Completely absent is quantity of jobs. If e.g. ml engineer job postings go +40% from 200 to 280 and writer job postings go -50% (over two years) from 20000 to 10000, then we have a better idea of the impact.
Without those data this report isn’t really quantifying impact on “180M jobs.”
Also decline in Security engineer by 0.35% doesn't make sense by conventional wisdom. Shouldn't it be increasing due to increased demand for security in all ai integrations?
I wouldn't be surprised by a drop in security postings. Quite a few companies view security as an "overhead" so the siren call of reducing that overhead by introducing AI is a thing.
Also for a lot of jobs in security it's pretty hard to measure how well it's being done, so if the AI based solutions are worse, that might not show up for a while
We also need to consider the confounding effect of corporate performance and recession expectations.
Cost centers in businesses are early canaries of expected pain, and a reduction in security roles may reflect belt-tightening irrespective of AI impact.
> for a lot of jobs in security it's pretty hard to measure how well it's being done
Nothing going wrong: “What are we paying you for?”
Everything going wrong: “What are we paying you for?”
It’s a no-win situation unless you manage to score a division manager who understands security and understands the reports a good security division produces. And most importantly, understands that no news is good news.
Security products and practitioners are the classic snake oil salesmen. They are actually sales and marketing roles for help closing deals by emphasizing some security aspect. True security comes from general IT practices followed by engineers themselves.
I would be wary of making categorical claims like this, but it's unfortunately true that "security" field hasn't been doing well in a long, long time now.
Half the field is B2B "magic bullet" solutions like CrowdStrike and all the associated sales tactics - with pitches that boil down to "you give us money, we make your security issues go away". Half of what remains is mandatory certifications and other flavors of checklist-obsessed cargo cultists - often CYA-driven, often demanding the adoption of the fancy acronym of the day, regardless of the real threat profiles. Then you get the "security snake oil" - "magic bullet" systems that don't work, never did and never will, but are supported by the right influence groups and get the right pockets lined, and so are used anyway. DRM systems like WideVine and PlayReady being the prime examples. Then there are the corporate "security of our business model" shills - who pay lip service to "security", but have the true aims of "prevent anyone we don't like from doing anything that can harm our revenue streams" - with Apple being a common example.
And about a fifth of the field is people who do actual security work, and keep the sky from falling.
I agree with you totally, although I'd venture to guess 20% is way too high. I'd say you have about 10% people doing security work, 15% doing compliance, and the rest are consuming oxygen.
It's a growth field, so you have lots of idiots getting certifications and stupid jobs. Reminds me of the 90s when I started, and companies were paying MCSE's (ie read a book, hit next-next-finish in Windows NT) more than software engineers in some markets.
> True security comes from general IT practices followed by engineers themselves.
I have yet to meet an org whose engineers care about security, or who would not compromise security if secure practices got in the way of shipping a product or feature.
I consistently see this commenter making a single comment, of questionable relevance, expressing a strong opinion which isn't particularly thoughtful or interesting or true. Then they ignore the pushback and move on to the next thread, where they post another tangential hot take. I'm not at all surprised at the result. Those comments attract a lot of downvote because they aren't very good.
This thread is a microcosm of that. They went on a tangent from a tangent to express how little they think of their colleagues working in security. It wasn't out of curiosity, it didn't raise interesting questions or provoke interesting debate. They didn't defend or substantiate their opinion so that they and we could learn something from it. It was just a drive-by flamebait to stir the pot and express derision. It should be downvoted; it's a bad comment.
Perhaps that pattern is difficult to see when their hot takes align with your own takes.
On the continuum of approval, where at one end there is endorsement and at the other disapproval, it's somewhere in-between. Even I who made the observation don't know exactly where. Sometimes something jumps out at me and I don't yet why.
It could be an incorrect observation. Some of what they said seems true, some false. I don't know enough about security specifically to say. I know a lot about other things to know he said some things that are true.
It's astonishment at perhaps some kind of law of the universe that things that seem one way may be a different way.
It's an exclamation at the poetic irony of someone expressing there's gray area in some things gets downvoted and their comments are in gray colour.
It's a way to introduce myself, to say hi Mr Monero user, and pass a super secret note.
I meant no disrespect.
Perhaps my reply here is astonishment at how interpretation of words may depend on imagination. As if words alone aren't enough.
I post my view that is against the HN hive mind and don't always feel like rebutting the same hive mind talking points again and again. I like to post to prove there is an alternative view out there
I'm also guilty of what they accuse you of. Sometimes my internet comments are not made for the purpose of sparking discussion, but more of a "vent" where I know my take is not popular but I feel the need to throw it out there anyway. The comment is more for "me" than anyone else. And, yeah.. that makes it a bad comment lol.
I also just love playing devil's advocate, and I'm adverse to hivemindy-feeling opinions (even when I share them). Maybe this all describes you, too.
I don't have a problem with people doing that as long as they don't pretend that every other commenter holds the same contrary opinion and that the downvotes indicate they're too sensitive to discuss such things, or other similar rationalizations. If you want to leave some drive-by snark without rationalizing it as being about other people, it's not my favorite kind of comment but I'm not going to object to it either.
The downvoting functionality here and in other forums can mean many things. It isn't a precise thing. If it is precise I'm not finding a clear definition. It can mean I disagree, this is boring, this is false, this made me sad, I don't like reading this, I don't like this user, I'm tired, etc.
One plausible interpretation of a downvote without a comment is drive-by snark without rationalizing.
I don't know your motivations but I know the "HN hive mind" isn't the problem. When you do engage with people who disagree with you, it usually becomes evident to me that there isn't much substance behind your views and that you struggle to disagree amicably. I also see lots of people on HN with a similar perspective to yours who don't have the same problems or engage in the same patterns of behavior.
The facts are that HN has a diverse set of perspectives with many conservative/libertarian commenters who would align with you, but that your comments are frequently shallow flamebait. Though I have seen a couple good points you've made, as well. Do with that information what you will.
I disagree entirely, I don’t even post very frequently so it’s surprising I have someone tracking my posts. The shorter a comment the better it is, if the same opinion that takes an essay can be distilled into a sentence
My first comment on this whole thread was how security in tech is theater, and the sellers mostly snake oil salesmen. I’m not the first to make this observation and I don’t think it’s wrong. Which is why employment in the sector is down, full circle to the OP
Go ahead, take that slim, speculative, tangential connection and interpret it as permission to inject your hot take into the discussion. Decline to elaborate when your supposition is challenged. You've every right to do that.
Just don't pretend that it's for our benefit or that we downvote it because we're unthinking drones, or that you decline to elaborate because we're simply not capable of having the discussion.
I tell you this because if I were insulating myself inside a bubble and rationalizing my interactions with those who disagree with me as being the reflexive behavior of a hive mind, I would hope someone would point that out to me. So here it is; again, do with the opportunity what you will.
> Just don't pretend that it's for our benefit or that we downvote it because we're unthinking drones
The reason many comments are downvoted on HN in general is most often unknown to me. One interpretation is that it's a major flaw in HN.
This design decision by HN could be intentional, as a trade-off to achieve something else. For example, it could be done to have high velocity of discussion. High velocity could preserve an invariant of keeping or pulling users on the site.
If it's a trade-off, which would suggest something is given up for it, it might be worth exploring what's given up.
This website is full of unthinking drones acting with hive mind behavior, that is my contention, and I think very differently, not just on here but with almost everyone I engage with. However I succeed over and over with asymmetric bets on a wide variety of things, including and especially tech and making money in tech, so if we compare bank accounts and career trajectories, investments, etc. it would be wise to let me speak
Overweighing people's opinions on matters they demonstrate a shallow engagement with on account of their success in other areas is cargo culting. Maybe you should worry less about HN and more about your own reflexes to accept bad ideas from yourself and others based on their wealth. Maybe succeeding in contrarian bets doesn't make you "correct" in some moral sense, but only successful in the trade.
It's easy for me to believe you're an intelligent person who's accomplished impressive things. But it wouldn't contradict anything I've said.
I own a software business with hundreds of employees I built from nothing. I know all about tech security, hiring security guys, etc. It is a cost center, the directors / VPs / CSOs are overpaid salesmen, 99% of the products you want are features from cloud vendors or provided by standard tools like device management or password managers. I totally get the patina desired by large corporations who need to show wall street they care about security. When I box check I use a super cheap cut rate firm who can check boxes for me at the lowest cost because it is total bullshit. I'm sorry this is the reality
It's understandable to have sour grapes after having some bad experiences, but what I'm hearing is unrelated to any impact AI is having on the job market. You're talking about security products you think are snake oil and executives you think are overpaid; that is unrelated to trends in job postings for security professionals, working at software companies, and how that might be impacted from AI.
This is what I've been saying. You've got some random grievance, you want to take this discussion as an opportunity to get it off your chest. But you don't want to engage with people challenging your ideas. And when for whatever reason you do explain yourself to me, your explanation is "I am wealthy and successful, so I must be right. Those who disagree with me are an undifferentiated mass of imbeciles that I have nothing to learn from."
If that's how you want to live, it's your right. You're only cheating yourself, so maybe I'll just shut up and let you get on with it.
I could easily see those just running tool and then printing report being replaced by script running the tool and passing results to LLM and then sending report.
From what I can see, being closer than the average engineer to the space (but not an expert on my own), a few things are happening:
* Engineers are being pushed for ownership of security more directly. You still need someone on the team to guide and support them, but they're not going to be directly involved all of the time.
* Significant amounts of automation and centralized security. Supply chain management is a double edge sword. It does open up vulnerabilities, but you can simply pay one of the SaaS companies in the space to help with a lot of the heavy lifting.
Bold of you to assume there is any demand for security in AI integrations. It's like 90s web browsers, everyone's running random MCP servers that do god knows what.
I think it's because companies are moving away from in-house security and hiring 3rd party companies for security work. It also depends on the time of year this was taken. Q4 is the busiest time for security. Q1 is the slowest.
I'm a security consultant and work with multiple companies that provide security services. Work has increased massively in the last year.
Comparing US job openings from 2024 to 2025 seems to be selective. Is the job in sudden decline? Or a reset to a norm? Or is the decline something that has been happening over even a longer timeframe?
Taking two years and drawing conclusions from those two years seems to miss larger pictures - especially when the mess of Covid and the pandemic years and that job market is mixed in with this. Yea, that's half a decade back but companies are still trying to figure out how much staff they need where.
Photography? Yes, that's likely been impacted. Compare https://web.archive.org/web/20230124085038/https://www.bls.g... and https://www.bls.gov/ooh/media-and-communication/photographer... and the rate of growth has changed. However I'll also draw attention to the estimates that there's only a few thousand open positions with that classification each year. This includes self employed stock photographers and artists - corporate photographers have been a hard thing to get for a much longer timeframe (I looked into it a little bit back in '09). Additionally, artistic photography is impacted by the amount of money that regular people are willing to spend at art festivals and the like - that's gone down irrespective of AI.
I agree with this. I'm a product designer, and apparently there's "only" been a decrease in job postings of 2.6%. That doesn't seem bad, however in late 2022 through all of 2023 people were getting canned left and right and the market was incredibly tough. Can't really have big decreases when you're near the bottom of the hole.
I found this interesting: "Still, despite all the hype about how AI coding tools will replace software engineers, software engineering is still one of the most secure jobs you can have today, relative to most other white-collar jobs."
There is a lot of induced demand in software engineering. We are still in the realm where cheaper software means that people want more and more complex software. And that demand increase is more than enough to offset any efficiency increases
Meanwhile the amount of accounting that has to be done is pretty inelastic. Whenever accounting gets more efficient you just reduce the number of accountants instead of doing more accounting
Creative is somewhere in between. Not completely static demand, but not extremely elastic either. The healthy rise in postings for creative directors indicates that the cost reduction has lead to more art being done, but the increase in demand isn't big enough to offset the job losses in the rank and file positions
It's funny, almost every conversation I have about the fact that I work as a software engineer with new people I meet nowadays seems to include them asking if I'm worried about AI stealing my job. Maybe it's something that people ask everyone nowadays regardless of what industry they work in, but at least as far as I can tell, the type of work I do doesn't seem in any apparent danger of being replaced by AI any time soon.
as of today that's probably true, but if the labs manage to keep increasing the 50% time horizon (defined by METR as 'the length of tasks (measured by how long they take human professionals) that it can complete autonomously with 50% probability') at the current pace, it might not be for long. Exponentials are hard enough to forecast if you kinda sorta know the parameters, and we don't have that comfort today.
It's because journalists are still big mad that the internet wrecked the newspaper business, therefore the news constantly reports lies about how the tech industry is collapsing. The more news you watch and the less personal contact you have with the industry, the more likely you are to believe that techies are jumping out of office windows in despair (hi, Mom).
The job market is terrible, pay is stagnate and remote work is being taken away and the biggest companies despite profits are laying off large amounts of people. I don't think that's fud
Is there something the the SV charter that says companies operating there must do that?
Nobody seems to want to answer the question so I assume nobody wants to confront that a company isn't the bad guy when it fires people it doesn't need. And, if it turns out it did need them, the company will suffer. Usually that doesn't happen though and so I tend to trust that these companies understand their business needs more than I do.
It sucks when people lose their jobs but if there's no job at their company for them anymore, their employer isn't a charity. Just like if a company could no longer afford to pay they employees, the employees shouldn't hang around and work for nothing. You might as well be demonizing people who file for divorce for making their partner single.
That's never how layoffs are done. If you survive them your workload just increases and the employees picked are based on nothing or something arbitrary.
Then your company has a reputation for being unreliable for stable employment and people that can pick where they want to go will go else where. It also kills morale and remaining high performers will go somewhere else. Now the company laying off is stuck with low performers that don't have other options.
I don't see how you can hold that this job market is "terrible" in a frame of reference of the last 50 years, unless you were born fully-grown in a vat exactly 4 years ago.
No, and no individual person can rely on their personal experience to gauge the job market, including you. However if you wish to assert that it is currently more difficult to get a job as a software developer than it was in late 2015, aggregate data suggest this is false. Also, looking back a mere 10 years does not dispel the general impression that you lack context.
If you look at employment in, for example, NAICS 518 "Computing infrastructure providers, data processing, web hosting, and related services" which is one of the larger BLS categories for our industry, the numbers are at all-time highs, having doubled since 2011. An example of a bad job market for software developers was 2001-2011, when this sector shrank by a third.
Until Software Engineers have automated away all the other jobs with AI & software they'll be safe. That's going to take a long time.
Replacing software engineers with AI only affects the bottom line of software companies. Companies are usually fine with increasing the bottom line if they can exponentially increase the top line. I think software engineers will provide that capability for at least the next 10-15 years.
I think that's an over-simplistic view - at the moment there are many, many software engineers hired by companies who are betting on AI being madly profitable. If those expectations change, we could see more cascading layoffs, which will mean those engineers will go looking at more traditional places like banks, which means they'll stop hiring, which means it'll be harder to someone who is looking for a new job to find one, even though not all jobs have yet been automated.
True, besides actual software engineering is small part of overall IT/computer related work. There are far more analysts, managers(project, program, IT, agile etc), QA, operations and so on. So engineers who employers think are capable of leveraging AI and do 20 s/w engineer worth of work with 5 people will remain in demand for time being.
Even without outright layoffs one can see how fast leverage of average IT engineer is disappearing. After 20 years of experience my value or feedback matters less than when I was 4 years into paid job. And it has far less to with AI so far.
Most custom work of past is just a library, component or framework to use. And those are mandated to be used as it much easier to hire/replace teams to work on those.
Now It may be always be true to have reusable components created but growth of IT industry kept people employed in ever greater numbers. However now it seems to be reaching limit. Leaving aside highly visible layoffs by US tech giants, growth is fading in countries like India with huge IT offshoring workforce. There are millions upon millions jobless fresh graduates waiting to get jobs with some IT degree.
Did the invention of compilers eliminate the need for programmers or make them more productive and valuable? LLM coding is really not in the most abstract sense any different from compiling a higher level language to a lower level language.
> LLM coding is really not in the most abstract sense any different from compiling a higher level language to a lower level language.
Hard disagree here. Anecdotally, know a few people who can't write a Java program that will compile, who can leverage ChatGPT to produce functional websites.
A good friend of mine ChatGPTed his way into a masters degree that involved a lot of coding. A good 97% of his degree was done by AI, and the other 3% was me helping him troubleshoot he couldn't get AI to solve.
LLM is vastly different from a compiler/translator. Despite the joke, you can't just fire up Python with import website and have a functional website. But you can basically do that with LLMs, which will then add features as requested. It's not perfect, nor guaranteed to be functional, but it is quite a bit more capable than a compiler is for such tasks.
At my work, the sales guys are using AI tools to rapidly prototype features on our website with prospects. While it doesn't do all the work, it can produce useful HTML templates that the front-end team can make functional.
Certainly the jobs that were around for bespoke compilers were eliminated with the unification of compilers (GCC, Microsoft Visual C++). How many of those people transitioned to other roles I don't know. But the number of compiler jobs has been declining forever at this point.
At the end of the day, "AI" is just another programming language, albeit one that is much more accessible to the layman. When using AI, you become a software engineer. So it stands to reason that software engineering jobs are strong.
But what about pay? Elevator operator jobs have never been more prevalent, but increased accessibility to the layman pushed the price to zero.
idk man, plenty of people have "ai" gf/bf/therapist, ask "ai" for vacation trip ideas, recipes, gym workouts, &c. I wouldn't even be surprised if most tokens were used on non software engineering tasks.
I have a zombie developer, coder, idk how to call them, in my team who doesn't talk to anyone, writes shit tier PRs and spends all day long talking to chatgpt. They're a prompter, a chatter, a waste of money, but certainly not an engineer
> idk man, plenty of people have "ai" gf/bf/therapist, ask "ai" for vacation trip ideas, recipes, gym workouts, &c.
Not without engineering a system to get the AI to generate a response they don't.
> non software engineering tasks.
There is no non-software engineering task when using an LLM. All LLM use first requires designing software to get the LLM to do anything. An LLM will not autonomously act on its own. It can only respond in accordance to the the program you have designed for it.
That program can be written in natural language, granted, but software programs are not limited by the particular language they are written in. Its not somehow not a program just because you didn't write machine code on punch cards.
> the definition of engineer says nothing of quality
The definition you used makes someone who maintain a twitter account or use a coffee machine an "engineer"... everyone is an engineer by that definition really
It's not just the definition I use, it is the most common definition people use according to the records we keep.
And yes, someone who builds a system to generate Twitter content or maintains a coffee machine are definitely engineers.
Professional engineering societies don't like that fact and often try to usurp the term for their own financial gain, but that's not how English works. Definitions derive from use, not what a small group of people wish were so, and the record that keeps track of use is abundantly clear how the word is most commonly used.
Slop doesn't suggest the difference is only in quality, but also in form. Words have meanings, ya know?
I can have an LLM generate me code-like text all day long, but that doesn't means it's building a system. Just like I can eat Chicken McNuggets all day long, but that doesn't mean I'm eating a roast chicken.
> I can have an LLM generate me code-like text all day long, but that doesn't means it's building a system.
I don't follow. An LLM doesn't magically output code-like text, or anything else for that matter, on a whim. You have to build a system that describes your intent to the machine. Only then might it output code-like text, if that's what your system describes. It's not the execution of your code that makes you an engineer. It's building a system that can be executed in the first place that makes you a (software) engineer.
> There are many things you can do with LLMs other than build systems.
Like what? The code-like output example clearly requires you to build a system before the LLM can evaluate your program. If I want an LLM to generate a bedtime story, I also need to build a system that defines that. Where do you find the escape?
Maybe you're thinking of AGI? While everyone has their own pet AGI definition, many see it as being the point where you no longer have to build the system and the machine can start to take on that role. But we don't have that yet, and isn't what we're talking about anyway.
In mechanical engineering you can be an engineer or a mechanic, or other things (like a hobbies or different degrees of casual work), and they're no interchangeable.
I think being an engineer implies it's a profession you've trained in and you're implementing the science behind it in a practical manor in some capacity (like a computer scientist studies computers, a computer engineer implements and builds the systems based upon this science).
> In mechanical engineering you can be an engineer or a mechanic, or other things
Sure. While some mechanics may sometimes find themselves having to design things to carry out their work (in which case they would be engineers during that time), generally speaking mechanics carry out the physical replacement of what engineers have already designed. They are not designing things themselves.
But there is no mechanic analog in computing. At least there isn't a human mechanic analog. One is always operating at the design level, regardless of whether the design is written in C, Rust, or natural language. All the engineering-adjacent work you find in other engineering disciplines is done by the computer in the software realm.
> I think being an engineer implies it's a profession you've trained in and you're implementing the science behind it in a practical manor in some capacity
That's what "professional engineer" implies, but we're talking about "engineer". There is no such connotations in the word engineer alone, hence the existence of the PE term.
I would say attaching things together with or from an LLM would be the analogy. Being a mechanic and being an engineer are both technical jobs which require skill, but one is design focused while the other is system focused.
I would also disagree that "professional engineer" is a meaningful distinction from engineer. I don't really think of an engineer as being something you can do casually, it's a profession.
Also re-reading my comment I should have said adjacent to mechanical engineering, generally you don't get a degree in mechanical engineering to become a mechanic.
> Being a mechanic and being an engineer are both technical jobs which require skill, but one is design focused while the other is system focused.
Right. And such software-based work is always design-focused. You're writing documents that specify how the computer is expected to function. The computer takes those specifications and turns it into something functional. I think you can make a good case the latter is system work, but not something humans participate in.
In the olden days we had to write those specifications in arcane languages that were a speciality to learn. Now you can write the specifications in plain English, more or less. That's evolutionary, but since the advent of the compiler we've always been moving towards the language being more and more like English (or insert other natural language, if you prefer). Such progress is nothing new.
Perhaps it's like the difference between a manual and automatic transmission. Gearheads worried about their personhood might try to argue otherwise, but in reality you aren't not a driver if your car has an automatic transmission. It is not the exact details of the tool that defines the act you are engaging in. Likewise, you don't suddenly stop being an engineer just because the tools have now made it "easy".
> I don't really think of an engineer as being something you can do casually, it's a profession.
You're always free to come up with your own pet definition, but per the record of how the word is most commonly used out there in the world, there is no connotations to one's skill, ability, quality, or anything of that nature that is normally associated with professional engineer. It only refers to the act itself in any capacity. Which, of course, is why we also coined the term "professional engineer". That would have be entirely pointless — and therefore would have never been adopted — if "engineer" already captured that sentiment under common use.
No. You do not. It may make you a developer, at best. I don't even call my self a software engineer, because I'm not. I'm a self taught coder that has spent 25+ years gaining experience, but I've never graduated from a school with any kind on engineering degree. I started CSE way back in the 90s, but stopped because life got in the way.
Maybe you're joking, but you just know people actually feel this way. They have no idea the difference of a coder and an SWE, and flippant comments don't help
> but I've never graduated from a school with any kind on engineering degree.
So? Per the dictionary, engineer is clearly defined as: A person who designs, builds, or maintains machines, structures, or systems. There is no mention of school or having an engineering degree.
It has always been a bit debatable if software fits into machine, structure, or system, granted, but we generally have come to agree that it does. And per the context of discussion, we've already established that it does for the sake of discussion. On that understanding, designing/building/maintaining a system in "LLM code" instead of C++ code is fundamentally no different.
You're likely confusing engineer with Professional Engineer™, but that's something else entirely. That obviously has nothing to do with anything that we're talking about here.
VR flight simulator software is accessible to the layman. Does that make them qualified to be a captain (pilot-in-command) for a commercial passenger plane?
> VR flight simulator software is accessible to the layman. Does that make them qualified to be a captain (pilot-in-command) for a commercial passenger plane?
No. They might be able to fly a plane poorly, though. Engineer doesn't imply being qualified, only engaging in the act of designing, building, or maintaining a machine, structure, or system. You don't have to be qualified, or even be good at it, to carry out those acts.
You're probably thinking of Professional Engineer™, which does represent recognized qualifications, but that's something completely different. Obviously if Professional Engineer™ was meant, Professional Engineer™ would have been written.
Huh, Pilot/ Captain job requirement for commercial plane is highly regulated by authorities like FAA etc but software engineer has no such requirement. Any random business with some basic software requirement can ask an employee or contractor to get something developed quickly and deployed it. They may start calling that person software engineer.
Further even if you have some strict ACM/IEEE definition of Software Engineer®, a person is not going to end up in jail if they don't fulfill those but call themselves software engineer nonetheless.
> Huh, Pilot/ Captain job requirement for commercial plane is highly regulated by authorities like FAA etc but software engineer has no such requirement. ny random business with some basic software requirement can ask an employee or contractor to get something developed quickly and deployed it. They may start calling that person software engineer.
Exactly the problem. Secondly, if I am building commercial plane software for pilots to use, you wouldn't want to hire unqualified / in-experienced 'engineers' for all the critical work and validation testing. (or even AI vibe-coders picked from anywhere.)
Because surely, that worked out for Boeing. [0] /s
> Further even if you have some strict ACM/IEEE definition of Software Engineer®, a person is not going to end up in jail if they don't fulfill those but call themselves software engineer nonetheless.
So we are now defending fraud if one calls themselves an SWE on their CV with zero experience other than an AI doing all the coding?
It's like you want to take the legal risk hoping that the employer / company won't sue you for fraud when that vibe-coded software goes all wrong and money is lost.
At the end of this day.. and of the next.
But at some point, the tool will "suddenly" turn into a versatile agent, and that time might be a lot sooner than most expect (c.f. "exponential growth surprise factor"...)
I've been getting lots of value out of AI coding tools; especially in the last few months. My assessment is that this will lead to more work to do, not less. It's not a zero sum game. More of what I do is baby sitting AIs. But together we do more than me with my team before. It will reflect in my hiring decisions as well. I want people that can do this responsibly (are able to tell good from bad code, are able to think for themselves, are able to get stuff done).
My observation so far is that micromanaging AIs still sucks
to play devils advocate when a company is faced with increasing pnl ( reducing headcount) or trying new ideas why won't they take the former since it is guaranteed
There's a ceiling on how much you can increase profit just by substituting existing human roles with compute. At some point you need to expand the workforce to raise profits further. And businesses are definitely willing to take risks for the potential of massively scaling up profits. The "risks" here being a larger workforce combined with more compute.
> First, let’s establish our benchmark: job postings dropped 8% in 2025 compared to the same period in 2024. Indeed reported a 7.3% year-over-year decline for US jobs recently, so this was a good sanity check, and told me that my data most likely was comprehensive.
Doesn't this also suggest that the job market is in such an unusual state, that any further analysis makes little or no sense?
Yeah, I was wondering something similar, especially because there was a huge surge in tech job hiring during the pandemic, and part of the story from 2020-2025 is the industry regressing back to the trendline before then. Some people say the rise of ChatGPT (in what, 2023?) is part of that but I'm not convinced.
I suspect much of the data is influenced by speculative firing/rehiring as c-suites are exploring what they can actually skimp on and what they cannot.
There is no causality analysis here, and really no justification for why AI is the reason for these job losses. An alternative explanation is the tariffs.
Same is true for sustainability related jobs. Given the backlash first from Texas, Florida, and other red states and now from the Trump administration, it is not at all surprising that these jobs are in decline. AI has nothing to do with it.
My experience using Claude code is that it's excellent at React/React Native and because apps are mostly a View layer the consequences of bad "vibes" are less far reaching.
It depends on a lot of things, a lot of mobile development has shifted overseas, and this report only focuses on US jobs. I can't tell if this is from domestic AI, or outsourcing that uses AI.
The decline in frontend engineering jobs matches my personal impression from the past year. Smaller companies can do a lot with vibe coding alone, while larger companies can multiply their FE team productivity without making any new hires.
Frontend code can be very repetitive / labour intensive, I bet that has made this more attainable than for other layers in the stack. Most mistakes in UI code are also easily corrected and have a very tiny impact radius.
Yeah, I've found LLMs to be pretty decent at frontend stuff.
I got a new job recently and my main task has been overhauling a Claude generated app into something functional. The backend is an unsalvageable disaster, but the frontend is actually pretty good. Which is great for me because I suck at UIs.
I think if you have a solid backend already in place, an LLM can produce a pretty functional frontend with far less effort to fix up.
I think that has more to do with the fact that JS is the most popular language in the world, especially available code to train LLMs on.
Backend code is far less likely to be available because it's more likely to be closed source and is spread over many languages (Java, Python, C#, JS, Ruby, Go, PHP...).
> While I acknowledge not all job postings result in a hire, and some are ‘ghost jobs’, since I was comparing the relative growth in job titles, this didn’t seem like a big issue to me.
Uhhh... That's a very big issue. In addition, there is also a glaringly obvious analysis error here related to layoffs/attrition.
No offence but is this analysis vibe analyzed or something?
All this is actually measuring is how the number of job listings in a specific industry have changed globally. That's not the same thing as "what jobs AI is replacing" at all!
Like nurses decreasing 11%. There is 0% chance that AI is disrupting nurses. If anything, nursing is probably in more demand due to continual aging. It might just be that fewer people are quitting because of covid being over. The total number of nurses might be going up still!
Are we just upvoting anything with AI in the title now? HN used to be pretty respectful of scientific methods. The methodology section here just reads like personal resume filler for showing that you can use AI. I mean I get it and I admire the hustle, but the quality here is pretty lacking overall. https://bloomberry.com/blog/i-analyzed-180m-jobs-to-see-what...
I perform a different analysis, focused more on the type of blue collar jobs that most people believe are the prime candidates for AI/robotic job replacement:
You switch from global automation to US car manufacturing in the first paragraph without a even a line break, basically equating them.
The first graph you show is missing the corresponding table from the source which shows the exact opposite of what you have stated: the workforce had been steadily increasing until ~2019 and is in decline since then. Not massively but still it contradicts what you stated. Next you switch to Korea and the wage increase without aligning a corresponding price increase or adjust for inflation. Then you switch to the history on the industrial revolution. At that point I gave up because I know the point you are trying to make and I can see, you adjusted your samples to match that point.
Would be cool to see cybersecurity covered here as well. I find it interesting how the machine learning engineer is exploding, but every engineer I talk to about it suggests these positions are mostly doing non-productive "bullshit" work and not really contributing much (arguably this applies to many tech jobs). Obviously that's anecdotal and probably not the real story, but it does seem like "machine learning engineer" is a broad term and probably doesn't tell the full story in itself.
I wonder what the numbers look like if you compare pre-pandemic to now.
To me it feels like during the pandemic companies found out they can fire a lot of people without a lot of backlash as long as they have a weak excuse.
I feel like AI taking over jobs is such an excuse, especially considering creative jobs and the layers that are supposedly being replaced by AI in the article they call it "Creative roles that “execute” ".
The US is in a recession when the AI bubble is taken out. So obviously jobs are in decline. Machine learning hires are up because there are billions of dollars in circular financing.
While consumers are generally receptive to assistive machine learning like speech recognition, translation, limited auto-complete I think there is a huge backlash towards it being used to generate content and as far as I know all the major AI companies are losing huge amounts of money with every transaction. I am sure AI is being widely used as a reason for staff reductions but the reality is more likely sales are down, costs are up and head count is too high for current business requirements.
AI is taking jobs but for a different reason. Big Tech is in an AI race so they will shrink their staff or move jobs to India as much as possible to free capital for GPU datacenters. To hell with the consequences.
Meta, Alphabet, and Oracle are even issuing almost $60 billions in bonds. Unheard of.
This is part of the picture, but in terms of impact one also needs to look at this in terms of absolute numbers not just percentages. Two job positions can both drop by 10% and in one case it represents 100,000 jobs and in the other case 500,000 jobs.
Hey, I liked the analysis thanks! Any reason why you removed SDRs, Account Executives and similar Sales related roles? I would love to see how that compares to the data you do present.
They probably have this data. But the data is most definitely heavily polluted: Fake job listings, jobs posted to several sites, sites that aggregate job listings etc. This data is probably more polluted by AI then AI has influenced job openings itself.
Flipping between GPTs and Gemini. Mostly having it help with interpretation of instructions and giving direction to find that data in my books. Stuff my CPA did. Help me categorize an expense (easy). And I had it make a web-form for all the data entry to produce the necessary PDFs.
Completely absent is quantity of jobs. If e.g. ml engineer job postings go +40% from 200 to 280 and writer job postings go -50% (over two years) from 20000 to 10000, then we have a better idea of the impact.
Without those data this report isn’t really quantifying impact on “180M jobs.”
Also decline in Security engineer by 0.35% doesn't make sense by conventional wisdom. Shouldn't it be increasing due to increased demand for security in all ai integrations?
I wouldn't be surprised by a drop in security postings. Quite a few companies view security as an "overhead" so the siren call of reducing that overhead by introducing AI is a thing.
Also for a lot of jobs in security it's pretty hard to measure how well it's being done, so if the AI based solutions are worse, that might not show up for a while
We also need to consider the confounding effect of corporate performance and recession expectations.
Cost centers in businesses are early canaries of expected pain, and a reduction in security roles may reflect belt-tightening irrespective of AI impact.
> for a lot of jobs in security it's pretty hard to measure how well it's being done
Nothing going wrong: “What are we paying you for?”
Everything going wrong: “What are we paying you for?”
It’s a no-win situation unless you manage to score a division manager who understands security and understands the reports a good security division produces. And most importantly, understands that no news is good news.
Security products and practitioners are the classic snake oil salesmen. They are actually sales and marketing roles for help closing deals by emphasizing some security aspect. True security comes from general IT practices followed by engineers themselves.
> True security comes from general IT practices followed by engineers themselves.
Thank goodness engineers pop up out of the ground fully trained on good general IT practices....
I would be wary of making categorical claims like this, but it's unfortunately true that "security" field hasn't been doing well in a long, long time now.
Half the field is B2B "magic bullet" solutions like CrowdStrike and all the associated sales tactics - with pitches that boil down to "you give us money, we make your security issues go away". Half of what remains is mandatory certifications and other flavors of checklist-obsessed cargo cultists - often CYA-driven, often demanding the adoption of the fancy acronym of the day, regardless of the real threat profiles. Then you get the "security snake oil" - "magic bullet" systems that don't work, never did and never will, but are supported by the right influence groups and get the right pockets lined, and so are used anyway. DRM systems like WideVine and PlayReady being the prime examples. Then there are the corporate "security of our business model" shills - who pay lip service to "security", but have the true aims of "prevent anyone we don't like from doing anything that can harm our revenue streams" - with Apple being a common example.
And about a fifth of the field is people who do actual security work, and keep the sky from falling.
I agree with you totally, although I'd venture to guess 20% is way too high. I'd say you have about 10% people doing security work, 15% doing compliance, and the rest are consuming oxygen.
It's a growth field, so you have lots of idiots getting certifications and stupid jobs. Reminds me of the 90s when I started, and companies were paying MCSE's (ie read a book, hit next-next-finish in Windows NT) more than software engineers in some markets.
As the security guy. I get the feeling that on average engineers are not exactly great at general IT practises. Or even doing basic things.
> True security comes from general IT practices followed by engineers themselves
Sounds exactly like something the average security practitioner would say...
`not_sure_if.jpg`
How does this affect hiring of security engineers?
> True security comes from general IT practices followed by engineers themselves.
I have yet to meet an org whose engineers care about security, or who would not compromise security if secure practices got in the way of shipping a product or feature.
I'm a bit amazed you consistently get downvoted while you seem to speak the truth. So much gray in your comments.
I consistently see this commenter making a single comment, of questionable relevance, expressing a strong opinion which isn't particularly thoughtful or interesting or true. Then they ignore the pushback and move on to the next thread, where they post another tangential hot take. I'm not at all surprised at the result. Those comments attract a lot of downvote because they aren't very good.
This thread is a microcosm of that. They went on a tangent from a tangent to express how little they think of their colleagues working in security. It wasn't out of curiosity, it didn't raise interesting questions or provoke interesting debate. They didn't defend or substantiate their opinion so that they and we could learn something from it. It was just a drive-by flamebait to stir the pot and express derision. It should be downvoted; it's a bad comment.
Perhaps that pattern is difficult to see when their hot takes align with your own takes.
A microcosm indeed.
I didn't write my comment to applaud them.
I don't understand what "you seem to speak the truth" means if it isn't an endorsement?
It's an observation.
On the continuum of approval, where at one end there is endorsement and at the other disapproval, it's somewhere in-between. Even I who made the observation don't know exactly where. Sometimes something jumps out at me and I don't yet why.
It could be an incorrect observation. Some of what they said seems true, some false. I don't know enough about security specifically to say. I know a lot about other things to know he said some things that are true.
It's astonishment at perhaps some kind of law of the universe that things that seem one way may be a different way.
It's an exclamation at the poetic irony of someone expressing there's gray area in some things gets downvoted and their comments are in gray colour.
It's a way to introduce myself, to say hi Mr Monero user, and pass a super secret note.
I meant no disrespect.
Perhaps my reply here is astonishment at how interpretation of words may depend on imagination. As if words alone aren't enough.
I post my view that is against the HN hive mind and don't always feel like rebutting the same hive mind talking points again and again. I like to post to prove there is an alternative view out there
I'm also guilty of what they accuse you of. Sometimes my internet comments are not made for the purpose of sparking discussion, but more of a "vent" where I know my take is not popular but I feel the need to throw it out there anyway. The comment is more for "me" than anyone else. And, yeah.. that makes it a bad comment lol.
I also just love playing devil's advocate, and I'm adverse to hivemindy-feeling opinions (even when I share them). Maybe this all describes you, too.
I don't have a problem with people doing that as long as they don't pretend that every other commenter holds the same contrary opinion and that the downvotes indicate they're too sensitive to discuss such things, or other similar rationalizations. If you want to leave some drive-by snark without rationalizing it as being about other people, it's not my favorite kind of comment but I'm not going to object to it either.
The downvoting functionality here and in other forums can mean many things. It isn't a precise thing. If it is precise I'm not finding a clear definition. It can mean I disagree, this is boring, this is false, this made me sad, I don't like reading this, I don't like this user, I'm tired, etc.
One plausible interpretation of a downvote without a comment is drive-by snark without rationalizing.
I don't know your motivations but I know the "HN hive mind" isn't the problem. When you do engage with people who disagree with you, it usually becomes evident to me that there isn't much substance behind your views and that you struggle to disagree amicably. I also see lots of people on HN with a similar perspective to yours who don't have the same problems or engage in the same patterns of behavior.
The facts are that HN has a diverse set of perspectives with many conservative/libertarian commenters who would align with you, but that your comments are frequently shallow flamebait. Though I have seen a couple good points you've made, as well. Do with that information what you will.
I disagree entirely, I don’t even post very frequently so it’s surprising I have someone tracking my posts. The shorter a comment the better it is, if the same opinion that takes an essay can be distilled into a sentence
Maybe you disagree, maybe you don't. Since you chose to veer in different direction and reply to something I never said, I don't know either way.
In any case, brevity is something great writing and shallow hot takes share.
My first comment on this whole thread was how security in tech is theater, and the sellers mostly snake oil salesmen. I’m not the first to make this observation and I don’t think it’s wrong. Which is why employment in the sector is down, full circle to the OP
Go ahead, take that slim, speculative, tangential connection and interpret it as permission to inject your hot take into the discussion. Decline to elaborate when your supposition is challenged. You've every right to do that.
Just don't pretend that it's for our benefit or that we downvote it because we're unthinking drones, or that you decline to elaborate because we're simply not capable of having the discussion.
I tell you this because if I were insulating myself inside a bubble and rationalizing my interactions with those who disagree with me as being the reflexive behavior of a hive mind, I would hope someone would point that out to me. So here it is; again, do with the opportunity what you will.
> Just don't pretend that it's for our benefit or that we downvote it because we're unthinking drones
The reason many comments are downvoted on HN in general is most often unknown to me. One interpretation is that it's a major flaw in HN.
This design decision by HN could be intentional, as a trade-off to achieve something else. For example, it could be done to have high velocity of discussion. High velocity could preserve an invariant of keeping or pulling users on the site.
If it's a trade-off, which would suggest something is given up for it, it might be worth exploring what's given up.
This website is full of unthinking drones acting with hive mind behavior, that is my contention, and I think very differently, not just on here but with almost everyone I engage with. However I succeed over and over with asymmetric bets on a wide variety of things, including and especially tech and making money in tech, so if we compare bank accounts and career trajectories, investments, etc. it would be wise to let me speak
Overweighing people's opinions on matters they demonstrate a shallow engagement with on account of their success in other areas is cargo culting. Maybe you should worry less about HN and more about your own reflexes to accept bad ideas from yourself and others based on their wealth. Maybe succeeding in contrarian bets doesn't make you "correct" in some moral sense, but only successful in the trade.
It's easy for me to believe you're an intelligent person who's accomplished impressive things. But it wouldn't contradict anything I've said.
I own a software business with hundreds of employees I built from nothing. I know all about tech security, hiring security guys, etc. It is a cost center, the directors / VPs / CSOs are overpaid salesmen, 99% of the products you want are features from cloud vendors or provided by standard tools like device management or password managers. I totally get the patina desired by large corporations who need to show wall street they care about security. When I box check I use a super cheap cut rate firm who can check boxes for me at the lowest cost because it is total bullshit. I'm sorry this is the reality
It's understandable to have sour grapes after having some bad experiences, but what I'm hearing is unrelated to any impact AI is having on the job market. You're talking about security products you think are snake oil and executives you think are overpaid; that is unrelated to trends in job postings for security professionals, working at software companies, and how that might be impacted from AI.
This is what I've been saying. You've got some random grievance, you want to take this discussion as an opportunity to get it off your chest. But you don't want to engage with people challenging your ideas. And when for whatever reason you do explain yourself to me, your explanation is "I am wealthy and successful, so I must be right. Those who disagree with me are an undifferentiated mass of imbeciles that I have nothing to learn from."
If that's how you want to live, it's your right. You're only cheating yourself, so maybe I'll just shut up and let you get on with it.
> This website is full of unthinking drones acting with hive mind behavior
Maybe not "full", statistically, but many times I receive a similar impression.
Thanks! I just take the downvotes, whatever
People are sleeping on AI in sec, lots of lazy sec engs and architects going to be SoL sooner rather than later.
I could easily see those just running tool and then printing report being replaced by script running the tool and passing results to LLM and then sending report.
And probably more useless architects.
Most companies don’t care about security beyond window dressing and getting whatever certification required to close deals.
Time for budget cuts? Cut the Security team!
Since most of what these teams do is box-checking for these certifications, it's true.
Security is a process. It's not a constellation of products or certifications.
From what I can see, being closer than the average engineer to the space (but not an expert on my own), a few things are happening:
* Engineers are being pushed for ownership of security more directly. You still need someone on the team to guide and support them, but they're not going to be directly involved all of the time.
* Significant amounts of automation and centralized security. Supply chain management is a double edge sword. It does open up vulnerabilities, but you can simply pay one of the SaaS companies in the space to help with a lot of the heavy lifting.
* Commoditization/Platform-ification drastically reduces attack vectors.
OWASP has a nice comparison from over the years: https://github.com/OWASP/Top10/blob/master/2021-2003_Compari...
The "S" in "AI" famously stands for "security", so no humans needed anymore.
if everything goes down 8%, the one that loses only 0,35% is a relative winner.
Bold of you to assume there is any demand for security in AI integrations. It's like 90s web browsers, everyone's running random MCP servers that do god knows what.
Seems like when the security market is low on the white hat side, it's high on the black hat one. Security people just need to learn to adapt.
I'm sure some of these ransomware groups probably offer health insurance and 401k matching.
I think it's because companies are moving away from in-house security and hiring 3rd party companies for security work. It also depends on the time of year this was taken. Q4 is the busiest time for security. Q1 is the slowest.
I'm a security consultant and work with multiple companies that provide security services. Work has increased massively in the last year.
Why do you think there is increased demand for security in AI integrations?
Also absent is any data in SDRs, Sales Account Executives and Sales Managers (all the Revenue vertical.
Comparing US job openings from 2024 to 2025 seems to be selective. Is the job in sudden decline? Or a reset to a norm? Or is the decline something that has been happening over even a longer timeframe?
Taking two years and drawing conclusions from those two years seems to miss larger pictures - especially when the mess of Covid and the pandemic years and that job market is mixed in with this. Yea, that's half a decade back but companies are still trying to figure out how much staff they need where.
Photography? Yes, that's likely been impacted. Compare https://web.archive.org/web/20230124085038/https://www.bls.g... and https://www.bls.gov/ooh/media-and-communication/photographer... and the rate of growth has changed. However I'll also draw attention to the estimates that there's only a few thousand open positions with that classification each year. This includes self employed stock photographers and artists - corporate photographers have been a hard thing to get for a much longer timeframe (I looked into it a little bit back in '09). Additionally, artistic photography is impacted by the amount of money that regular people are willing to spend at art festivals and the like - that's gone down irrespective of AI.
A few more steps (<< post-covid << covid << zero interest-rate << ...) to understand that there is no reliable baseline for this use case.
This subset is not perfect, but good enough.
I agree with this. I'm a product designer, and apparently there's "only" been a decrease in job postings of 2.6%. That doesn't seem bad, however in late 2022 through all of 2023 people were getting canned left and right and the market was incredibly tough. Can't really have big decreases when you're near the bottom of the hole.
I found this interesting: "Still, despite all the hype about how AI coding tools will replace software engineers, software engineering is still one of the most secure jobs you can have today, relative to most other white-collar jobs."
There is a lot of induced demand in software engineering. We are still in the realm where cheaper software means that people want more and more complex software. And that demand increase is more than enough to offset any efficiency increases
Meanwhile the amount of accounting that has to be done is pretty inelastic. Whenever accounting gets more efficient you just reduce the number of accountants instead of doing more accounting
Creative is somewhere in between. Not completely static demand, but not extremely elastic either. The healthy rise in postings for creative directors indicates that the cost reduction has lead to more art being done, but the increase in demand isn't big enough to offset the job losses in the rank and file positions
It's funny, almost every conversation I have about the fact that I work as a software engineer with new people I meet nowadays seems to include them asking if I'm worried about AI stealing my job. Maybe it's something that people ask everyone nowadays regardless of what industry they work in, but at least as far as I can tell, the type of work I do doesn't seem in any apparent danger of being replaced by AI any time soon.
I assume that they think SWEs have a better grasp on it than non-tech folk.
as of today that's probably true, but if the labs manage to keep increasing the 50% time horizon (defined by METR as 'the length of tasks (measured by how long they take human professionals) that it can complete autonomously with 50% probability') at the current pace, it might not be for long. Exponentials are hard enough to forecast if you kinda sorta know the parameters, and we don't have that comfort today.
sorry, new to this, any helpful links on what metr is?
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-com...
It's because journalists are still big mad that the internet wrecked the newspaper business, therefore the news constantly reports lies about how the tech industry is collapsing. The more news you watch and the less personal contact you have with the industry, the more likely you are to believe that techies are jumping out of office windows in despair (hi, Mom).
The job market is terrible, pay is stagnate and remote work is being taken away and the biggest companies despite profits are laying off large amounts of people. I don't think that's fud
> the biggest companies despite profits are laying off large amounts of people.
Hi is the need for employees related to profit?
The decision to layoff has side effects beyond net income on a balance sheet.
What's your point? If there's no work for 10k people to do, should companies just continue employing them to do... nothing?
One of SV's strengths is cutting your highly qualified and experienced staff loose to go do something else.
Is there something the the SV charter that says companies operating there must do that?
Nobody seems to want to answer the question so I assume nobody wants to confront that a company isn't the bad guy when it fires people it doesn't need. And, if it turns out it did need them, the company will suffer. Usually that doesn't happen though and so I tend to trust that these companies understand their business needs more than I do.
It sucks when people lose their jobs but if there's no job at their company for them anymore, their employer isn't a charity. Just like if a company could no longer afford to pay they employees, the employees shouldn't hang around and work for nothing. You might as well be demonizing people who file for divorce for making their partner single.
Agreed. I intended that last comment to reinforce your point.
That's never how layoffs are done. If you survive them your workload just increases and the employees picked are based on nothing or something arbitrary.
Then your company has a reputation for being unreliable for stable employment and people that can pick where they want to go will go else where. It also kills morale and remaining high performers will go somewhere else. Now the company laying off is stuck with low performers that don't have other options.
Sounds like a recipe for bankruptcy. How many of these companies doing mass layoffs are going to go out of business?
I don't see how you can hold that this job market is "terrible" in a frame of reference of the last 50 years, unless you were born fully-grown in a vat exactly 4 years ago.
Have you looked for a software job in the last few years?
It's worse then 10 years ago. Idk why everyone who comments on the job market only looks back to the covid year
No, and no individual person can rely on their personal experience to gauge the job market, including you. However if you wish to assert that it is currently more difficult to get a job as a software developer than it was in late 2015, aggregate data suggest this is false. Also, looking back a mere 10 years does not dispel the general impression that you lack context.
If you look at employment in, for example, NAICS 518 "Computing infrastructure providers, data processing, web hosting, and related services" which is one of the larger BLS categories for our industry, the numbers are at all-time highs, having doubled since 2011. An example of a bad job market for software developers was 2001-2011, when this sector shrank by a third.
Which numbers are at all time highs.
Is that really representative of what people mean by tech? Or does that include companies that happen to make software?
This makes sense to me.
Until Software Engineers have automated away all the other jobs with AI & software they'll be safe. That's going to take a long time.
Replacing software engineers with AI only affects the bottom line of software companies. Companies are usually fine with increasing the bottom line if they can exponentially increase the top line. I think software engineers will provide that capability for at least the next 10-15 years.
I think that's an over-simplistic view - at the moment there are many, many software engineers hired by companies who are betting on AI being madly profitable. If those expectations change, we could see more cascading layoffs, which will mean those engineers will go looking at more traditional places like banks, which means they'll stop hiring, which means it'll be harder to someone who is looking for a new job to find one, even though not all jobs have yet been automated.
True, besides actual software engineering is small part of overall IT/computer related work. There are far more analysts, managers(project, program, IT, agile etc), QA, operations and so on. So engineers who employers think are capable of leveraging AI and do 20 s/w engineer worth of work with 5 people will remain in demand for time being.
Even without outright layoffs one can see how fast leverage of average IT engineer is disappearing. After 20 years of experience my value or feedback matters less than when I was 4 years into paid job. And it has far less to with AI so far.
Most custom work of past is just a library, component or framework to use. And those are mandated to be used as it much easier to hire/replace teams to work on those.
Now It may be always be true to have reusable components created but growth of IT industry kept people employed in ever greater numbers. However now it seems to be reaching limit. Leaving aside highly visible layoffs by US tech giants, growth is fading in countries like India with huge IT offshoring workforce. There are millions upon millions jobless fresh graduates waiting to get jobs with some IT degree.
"There are millions upon millions jobless fresh graduates waiting to get jobs with some IT degree".
Quality of those graduates is not always great. A talented and passionate SWE will always make a good living.
what about amount of churn in software engineering?
Did the invention of compilers eliminate the need for programmers or make them more productive and valuable? LLM coding is really not in the most abstract sense any different from compiling a higher level language to a lower level language.
> LLM coding is really not in the most abstract sense any different from compiling a higher level language to a lower level language.
Hard disagree here. Anecdotally, know a few people who can't write a Java program that will compile, who can leverage ChatGPT to produce functional websites.
A good friend of mine ChatGPTed his way into a masters degree that involved a lot of coding. A good 97% of his degree was done by AI, and the other 3% was me helping him troubleshoot he couldn't get AI to solve.
LLM is vastly different from a compiler/translator. Despite the joke, you can't just fire up Python with import website and have a functional website. But you can basically do that with LLMs, which will then add features as requested. It's not perfect, nor guaranteed to be functional, but it is quite a bit more capable than a compiler is for such tasks.
At my work, the sales guys are using AI tools to rapidly prototype features on our website with prospects. While it doesn't do all the work, it can produce useful HTML templates that the front-end team can make functional.
Certainly the jobs that were around for bespoke compilers were eliminated with the unification of compilers (GCC, Microsoft Visual C++). How many of those people transitioned to other roles I don't know. But the number of compiler jobs has been declining forever at this point.
At the end of the day, "AI" is just another programming language, albeit one that is much more accessible to the layman. When using AI, you become a software engineer. So it stands to reason that software engineering jobs are strong.
But what about pay? Elevator operator jobs have never been more prevalent, but increased accessibility to the layman pushed the price to zero.
> When using AI, you become a software engineer.
When using a pen you become a poet ? lol
Most people who code aren't software engineers, you certainly can't extend the definition to every AI users
> When using a pen you become a poet ?
No. By definition, a poet writes poems. Not all pen use leads to poems.
By definition, engineers build systems. What else can you do with code (and LLMs; same thing) other than build systems?
idk man, plenty of people have "ai" gf/bf/therapist, ask "ai" for vacation trip ideas, recipes, gym workouts, &c. I wouldn't even be surprised if most tokens were used on non software engineering tasks.
I have a zombie developer, coder, idk how to call them, in my team who doesn't talk to anyone, writes shit tier PRs and spends all day long talking to chatgpt. They're a prompter, a chatter, a waste of money, but certainly not an engineer
> idk man, plenty of people have "ai" gf/bf/therapist, ask "ai" for vacation trip ideas, recipes, gym workouts, &c.
Not without engineering a system to get the AI to generate a response they don't.
> non software engineering tasks.
There is no non-software engineering task when using an LLM. All LLM use first requires designing software to get the LLM to do anything. An LLM will not autonomously act on its own. It can only respond in accordance to the the program you have designed for it.
That program can be written in natural language, granted, but software programs are not limited by the particular language they are written in. Its not somehow not a program just because you didn't write machine code on punch cards.
You can make a lot of slop of all different sorts with LLMs. That has very little in common with building systems.
"Slop" suggests that the only difference is in quality, but the definition of engineer says nothing of quality.
Perhaps you might consider using an LLM to build a system that creates a response that is more coherent?
> the definition of engineer says nothing of quality
The definition you used makes someone who maintain a twitter account or use a coffee machine an "engineer"... everyone is an engineer by that definition really
It's not just the definition I use, it is the most common definition people use according to the records we keep.
And yes, someone who builds a system to generate Twitter content or maintains a coffee machine are definitely engineers.
Professional engineering societies don't like that fact and often try to usurp the term for their own financial gain, but that's not how English works. Definitions derive from use, not what a small group of people wish were so, and the record that keeps track of use is abundantly clear how the word is most commonly used.
Slop doesn't suggest the difference is only in quality, but also in form. Words have meanings, ya know?
I can have an LLM generate me code-like text all day long, but that doesn't means it's building a system. Just like I can eat Chicken McNuggets all day long, but that doesn't mean I'm eating a roast chicken.
> I can have an LLM generate me code-like text all day long, but that doesn't means it's building a system.
I don't follow. An LLM doesn't magically output code-like text, or anything else for that matter, on a whim. You have to build a system that describes your intent to the machine. Only then might it output code-like text, if that's what your system describes. It's not the execution of your code that makes you an engineer. It's building a system that can be executed in the first place that makes you a (software) engineer.
You said a few comments upthread (and I quote): "What else can you do with code (and LLMs; same thing) other than build systems?"
There are many things you can do with LLMs other than build systems. That's my point. Using an LLM doesn't make you an engineer; that's preposterous.
> There are many things you can do with LLMs other than build systems.
Like what? The code-like output example clearly requires you to build a system before the LLM can evaluate your program. If I want an LLM to generate a bedtime story, I also need to build a system that defines that. Where do you find the escape?
Maybe you're thinking of AGI? While everyone has their own pet AGI definition, many see it as being the point where you no longer have to build the system and the machine can start to take on that role. But we don't have that yet, and isn't what we're talking about anyway.
wow epic comeback man can you share the prompt you used to generate it?
I'm told you don't need to build a prompt. That would be engineering, which apparently isn't involved in the use of LLMs.
In mechanical engineering you can be an engineer or a mechanic, or other things (like a hobbies or different degrees of casual work), and they're no interchangeable.
I think being an engineer implies it's a profession you've trained in and you're implementing the science behind it in a practical manor in some capacity (like a computer scientist studies computers, a computer engineer implements and builds the systems based upon this science).
> In mechanical engineering you can be an engineer or a mechanic, or other things
Sure. While some mechanics may sometimes find themselves having to design things to carry out their work (in which case they would be engineers during that time), generally speaking mechanics carry out the physical replacement of what engineers have already designed. They are not designing things themselves.
But there is no mechanic analog in computing. At least there isn't a human mechanic analog. One is always operating at the design level, regardless of whether the design is written in C, Rust, or natural language. All the engineering-adjacent work you find in other engineering disciplines is done by the computer in the software realm.
> I think being an engineer implies it's a profession you've trained in and you're implementing the science behind it in a practical manor in some capacity
That's what "professional engineer" implies, but we're talking about "engineer". There is no such connotations in the word engineer alone, hence the existence of the PE term.
I would say attaching things together with or from an LLM would be the analogy. Being a mechanic and being an engineer are both technical jobs which require skill, but one is design focused while the other is system focused.
I would also disagree that "professional engineer" is a meaningful distinction from engineer. I don't really think of an engineer as being something you can do casually, it's a profession.
Also re-reading my comment I should have said adjacent to mechanical engineering, generally you don't get a degree in mechanical engineering to become a mechanic.
> Being a mechanic and being an engineer are both technical jobs which require skill, but one is design focused while the other is system focused.
Right. And such software-based work is always design-focused. You're writing documents that specify how the computer is expected to function. The computer takes those specifications and turns it into something functional. I think you can make a good case the latter is system work, but not something humans participate in.
In the olden days we had to write those specifications in arcane languages that were a speciality to learn. Now you can write the specifications in plain English, more or less. That's evolutionary, but since the advent of the compiler we've always been moving towards the language being more and more like English (or insert other natural language, if you prefer). Such progress is nothing new.
Perhaps it's like the difference between a manual and automatic transmission. Gearheads worried about their personhood might try to argue otherwise, but in reality you aren't not a driver if your car has an automatic transmission. It is not the exact details of the tool that defines the act you are engaging in. Likewise, you don't suddenly stop being an engineer just because the tools have now made it "easy".
> I don't really think of an engineer as being something you can do casually, it's a profession.
You're always free to come up with your own pet definition, but per the record of how the word is most commonly used out there in the world, there is no connotations to one's skill, ability, quality, or anything of that nature that is normally associated with professional engineer. It only refers to the act itself in any capacity. Which, of course, is why we also coined the term "professional engineer". That would have be entirely pointless — and therefore would have never been adopted — if "engineer" already captured that sentiment under common use.
> When using AI, you become a software engineer.
No. You do not. It may make you a developer, at best. I don't even call my self a software engineer, because I'm not. I'm a self taught coder that has spent 25+ years gaining experience, but I've never graduated from a school with any kind on engineering degree. I started CSE way back in the 90s, but stopped because life got in the way.
Maybe you're joking, but you just know people actually feel this way. They have no idea the difference of a coder and an SWE, and flippant comments don't help
> but I've never graduated from a school with any kind on engineering degree.
So? Per the dictionary, engineer is clearly defined as: A person who designs, builds, or maintains machines, structures, or systems. There is no mention of school or having an engineering degree.
It has always been a bit debatable if software fits into machine, structure, or system, granted, but we generally have come to agree that it does. And per the context of discussion, we've already established that it does for the sake of discussion. On that understanding, designing/building/maintaining a system in "LLM code" instead of C++ code is fundamentally no different.
You're likely confusing engineer with Professional Engineer™, but that's something else entirely. That obviously has nothing to do with anything that we're talking about here.
> When using AI, you become a software engineer.
Stopped reading.
VR flight simulator software is accessible to the layman. Does that make them qualified to be a captain (pilot-in-command) for a commercial passenger plane?
> VR flight simulator software is accessible to the layman. Does that make them qualified to be a captain (pilot-in-command) for a commercial passenger plane?
No. They might be able to fly a plane poorly, though. Engineer doesn't imply being qualified, only engaging in the act of designing, building, or maintaining a machine, structure, or system. You don't have to be qualified, or even be good at it, to carry out those acts.
You're probably thinking of Professional Engineer™, which does represent recognized qualifications, but that's something completely different. Obviously if Professional Engineer™ was meant, Professional Engineer™ would have been written.
Huh, Pilot/ Captain job requirement for commercial plane is highly regulated by authorities like FAA etc but software engineer has no such requirement. Any random business with some basic software requirement can ask an employee or contractor to get something developed quickly and deployed it. They may start calling that person software engineer.
Further even if you have some strict ACM/IEEE definition of Software Engineer®, a person is not going to end up in jail if they don't fulfill those but call themselves software engineer nonetheless.
> Huh, Pilot/ Captain job requirement for commercial plane is highly regulated by authorities like FAA etc but software engineer has no such requirement. ny random business with some basic software requirement can ask an employee or contractor to get something developed quickly and deployed it. They may start calling that person software engineer.
Exactly the problem. Secondly, if I am building commercial plane software for pilots to use, you wouldn't want to hire unqualified / in-experienced 'engineers' for all the critical work and validation testing. (or even AI vibe-coders picked from anywhere.)
Because surely, that worked out for Boeing. [0] /s
> Further even if you have some strict ACM/IEEE definition of Software Engineer®, a person is not going to end up in jail if they don't fulfill those but call themselves software engineer nonetheless.
So we are now defending fraud if one calls themselves an SWE on their CV with zero experience other than an AI doing all the coding?
It's like you want to take the legal risk hoping that the employer / company won't sue you for fraud when that vibe-coded software goes all wrong and money is lost.
[0] https://www.industryweek.com/supply-chain/article/22027840/b...
At the end of this day.. and of the next. But at some point, the tool will "suddenly" turn into a versatile agent, and that time might be a lot sooner than most expect (c.f. "exponential growth surprise factor"...)
I've been getting lots of value out of AI coding tools; especially in the last few months. My assessment is that this will lead to more work to do, not less. It's not a zero sum game. More of what I do is baby sitting AIs. But together we do more than me with my team before. It will reflect in my hiring decisions as well. I want people that can do this responsibly (are able to tell good from bad code, are able to think for themselves, are able to get stuff done).
My observation so far is that micromanaging AIs still sucks
Yes, my observation so far is that management is less judicious with what projects we take on, not that we can get by with fewer people.
to play devils advocate when a company is faced with increasing pnl ( reducing headcount) or trying new ideas why won't they take the former since it is guaranteed
There's a ceiling on how much you can increase profit just by substituting existing human roles with compute. At some point you need to expand the workforce to raise profits further. And businesses are definitely willing to take risks for the potential of massively scaling up profits. The "risks" here being a larger workforce combined with more compute.
> First, let’s establish our benchmark: job postings dropped 8% in 2025 compared to the same period in 2024. Indeed reported a 7.3% year-over-year decline for US jobs recently, so this was a good sanity check, and told me that my data most likely was comprehensive.
Doesn't this also suggest that the job market is in such an unusual state, that any further analysis makes little or no sense?
Yeah, I was wondering something similar, especially because there was a huge surge in tech job hiring during the pandemic, and part of the story from 2020-2025 is the industry regressing back to the trendline before then. Some people say the rise of ChatGPT (in what, 2023?) is part of that but I'm not convinced.
I suspect much of the data is influenced by speculative firing/rehiring as c-suites are exploring what they can actually skimp on and what they cannot.
Why would that be?
There is no causality analysis here, and really no justification for why AI is the reason for these job losses. An alternative explanation is the tariffs.
Same is true for sustainability related jobs. Given the backlash first from Texas, Florida, and other red states and now from the Trump administration, it is not at all surprising that these jobs are in decline. AI has nothing to do with it.
Such an interesting thing! I wonder why mobile engineer is down 5%. Maybe companies are moving to React Native / Flutter / Tauri / Electron ?
Maybe Apps are less of a priority now?
I'm not familiar in this industry so someone help me out here
My experience using Claude code is that it's excellent at React/React Native and because apps are mostly a View layer the consequences of bad "vibes" are less far reaching.
It depends on a lot of things, a lot of mobile development has shifted overseas, and this report only focuses on US jobs. I can't tell if this is from domestic AI, or outsourcing that uses AI.
The decline in frontend engineering jobs matches my personal impression from the past year. Smaller companies can do a lot with vibe coding alone, while larger companies can multiply their FE team productivity without making any new hires.
Frontend code can be very repetitive / labour intensive, I bet that has made this more attainable than for other layers in the stack. Most mistakes in UI code are also easily corrected and have a very tiny impact radius.
Yeah, I've found LLMs to be pretty decent at frontend stuff.
I got a new job recently and my main task has been overhauling a Claude generated app into something functional. The backend is an unsalvageable disaster, but the frontend is actually pretty good. Which is great for me because I suck at UIs.
I think if you have a solid backend already in place, an LLM can produce a pretty functional frontend with far less effort to fix up.
I think that has more to do with the fact that JS is the most popular language in the world, especially available code to train LLMs on.
Backend code is far less likely to be available because it's more likely to be closed source and is spread over many languages (Java, Python, C#, JS, Ruby, Go, PHP...).
> While I acknowledge not all job postings result in a hire, and some are ‘ghost jobs’, since I was comparing the relative growth in job titles, this didn’t seem like a big issue to me.
Uhhh... That's a very big issue. In addition, there is also a glaringly obvious analysis error here related to layoffs/attrition.
No offence but is this analysis vibe analyzed or something?
All this is actually measuring is how the number of job listings in a specific industry have changed globally. That's not the same thing as "what jobs AI is replacing" at all!
Like nurses decreasing 11%. There is 0% chance that AI is disrupting nurses. If anything, nursing is probably in more demand due to continual aging. It might just be that fewer people are quitting because of covid being over. The total number of nurses might be going up still!
Are we just upvoting anything with AI in the title now? HN used to be pretty respectful of scientific methods. The methodology section here just reads like personal resume filler for showing that you can use AI. I mean I get it and I admire the hustle, but the quality here is pretty lacking overall. https://bloomberry.com/blog/i-analyzed-180m-jobs-to-see-what...
I perform a different analysis, focused more on the type of blue collar jobs that most people believe are the prime candidates for AI/robotic job replacement:
https://emusings.substack.com/p/is-automation-going-to-eat-y...
You switch from global automation to US car manufacturing in the first paragraph without a even a line break, basically equating them.
The first graph you show is missing the corresponding table from the source which shows the exact opposite of what you have stated: the workforce had been steadily increasing until ~2019 and is in decline since then. Not massively but still it contradicts what you stated. Next you switch to Korea and the wage increase without aligning a corresponding price increase or adjust for inflation. Then you switch to the history on the industrial revolution. At that point I gave up because I know the point you are trying to make and I can see, you adjusted your samples to match that point.
I go where I have data.
If you have additional data, I would be grateful for it and I will amend my article.
Would be cool to see cybersecurity covered here as well. I find it interesting how the machine learning engineer is exploding, but every engineer I talk to about it suggests these positions are mostly doing non-productive "bullshit" work and not really contributing much (arguably this applies to many tech jobs). Obviously that's anecdotal and probably not the real story, but it does seem like "machine learning engineer" is a broad term and probably doesn't tell the full story in itself.
I wonder what the numbers look like if you compare pre-pandemic to now. To me it feels like during the pandemic companies found out they can fire a lot of people without a lot of backlash as long as they have a weak excuse. I feel like AI taking over jobs is such an excuse, especially considering creative jobs and the layers that are supposedly being replaced by AI in the article they call it "Creative roles that “execute” ".
> I developed a taxonomy of 650 distinct, normalized job titles (graphics designer, nurse, etc)
O’NET taxonomy is better for this task, IMHO.
The US is in a recession when the AI bubble is taken out. So obviously jobs are in decline. Machine learning hires are up because there are billions of dollars in circular financing.
While consumers are generally receptive to assistive machine learning like speech recognition, translation, limited auto-complete I think there is a huge backlash towards it being used to generate content and as far as I know all the major AI companies are losing huge amounts of money with every transaction. I am sure AI is being widely used as a reason for staff reductions but the reality is more likely sales are down, costs are up and head count is too high for current business requirements.
using job listings as input instantaneously garbage in/garbage out this experiment
AI is taking jobs but for a different reason. Big Tech is in an AI race so they will shrink their staff or move jobs to India as much as possible to free capital for GPU datacenters. To hell with the consequences.
Meta, Alphabet, and Oracle are even issuing almost $60 billions in bonds. Unheard of.
I am curious to see if jobs are actually moving offshore. Is there any data suggesting this?
I was under the impression Big Tech is cutting jobs across the board, all regions.
Microsoft announces US $3bn investment over two years in India https://news.microsoft.com/en-in/microsoft-announces-us-3bn-... (Jan 2025)
Google announces $15B investment in AI hub in India https://apnews.com/article/google-artificial-intelligence-vi... (3 weeks ago)
[Indian] ex-Accenture CTO named Google Cloud’s Chief Product https://www.hindustantimes.com/trending/us/who-is-karthik-na... (last week) (a lot of people speculate they named an Indian Accenture guy to move as much as possible to India)
Big Tech giants defy US-India trade tensions, record strongest 12-month headcount growth in India in 3 years https://www.moneycontrol.com/technology/big-tech-giants-defy... (September)
https://www.reuters.com/world/india/openai-launch-first-indi...
https://www.anthropic.com/news/expanding-global-operations-t...
...
Okay thanks for the links. I have supposedly not been reading the news.
Cannibalising their businesses....
what does engineering look like if you remove ml engineer?
how does one move from backend to ml engineer
This is part of the picture, but in terms of impact one also needs to look at this in terms of absolute numbers not just percentages. Two job positions can both drop by 10% and in one case it represents 100,000 jobs and in the other case 500,000 jobs.
>180M jobs
I doubt they have this data. I highly doubt they have this data.
OP here. You're right to be skeptical, there's an API here if you want to see a sample of the data: https://docs.bloomberry.com/reference/get_signals-jobs-json#...
Hey, I liked the analysis thanks! Any reason why you removed SDRs, Account Executives and similar Sales related roles? I would love to see how that compares to the data you do present.
Thanks for sharing your work.
I added a section on sales jobs just now: https://bloomberry.com/blog/i-analyzed-180m-jobs-to-see-what...
You rock man! :) really thanks a lot for sharing this.
The short answer is because I ran out of time :) but I will definitely do some analysis on sales jobs, and either add it to the post or make a new one
Thanks! Great data source.
It's just that 180M seemed a bit high.
10x more than this one (also yours): https://bloomberry.com/blog/how-ai-is-disrupting-the-tech-jo...
Btw. I'm not arguing about the quality of your posts, which I consider really good :).
Back after 1 day. I did the homework.
I analyzed the data, only ~100M job postings. There's a lot of repetition (some jobs are posted as much as >10 times).
The actual number of real, unique jobs in there should be on the order of 10M and I'm being generous.
They probably have this data. But the data is most definitely heavily polluted: Fake job listings, jobs posted to several sites, sites that aggregate job listings etc. This data is probably more polluted by AI then AI has influenced job openings itself.
Demand for senior leadership, like director and VP positions, is likely growing because the boomers are retiring.
I've got AI doing my taxes. It's saved me both bookkeeping and CPA costs.
which model? what is your workflow like
Flipping between GPTs and Gemini. Mostly having it help with interpretation of instructions and giving direction to find that data in my books. Stuff my CPA did. Help me categorize an expense (easy). And I had it make a web-form for all the data entry to produce the necessary PDFs.