesses improved. I used to have to manage context by hand, I used to review every line. I no longer do.
However, I will tell you this from my experience with AI. Software is not any easier. The same complexities are still there. What's different isn't the kind but the scale. A 2 person team can do what a 20 person team did 5 years ago.
What is still required is a shit ton of feedback to build what you want. It's still hard to build complex software even with AI. In fact in some ways harder because AI agents will go down rabbit holes and create entropy faster than you can say Bingo.
Productivity is waste, it has to be, it's thermodynamics. The person's job is to fight entropy in the system. That has been true before AI and true today.
We have to remember, software is there to serve us and fix our problems. AI doesn't change that.
Just today Opus 5.5 tried to send a Slack message to one of my teammate when it had to edit a code file. When asked why, it said it just "picked the wrong tool".
LLMs are borderline useless, AI has led to 0 increase in actual value produced. It's all powder in front of our eyes.
I genuinely cannot take anyone seriously when they say they're afraid of being replaced by AI. The only explication possible (unless they're astroturfing), is that they literally never look at what LLMs produce and always run in "Trust all tools" mode without ever looking.
It’s a dark coincidence that “Turkey Day” is next week in the USA. Many turkeys that will be on our plates never imagined THAT fate.
The gullibility of people is astounding, and they never learn.