Software engineering job openings hit five-year low?
blog.pragmaticengineer.com
blog.pragmaticengineer.com
The easiest explanation for the whole chart is "Software dev is more reactive to the monetary policy environment than most other industries, because more of it is funded by new capital investment -- whether VC money or otherwise -- instead of ongoing operations of stable firms compared to most other industries."
Trying to add other explanations is fun, but there's not a lot of evidence for any of them, or even that more explanations are needed.
No expensive machines or lab equipment, sometimes not even office space needed, just people with laptops
But the cycle wave length should ... Be similar to how many years people spend studying? That doesn't seem to be the case here?
Quoting the Wikipedia link:
"high salaries in a particular sector lead to an increased number of students studying the relevant subject; when these students enter the job market at the same time after several years of studying, their job prospects and salaries are much worse due to the new surplus of applicants."
Maybe you could say it's related to investors' pork cycle (of different/unknown wavelength) not students' pork cycle :-)
Or maybe just one-time opportunism: low interest rates etc
Like, forcing some not very edge case friendly set of text field validation rules onto the beurocrats such that he can't just do what he tries to do as he could with paper work.
In many orgs. nowadays the processes seem to be made to fit the computer programs not the needs of the beurocracy.
However, the flexibility and adaptability of human interaction is lost when an intermediate digital system enforces rules without exception, and that loss can absolutely result in inefficiencies because the world is always changing and your digital models rarely reflect reality.
Let's say you work for a government agency processing stolen bikes. To fill in your form you must select a category of bike, but "fatbike" is not an option as the software was built in 2012. As a human using paper, you could easily talk to your colleagues and agree that you can now check the "e-bike" checkmark and write "fat" beside it to indicate a fatbike. In the software world, some company (hopefully the one that built the original software) will quote you 50k to do it, it won't work correctly, and you'll thank them for the pleasure of dealing with them.
In another example, let's say you have to get a building permit to build an extension to your house, and based on your postal code the system tells you that you are too close to a nature reserve and are not allowed to do any construction, and the system rejects your application. The government has recently ruled that this does not apply anymore to residential housing development, to improve the housing crisis situation, but the software is not up to date with this change. In a paper world the human could simply approve the permit based on their knowledge of the real world, in the software world the computer program will never approve your permit.
This whole thing may sound anti-automation. Its not meant to be, rather its a reminder that the processes we automate should leave enough flexibility and adaptability to deal with the real world. This does not come for free though: the bike system would need to allow you to add categories, or a free-form details field. The permit system should work in an advisory manner, but allow overrides by a professional specifying a reason, or it should allow you to disable/modify/create rules on the fly.
However, no matter what you do, form validation will always leave somebody out in the cold, some edge case uncovered, that a human could always resolve, but a system cannot.
The local IT market is quite small and Google is a very large employer here.
Two things had happened then:
1. Other companies froze hiring, because they got scared of „what does FAANG know that we don’t that they’re reducing employment” - no other reason really
2. A few hundred of pretty good Google engineers flooded the small market.
2023 was very rough, things got much better since then though.
... to begin the loop again
Sec 175 especially hurt small to mid-size tech companies (especially fast growing ones...) which I imagine amplified the problem across the board.
You mean the change from deductible as a current expense to five-year amortization under changes to Sec 174? In an easy money situation, the way it increases tax liability in year one while decreasing in year 2-5 is not a big deal, it becomes more of an issue as the cost of financing goes up, so the main effect is to increase the already-high sensitivity of software development to tightening credit.
FRED is just a descriptive tool: https://www.nytimes.com/2024/12/06/business/economy/fred-eco...
That interest rates started to rise right when the tax changes to make it less advantageous happened? 1-2 punch.
This impacts all other markets.
You are not getting rich by throwing $50k to trash.
But I don’t believe they get 10% more like 2-3% and most likely they pay back any amount as soon as their stock value goes up to cover the loan not to accumulate compounding.
They have deals normal people don’t get. For banks those are free and sure money deals as well.
A reasonably rich person with good credit is paying nowhere near that.
And a very rich person with great credit and who will put their assets up as collateral is paying low single digits in interest, if even.
But yes. If you're very ambitious and very rich, you can probably find a decent loan for your idea. This odd economic situation is ripe for uncontested disruption, but has an extremely high cost of entry (hence, being unconstested).
I suspect its more a structural issue
e.g. because of Big VC (a16z, other firms with dozens or hundreds of investing staff), VCs don't stick around firms long enough for real returns (cash distributed) to matter. If you're at a place and your stuff is marked up 10x after 4 years, you just hop to become a GP and try to ride the next markup wave. Even if these people exit VC after 10 years that is still 10 years of deals that all flopped.
This might not even be the individual VC's fault -- there may just be too many venture dollars chasing too few power law returns, so you get a surge of startups circa 2019-2022 that all disappear
This feels right to me: the common trend between the dotcom bubble, mortgage bubble, tech bubble, blockchain bubble, and now the AI bubble has been big investors very high returns to make up for previous losses, like a gambler trying to win back, and it’s destroyed a lot of companies which had an idea which could have been a sustainable medium-size business but were funded and tried to grow as the next Google/Facebook without apparent recognition of how unusual advertising is for the ability to grow profits rapidly without scaling costs at the same rate.
The problem is that in this non-ideal world value creation can become completely detached from returns: when instead of investing into people creating value, or distant future returns, or even medium term returns, it becomes about investing into other people investing in it—not too different from a pyramid scheme and various pump-and-dumps.
Where is there any indication that FRED wants this?
Last week I see the exact job I applied to posted yet again. Ghost jobs really need to be punished. If you aren't actively talking to candidates within X days of posting or are reposting a job before Y days, there needs to be some disincenive to not do that.
Take a look at this, the previous administration hired hundreds of thousands of gov workers to inflate the employment numbers.
EDIT: here you go, just looking at Federal employees [2]:
Jan 2021: 2.89 million
Jan 2025: 3.02 million
The Federal government added ~130k employees over the four years that Biden was in office, an increase of about 4.4%.
During that exact same period, all private employment went from 121 million to 135 million, an increase of about 11%. So Federal employment grew slower under Biden than private employment did. [3]
Sources:
1. Total government employees: https://fred.stlouisfed.org/series/USGOVT
2. Total Federal government employees: https://fred.stlouisfed.org/series/CES9091000001
3. Total private employment: https://fred.stlouisfed.org/series/USPRIV
source: https://www.npr.org/2025/02/15/nx-s1-5298182/trumps-probatio...
If this has general ghost jobs caught in the crossfire too, that'd be amazing. Just don't waste my time on something you already know doesn't exist.
the whole point is to obfuscate growth or decline to blur the ability of business intelligence units to make prediction about the company.
it's a scam for applicants, and a scam on the market.
I'm not convinced H1B issues are related, though I'm sure there are fake adverts for jobs that are unfillable to get H1Bs in, too. Unfortunately Musk, Ramaswamy, and DOGE are all about cheap workers.
It's happening, and happening fast.
I used to work at a SF startup before the FAANG about 4-5 years back. All the engineers there have unfortunately been replaced with those in LATAM too. LATAM engineers were making like $80-$90K in USD, which is apparently really good money there, whereas US engineer asked for base pay twice that. So it seems like a win-win deal for the cofounders and LATAM engineers.
Just not so great for those of us in the US.
Whether working collaboratively in an office was beneficial or not, employers certainly believed it to be so. They believed it so much that employers, pre pandemic, were willing to hire employees a single time zone away from SF, and then pay them a significant one time amount along with a much higher annual salary to relocate them to SF.
This belief in higher productivity when teams were geographically collocated was the entire basis for which employers were willing to pay American employees several multiples of what they would pay employees abroad.
But the same employees who were benefiting from this decided they didn’t want to keep that advantage anymore.
And now since WFH has become the norm in the US, employers are realizing there’s no need to pay to keep employees in the U.S. and outsourcing to significantly cheaper locations instead.
It’s also much harder in general to communicate less formally and form deep relationships.
For execs and managers, this can be a big problem.
If it isn’t possible to do it the ‘old way’, remote work can and does work (albeit in different ways and with different constraints). But if doing remote work, why not do it somewhere else?
If you're totally disconnected from the work, sure. But for someone actually paying, it shouldn't be that hard to see whether the list of priorities for this month is actually getting finished or not.
Outsourcing manufacturing had the same problem - people thought they could send drawings to China, and with some straightforward checks, could get what they wanted cheaper. That was not at all the case, however, and there is a LOT of QA, additional checks, additional data leakage and competitive risks, etc.
For instance, if in the US people might actually follow the law around things like NDAs and non-competes most of the time, what about the jurisdiction you’re outsourcing too? How would you even know? How would you enforce consequences? How about data security?
Sounds more like a personal problem than a technical one. Besides, if it was actually enough of an issue there wouldn't be so much outsourcing. I just see it as more smokescreens to cut costs.
It saves a ton of money if you can get people to buy the end product (costs are often literally 1/2 or 1/10th). Profit is a major motivator.
Only if you are thinking like a McDonald’s shift manager. Measuring software developers by time-in-seat is tacitly acknowledging that your managers aren’t doing their jobs up to the C-suite.
“Whoa we can do this cheaper in another country???” is not a new phenomenon. Companies have been trying to do this for literally decades at this point. The reason so many jobs still exist in the US is because it doesn’t end up working that well. I’m not _that_ old and I was around at IBM for two separate rounds of “outsource-it-to-India-no-wait-bring-it-back” and IBM was and still is very keen to ship everything overseas.
I’m not saying it won’t eventually stick, but I think it’s important to dispel this notion that WFH made everyone have this sudden epiphany that US workers aren’t necessary anymore. It’s been going on for ages.
Right now (and back then), primary focus was on cost. Got to juice those margins somehow, and it isn’t going to happen by opening new markets (probably!).
When outsourcing is less envogue, it’s usually because there is some big growth opportunity and they are trying to get things moving as fast as possible. That tends to require higher end (but more expensive!) staff working in closer communication with management/execs.
It’s the difference between trying to improve margins on an existing product, and trying to grow the top end of a new product.
5 years won't change that lol. tata is still going to suck, the difference is now there is TCS in LATAM. same for cap gemini, cognizant, hcl, etc.
If you look at the table of salaries of a big union [1], you can multiply that number by about 13.5 to get the annual salary. Can vary a little %, but a good estimation.
The salaries in the EU stagnated in the last years, whereas in USA they were moving.
[1] https://www.igmetall.de/download/20240809_Metall_Elektroindu...
For 50k net per year, or ~4k per month, in most other countries you need to earn ~80k EUR per year (gross). In Bulgaria, you need ~57k EUR.
For 75k net per year, or ~6k per month, in most other countries you need to earn ~130k EUR per year (gross). In Bulgaria, you need ~85k EUR.
Go figure, these are really large differences and that very well might be the reason why we see a lot of big companies there. Tax is unusually low.
$ python zaplata.py 10000
10000
neto : 10000 -> bruto= 11680.23, razhod: 12445.10
bruto: 10000 -> neto = 8487.80, razhod: 10764.88
$ python zaplata.py -2024 10000
-2024
10000
neto : 10000 -> bruto= 11627.86, razhod: 12322.36
bruto: 10000 -> neto = 8534.93, razhod: 10694.50
p.s. And no, i do not see a change here - all is still dead - am looking since september
The boring reality is that the price of labor roughly reflects the productivity companies typically get. Because if there is ever some clear win in some location, people start hiring there, and things start to balance out again.
They were somewhere between ok and not good. Felt like we got about what we paid for - their cost was about 50% of a US dev. They were as productive as a low end US dev, so in some cases ok value but since hiring low performers tends to hurt a team overall they weren’t worth it.
We’re a mid range company (decent cash comp, not much else since startup). I’m sure if you’re paying more you can do better. But that’s how it’s always been hasn’t it?
More specifically:
To be in the top 10% of earners in the USA, you need an annual income of at least $148,812
To be in the top 5% of earners in the USA, you need an annual income of at least $352,773
The median salary in the USA is approximately $59,428 per year
This is the usual ebb and flow of offshoring in the tech industry that's been going on since the 90s.
There has never been a scalable arbitrage for tech talent by going over to another geo, and there won't be one in the foreseeable future. What US HCoL talent gets paid is unfortunately the fair and natural market rate for that caliber of talent, at least if you need to hire more than like 10 people.
Companies in high income areas have two options for outsourcing:
1. Just go for the cheapest salary, quantity over quality style.
2. Find a replacement that has a comparable level of skill, experience and knowledge to the local talent they have or would need.
Now for (1), there is almost no floor. You can get devs for five bucks an hour. For some people, it's the only way they can make any money, so they take it.
For (2), people will gravitate towards earning 10-30 % less than a local equivalent developer. Low prices drive up demand, and since there isn't that much supply of strong candidates, that drives up their fees. At the same time, the employer pays a bit of a tax in terms of time zone, proximity and cultural differences, which introduces overhead, so it can't quite reach the local salaries. But it can get pretty close.
A lot of the outsourcing horror stories are (1). A lot of the actual success with outsourcing is, in my experience, (2).
I believe (1) is primarily responsible for these ebbs and flows you mention. The feedback loop for software development (in terms of whether it pays off economically) are _long_. That seems to force history to repeat all the time, because decision makers rarely see the consequences of their actions. The incoming generation of leadership did though. And then the generation after that didn't.
1. One $unit of $resource from supplier A is cheaper than from supplier B.
2. The price difference is sufficient to allocate a certain amount to dealing with quality issues (from lower efficiency to lawsuits).
3. If it turns out the quality issues are higher than expected, switch to supplier B.
The thing is, this doesn't work with software development. (2) has that pesky long feedback cycle. (3) typically is about equivalent to starting over from scratch.
First, my experience with devs at mid level companies is that the average dev ranges from competent (as in, can work within an existing framework to accomplish tasks), to incompetent (cannot do the same consistently), with there being a top 10-20% who are truly good (able to deliver complex high quality software from a blank sheet, capable of working on complex existing systems meaningfully etc.). This applies to all companies (including ones in the US).
Second, at high level companies, there's lots of European talent (usually in West Europe) - these people are as good as their US peers (by definition, and usually has been my experience), but make less than what you would take home in a mid level US company.
>Second, at high level companies, there's lots of European talent
Lots is subjective. The problem is the scale. You can easily hire 100 great devs in London over some period of time. 10000? Maybe not. It would take you too long to pick them up.
There is friction in turning fresh graduates into experienced developers. There is friction in hiring people in a region where you are not operating already. There is friction in using additional middlemen. There is friction in learning to do business in a new situation or in a new culture. But the friction can be overcome – in the long term and when the scale is large enough.
In no small part because if you look at that talent you’ll notice that… they’re mostly not natives to those areas. i.e. the highly paid in San Francisco are mostly immigrants from LCOL countries. Generous immigration policies ensure it. If immigration policies do change, you might start to find tech top talent in other geos but for the most part the people they really wanted to outsource to in the first place have already moved next door.
There are great engineers around the world but why would they accept ten beans an hour if they can take a plane to a different country and earn fifty?
The talent from LATAM are not just sitting in the rainforest waiting for your phonecall.
Also really top devs (at least in Brazil) can make more than 50k USD per year (fluctuates widely based on exchange rates) at Brazilian companies, it is cheaper than US but the tail-end of talent can still be expensive.
If you really want to attract good devs in cheaper areas the best way is to just open a branch of your office there, pay top-salary and don't just use it a dump ground for the projects no one at home wants to take. So treat them the same as at home. I had a roommate who worked in one of those companies that take offshoring projects, lets just say it is not the best talent pool and the good devs there leave fast.
Not that you're implying it, but I think it's fair to point out that people in ex. Brazil aren't worse at dev work, just that there is probably lower incentive to be good.
Just as you compete for top talent locally you also compete for top talent globally when hiring remote.
I live in Sweden now and have dealt with a lot of Swedish companies over the years. It is hard to explain this to the locals because they see Sweden as a big tech hub where all the talent is at, but the scale of things in Brazil is _crazy_. I worked in a telecom project where this one single company had more active (paying) SIM cards than the population of Sweden several times over.
Just because it is a poorer country doesn't mean it is easier engineering, if anything the engineering is harder.
I’ve been hearing about outsourcing destroying jobs since the 90s and that’s how it’s always gone: the suits salivated at the prospect of cutting wages by, say, 90% and had total write-offs because of it, because they missed that even in a poor country people have options and anyone smart enough to be a good developer is also smart enough to recognize that their skills are worth more. Outsourcing has some big costs related to communications so while you could find decent people at 50-70% of local labor rates, coordination overhead makes that a net loss even before you hit things like the security risks.
Part of the problem there is that the business people really want to think that they understand their business so well that they can give perfect instructions, and it takes a certain humility to recognize that more time goes into knowledge transfer and discovering the true needs than might be obvious.
I reckon there's a lot of "garbage in garbage out" going on, and if an org took the time and effort to actually treat Indian (or any outsourced team) devs like their own, things will be much better. But when you do that the overhead shoots up, and that cost/benefit analysis you did during the proposal to offshore flies out the window.
Outsourcing is a cost savings measure. It's a symptom, not the cause of the issue
The demand for engineers during the pandemic may have been artificial, but there is still need for people today. Some companies don't hire because they were built to be sold.
I can't agree with the conclusion. I can come with tens of examples of when using AI doesn't work.
Also, software engineering means much more than writing code.
What if it speeds up the part of pumping out widgets in the widget factory, shipping Spring Boot / Laravel / whatever projects more quickly and therefore individual devs becoming more productive on cookie cutter projects (without necessarily getting more money for it either).
Now in the AI boom era, 90% of the capital goes into the AI while 10% goes into SE jobs. That could explain the drop in SE jobs.
The headline doesn’t tell the whole story. SE openings precipitated in the year that followed the peak. Compared to that massive decline, openings over the last year have been relatively stable but on a downward trajectory.
Most of us are personally familiar with the reasons for this. Companies overhired during COVID and the demand for engineers was met by higher CS enrolment and to smaller extent, boot camps. But then remote work made companies realize that they could hire all over the world. Right now Netflix is only hiring engineers in Poland and other Faang companies are mostly interested in South America.
Focus on AI shifts investment in technology into that field, regardless of whether it makes sense or not. Likewise management expects that AI will drive up engineer productivity.
There has been no significant decline in software engineering positions since the advent of AI as all the factors I mentioned above still apply in the post 2023 period.
This is patently false. Source: I’m an engineering manager at Netflix who’s hiring in the US and whose peers are all doing the same.
When was the peak of AI? My impression was around 2022, and the crash started the next year.
I don't think AI is the cause so much as projected economic headwinds and S174. But I can see AI being indirectly responsible. Not because of replacing jobs, but because Companies want to redirect dev funds saved into powering AI servers. I believe Meta said as much at the beginning of 2025 (or maybe it was Google?)
Just think how much the investor narrative has shifted from early 2020s. Back then everyone talked about the business areas they were expanding into using tech. Now, investors get hyped up when you do the same thing, at lower cost. Less about growth, more about predictability. Which makes sense. Do investors view Amazon as a high tech company (as in 2020 and before) or a predictable business? dare I say utility?
It explains layoffs, forced RTO, etc. They simply don’t care if even their best leave. Their MBA executives know optimization/keeping the lights on - and less about innovation.
So it's a bit of both extremes at once.
Few, maybe 15%, of those people really seemed like they could program. Everybody else talked about their favorite tech stack and their opinions on code styles. It kind of felt like code masturbation. A shrinking job market is not surprising.
It's rapidly looking like nerding out while pining for the old days where engineering knowledge and education was valuable is maladaptive. The sunk cost fallacy can be real.
It's best to start hedging and learning other skills. Or focus on making as much money as possible ASAP so you have some financial safety net to fall back to as AI and Trump/Musk and offshoring has increased in recent years. All of us need an escape plan ready.
I don't mean to spread FUD. Just hedge please. I assume I have at most a decade of tech career left in the best case. A couple years in the worst case.
The Internet is basically a young, small portion of writings in english, as far as I can tell.
Even if everything being done with LLMs is useless, if you can do the same amount of useless with half as many developers, it's bad news for us.
Later moved to Oracle / SQL Server / C# / microservices / CICD etc etc) but hell, that was my whole career 2002-2012 across various companies. Making cool tools for smart people, things the accelerated the company in smal;l but measurable ways, that unbeknownst to me would absolutely have directly resulted in a team of 3-5 being reduced to a team of 1, or a team of 1 not being expanded to be a team of 3-5.
The same will happen (IS happening) with LLM's.
"No doubt, in turning them out of this “temporal” world, the machinery caused them no more than “a temporary inconvenience.” For the rest, since machinery is continually seizing upon new fields of production, its temporary effect is really permanent. Hence, the character of independence and estrangement which the capitalist mode of production as a whole gives to the instruments of labour and to the product, as against the workman, is developed by means of machinery into a thorough antagonism. Therefore, it is with the advent of machinery, that the workman for the first time brutally revolts against the instruments of labour."
soon we'll see if the old man was wrong and short-sighted, or if what has happened since the internet revolution is just an abnormal period in centuries
I’m not sure that’s the case even if in retrospect we can clearly argue this
In 1998, Paul Krugman, winner of the Nobel memorial prize in economic sciences, infamously predicted that “the growth of the Internet will slow drastically, as the flaw in ‘Metcalfe’s law’—which states that the number of potential connections in a network is proportional to the square of the number of participants—becomes apparent: most people have nothing to say to each other! By 2005 or so, it will become clear that the Internet’s impact on the economy has been no greater than the fax machine’s” [1] and even though that turned out to be spectacularly wrong it shows the attitude in the early days wasn’t one of absolute certainty.
And certainly the current AI boom is most visibly known for LLMs but there is a lot more happening beyond chatbots.
[1] https://www.snopes.com/fact-check/paul-krugman-internets-eff...
Sears famously laughing at online orders is a great example. They already had the mail over market owned, and already had the distribution, sourcing, and utterly dominant brand recognition in that space. People used to order homes from Sears!
They just needed the online catalog. But the CEO was a psychopath Randian who thought the internet was a fad and now Amazon runs things.
Same with Toy R Us, and Radioshack, which did a lot of mail order.
https://www.youtube.com/watch?v=3yX9D3h2F_0
Ironic that Sears back in the day got started with mail order.
Seems like bad management is a cancer on a lot of organizations. Their size protects them for a while, but eventually they get outcompeted.
It certainly feels this way if your only perspective on the tech industry comes from headlines and social media.
Step outside of the doomerism whirlwind that is internet headlines and it's a very different story. There are a lot of companies doing great things and delivering products that people use, but they don't make for catchy headlines. So you don't hear about them.
The automation industry with physical products interaction is where my labor shortage lives.
I will be requesting support this year. Software development is about solutions and all industry pivots have a learning curve.
This also means it is a great time to start a small, profitable business because labor is undervalued.
The final phase of this is where Americans start moving to some hut in the Congo to survive off the market wages offered when the supply side of the supply/demand curve blows up.
Edit: I am an idiot, there is a mention. Or possibly the article was edited and I'm not an idiot, it does say there was an edit recently.
I originally didn’t mention this because S174 impacts the US and US-HQ’d companies. In this data other countries like Germany, UK, France all see a similar drop. Also, S174 impact likely really started from early 2024, when companies impacted had to pay high taxes and realized the change is here to stay with no end in sight. Doesn’t explain the drop since 2022.
Updated the article to make this clear though. It was not in the original version - thanks for the note!
Bullshit. Spotting hallucinations requires expertise in the subject matter, and most devs wouldn't be using LLMs if they intimately knew the programming languages and APIs the LLM is generating.
The reality is that spotting hallucinations often takes more effort than reading the source documentation and writing the code from scratch, since the dev also needs to review and check the code for correctness.
With Cursor and Copilot, the days of $190k entry-level SWE roles in the US are GONE.
Juniors need to learn by falling on their face a few times, if companies don't want to train them then they'll never become seniors. Lovely cycle
It's rare that anyone gets trained by a company who doesn't already know something and is pivoting to a new role. It's unfortunate but that's just the way it's always been. Juniors have to be proactive and bust their ass to become mid level, usually by learning new stuff at home through building stuff on their own and reading obsessively. The hardest part for a junior to get is their first paid job, and more often than not, that first job will be at a shit hole. Juniors should take advantage of it and really bust their ass to impress them and improve themselves. Once they get a couple of strong learning years under their belts, it gets a bit easier.
If you are a junior and the company is paying you to learn stuff, consider yourself extremely lucky.
Companies are in desperate need of strong talent. To succeed, juniors need to force themselves to become strong talent.
When it does save me time, it's almost just some boilerplate I didn't have to Google or type out. Honestly, I feel like gen AI pleataued a year or more ago.
LLMs are just really convincing bullshit generators. They look impressive on the surface, and the times when they spit out a whole bunch of useful boilerplate feels like magic, but that stuff isn't super useful for a majority of the work you spend your time doing. Throwing more money at these companies is not magically going to yield AGI. I think the AI CEOs are basicallly selling us a lie.
I work in typescript, rust, and go - mostly typescript. LLM coding is an order of magnitude better at typescript than these other languages. To contrast, LLMs seem mostly useless for rust.
I also use cursor, which is a big jump over other code assistants.
And finally I understand how to prompt LLMs accurately. This is a new tool and from interacting with coworkers in the same codebase I can say many smart people have not yet learned how to use the tools effectively.
But if you just assume that everyone catches up to where I am today with cursor + typescript the change is massive.
It takes a bit time to build intuition around the workflow, but when you get going, it seems surprisingly useful. I was a skeptic before as well, btw.
Some hallucinations stay in codebases longer than others! If there were zero hallucinations there would really be no novel output. Some hallucinations are useful and some are not.
Facts don't really play a part there, if a response is factual its only a sign that the training set largely agreed on the facts (meaning the correlation of token sequence was high).
Also, is this the downside of the 'wait and see'
But, there could be a rebound from this.
" Software by non-developers creates more opportunities for devs. Imagine a situation where the number of non-developers creating software increases by 10x or 100x, due to AI, thanks to non-technical people creating software with AI tools and agents. "
Just like with arts/video editing.
There could be a rebound when so many people start creating things with AI,
THEN realize it is actually a bit harder than they thought, and boom, need to hire video or software engineers to finish what they started.
- Offshoring, nearshoring
- Generative AI
- Coding camps
- Better and more open source libraries. More high level open source libraries
- Low code and no code
- Easier to use technology, more documentation
Smaller teams and more bootstrapping (as opposed to venture-funded rapid growth) seems likely to reduce the reliance on recruiters who are a plague on the industry, with few exceptions.
On the other hand, maybe a lot of tasks you could previously get a lot of leverage on with a simple Perl script will just be done directly by LLMs. Not if they're customer-facing, maybe, but in cases where you just need to get some data in a different format or something.
> When disagreeing, please reply to the argument instead of calling names. (...) Please don't sneer, including at the rest of the community. (...) Please don't post insinuations about astroturfing, shilling, brigading, foreign agents, and the like. It degrades discussion and is usually mistaken.
Your comment flagrantly violates all three of these rules.
This trivial cryptogram contains more details: Tiqy mcif ufobrtohvsf'g dwgg-ghwbywbu oggvczs, mci dihfwr dirrzs ct dig.
Assuming that AI coding is already helping the average dev with shipping more.
A better graph would be of the number of people who self-report as being employed as software developer. I doubt that would show the same decline--there have not been many stories of people abandoning the profession because they could not find employment...
In 2000, none of the ideas ended up being bad. They were just too early and internet wasn’t ubiquitous.
But even then, if you were working as a standard enterprise dev for a profitable business - banks, insurance, companies, etc, outside of Silicon Valley, jobs were plentiful. I was a Windows dev living in Atlanta with 4 years of experience and had no issue of getting offers.
Things improved when internet became ubiquitous both at home and in their pockets with smart phones and the App Store by 2012.
On the B2B side, there was the rise of SaaS.
There will be no next doubling of jobs to absorb all of the people looking. Every application gets literally 1000 applications within the first day of posting.
It’s going to take at least a decade for the supply/demand to balance out.
No Section 174 is not to blame either.
If H1Bs were so destructive to the software industry, why would the software industry, which has the highest prevalence of H1Bs, be the industry with the highest growth in salaries for basically the past 2 decades?
Also, the idea that eliminating H1Bs will bring more jobs makes absolutely no sense.
If a company is asked to eliminate all their H1B positions, what do you think is more likely:
- they hire a completely new employee in the U.S. at a similar, or if the critics are right and H1B is suppressing salaries, higher salary, or
- they hire the exact same person but now with that person back in their home country and pay them a tiny fraction of what they were paying them in the U.S.?
There may be a handful of jobs where the employee needs to be in the U.S. where sure, a few more Americans will get jobs. But the overwhelming jobs will simply move abroad with the H1 employee, and along with that a whole bunch of dollars that were being spent within the U.S. and was paying US taxes will move abroad and be spent abroad and pay taxes abroad.
Is DeepMind an American company? It's owned by Google, but it was founded in, and is still HQed in, the UK, but has offices worldwide. How much appetite do you recon Canada, France, Germany, and the UK have for America's trillion-dollar-club at this point?
Now sure, you could demand any corporations selling in the US, even if not based there, hand over payroll etc. documents and then tax them at whatever the multiplier is to go from that country's salary to US salaries… if you don't mind that means some countries are banned by existing privacy laws from doing business with you, or that even this is easily gamed, or that American exceptionalism is increasingly overstaying its welcome.
I think that the internet is incompatible with late 20th century models of national sovereignty. Globalisation itself isn't the problem, it's that services performed across an international border have a very messy relationship with legal obligations, everything from surveillance obligations[0] to minimum wage laws. This will get worse when robots can be tele-operated from a different nation, blurring the division between service and non-service (primary, raw materials; secondary, manufacturing) labour.
I don't think we here in Europe would mind so much, if Big Tech obeyed local laws (even when I disapprove of the law[0], I know I can't pick and choose which laws I follow, every jurisdiction's laws are a package deal). But Big Tech seems to treat European fines for non-compliance as if they were taxes, even though a tax on an import is called a tariff and the US is fine with those.
[0] https://en.wikipedia.org/wiki/Investigatory_Powers_Act_2016, one of two reasons I left the UK
but also https://www.theverge.com/news/608145/apple-uk-icloud-encrypt...
and note that Apple disclosing they've received such an order is itself an offence.
The EU is fine with tariffs as well though. It’s not a US only activity. However, the US has a history of tariffs as income tax wasn’t even in existence until early 20th century. I’m interested to see what impacts tariffs are going to play as it does seem a reasonable tool for sovereign nations to utilize. Otherwise, how else do they prevent the hollowing out of their working class in countries that do not provide the cheapest labor? Libertarian types will simply say pick up your bootstraps and compete, but that’s not realistic against labor that is a fraction of their cost. Protectionist measures may be the only logical response for the working class to be protected.
Indeed, I don't wish to imply otherwise, I'm just saying it's a bit hypocritical to complain about something being taxed when also implementing a tax of the kind being complained about — even if the thing being complained about is actually a fine rather than a tax, complaining about it as if it were a tax makes it hypocritical. (Also, it's not like the US doesn't fine businesses for not following laws).
I, too, have no idea how this will play out. Even just from what you say, and ignoring the potential bull-in-the-China-shop that would be the economic impact of AI performing useful cognitive labour at below-subsistence-farming wages.
Maybe because that’s just not true? Salaries have generally been stagnant over the last 20 years in real terms for most in the industry.
Most developers do not work at FAANG positions, and even those positions are only found in the extremely high cost of living areas like Silicon Valley.
Companies are always free to hire people from wherever they want. However, it could be foreseeable that “American” companies that are largely overseas, maybe at a disadvantage when it comes to taxes or other benefits conferred by the US government. Such companies may find that their products are subject to tariffs when trying to sell those products back in the United States or other active measures to prevent offshore positions. There is a risk as section 147 changes already has shown.
I thought it was the MySpace of job boards.
Today: "I know this pain! It's real."
Rough out there.
this is a very stupid thing to say. The formal structure of a language doesn't easily determine the complexity of what can be done with a language, nor is it a way of predicting the difficulty of expressing an equivalent idea vs other formal or natural languages.
Well, it does have the virtue of being straightforwardly true. For example, programming languages can't be ambiguous.
Yes, in C++ we call it "unspecified behavior" and "undefined behavior"
† I'm aware that that isn't an unambiguous statement. This is for the simple reason that it's next to impossible to make an unambiguous statement in a natural language; that's why legal documents use so many clauses. I'm relying on the reader here to realize which meaning I had in mind, which is the way all natural language works.
The case with legal documents is equivalent in C sequence points for comma operator with something like `print(i++, i++)`. Imagine Boeing documentation with text "In case of blinking indicator press button A and stop immediately". Button "A and stop"? Button "stop" after button A? Authors can hope that a sane human can resolve this ambiguity, but if it is done by compiler/interpreter/robot, it can have an avalanche effect.
This is "straightforwardly" false.
Speaking specifically about the "language" part - while there is a formal specification for C++, many pieces are implementation dependent. Then there is actual undefined behavior that is part of a specification.
Actual programs implemented in a language can be ambiguous. consider multi-threaded programs where data arrive at different times in different threads, leading to different outcomes. Or just pure ambiguity of intent. Or a program which incorporates undefined behavior intentionally.
Formal and natural languages may overlap in some ways but it is ridiculous to compare them in this way and claim a probabilistic model is better at the formal language. Translation tasks are an example where LLMs perform extremely well, I would argue much better than in programming. Should I make the claim it's because of some intrinsic attribute of natural language vs formal language?
And what happens if you do?