The charts show software dev gaining more jobs under the COVID demand and stimulus swing, and then losing more as those went away and monetary policy got tighter than even it had been even before the COVID stimulus.
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.
How are sec 175 changes in the US not also a top explanation? This has a massive cost to small businesses and startups trying to hire software engineers, and a big impact to any team trying to grow.
Because the data doesn't match. The author addresses this in his blog post -- essentially every country has the same graph of SWE jobs, but only the US has sec 175.
Big tech layoffs were a global issue and U.S. tech firms hire (and fire) globally. So like it or not, U.S. economic issues have a downstream impact around the world.
Sec 175 especially hurt small to mid-size tech companies (especially fast growing ones...) which I imagine amplified the problem across the board.
> How are sec 175 changes in the US not also a top explanation?
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.
... partly because it's quicker to get started with software projects?
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
I tend to agree with you, if only because the swing in Europe on the IT job market has had less amplitude and is more spread out, similar to the difference in stimulus policy.
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.
I believe many software tools work like a vice for beurocrats making the beurocracy way less efficient.
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.
That seems a little backwards. I'm not sure exactly what you mean with making bureaucracy less efficient, but form validation is an example of automated, technologically enforced bureaucracy - in some sense, it's a more efficient form of bureaucracy. It's certainly less flexible, but lack of flexibility is arguably the point of bureaucracy; less flexibility suggests more rather than less efficient bureaucracy.
Yes and no, form validation is a double edged sword: without it whoever receives the data may have to deal with inconsistencies, so removing that will increase efficiency.
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.
Great points! To be clear, I was not trying to say that automated and strictly enforced bureaucracy is a good thing. The bureaucracy of the form is very efficiently implemented. That's mostly not a good thing (or at least it often has costly consequences), but it is an example of software creating more efficient bureaucracy - efficient as bureaucracy, but often not efficient in fulfilling its supposed goals.
Aha, ok. I'm not 100% sure I understand but I imagine you mean that bureaucracy, viewed negatively as simply the execution of some process, is expedited by form validation. I have to agree then indeed that this is true. I thought of bureaucracy more as the whole of the functioning, so efficiency would include successful resolution, not just successful process.
Thanks for the write up. I am having an emotional crisis having to defend bureaucrats against us. (I conveniently blame their systems not being FOSS, not us programmers blindly following orders. No... not our fault).
I love me some FOSS but I don't think that helps here, does it? In the end the person having domain knowledge is no longer in control of the process, because its been transferred to an automated system. Therein lies the issue, even if the software can be modified it cannot be done by the domain expert, or on the fly, or at low effort.
And even if they produce "real" products. How many of them are needed. How many of them are redoing the same thing. And how many could in perfect world be done more efficiently by resusing some common solution.
Jobs funded by speculative investment that are highly sensitive to monetary policy are pretty orthogonal to the kind of thing addressed by "bullshit jobs" theory. They may be societally harmful or low-social-value more often than other jobs, but generally for very different reasons than bullshit jobs. They are "lottery ticket" jobs, where many are wasted, some are very valuable, and its hard to predict in advance which are which (for investor value; for various reasons endemic to capitalism more generally, this is a very imperfect proxy for social value, but that problem is also mostly orthogonal to the kind off harmful job "bullshit jobs" addresses.)
FRED wants the implication to be AI related, but it seems pretty clear that all/most of this decline is related to their interest rate policies creating a raft of bullshit companies funded by too many dollars chasing ever more far-fetched returns...
Returns are no longer the actual goal. The actual goal is the perception of far reaching future returns. The perception of those returns is much more valuable than the actual returns. Because the wealthy don't want profits and dividends that get taxed. They want ever increasing stock prices so they can continuously grow their asset base to borrow against so they can never sell said assets this triggering capital gains taxes.
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 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.
Investing for distant future returns is not bad and favours long-term thinking over short-term profit. It is closer to the best (though abstract) metric, net positive value created, which naturally translates to ROI in an ideal world.
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.
Companies with actual profits can, and do, simply buy back their own stock. Inflating market expectation of future cashflows is not the only way to realize asset appreciation.
Wants is maybe too strong, but I've seen this chart in a couple of different substacks / editorials, which I assumed is the fed's press office shopping this story around because "AI taking jobs" is zeitgeist-y right now.
It's more likely that it became popular to use in commentary because this chart from FRED (both the software dev postings on Indeed chart, and, in the comments, the comparison to the overall job postings on Indeed chart) got attention on a certain orange-tinted site that is popular with people in and around and commenting on the software development field a few months ago.
People underestimate just how much of the industry was total garbage, pre-AI, pre-pandemic, the signs were all there in 2015. Who remembers "the average tenure of a software engineer in the SF bay area is 18 months". Was it ever really sustainable? That means that after just 18 months you'll get an offer from a company more fantastical than the one you're currently at, which has even more frothing-at-the-mouth investors and converted some of that into higher salary for you. And your current employer can't hope to match it, because they have already been written off by investors. We have a LONG way to fall to reach the true software industry value.
Then, assuming that the fake job posters are still there and that they now have AI help as well, the fall in the number of postings is more concerning.
Sometimes the position is simply shit. “My” job ad is open again. Looks like the third guy lasted there 6 months too. Every day new agency posts the same job. But Python developer, who can fix custom hardware from the 90s in the night shift is rare. And the salary was very average too.
I don't know how this isn't considered SEC fraud by publicly traded companies.
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.
The scams and fake job postings on the market are also increasing rapidly, outpacing the growth of overall job listings. This trend is significantly affecting the accuracy of employment and job availability tracking.
Any H1B related postings are a tiny fraction of the postings by recruiters and recruitment agencies trying to build up their databases, and companies wanting to look like they’re continuing to grow.
BLS says there are 140,000 of new job postings for sw engs annually. DOL stats show ~120,000 PERM applications (for which job postings are required - they're not required for H1Bs), of which probably only 20% are for software engineers. So 30,000 / 140,000 - ~21%. Pretty small fraction.
I am not a Trump supporter and I didn't vote for him, but the fact is that Biden's BLS numbers were always manipulated to portray the former president as successful. These statistics are unreliable, in addition, the underlying premise of relying on job postings was flawed from the start which was my initial argument.
Take a look at this, the previous administration hired hundreds of thousands of gov workers to inflate the employment numbers.
Doesn't pass the smell test since the chart for government employees clearly includes more than just Federal. The Federal workforce is about 3mm, but that graph shows 23mm [1], so it clearly can't just be Federal employees.
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]
Thanks for the additional information. The thing is the Trump administration is about to fire 200k probationary federal employees meaning they were hired within in the last year or so. These numbers don't add up either.
"A probationary employee is often a recent hire to the agency or *a long-serving employee who was moved or promoted into a new position*. They are put on a "probationary" period that typically lasts for one or two years, though it can be longer at some agencies. It's like a trial period during which the worker and their performance are under heightened scrutiny."
Our company just let go of all our remote LATAM hires. We paid high end in their LCOL areas.
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?
As a Brazilian dev living in Europe, one thing people just don't get about offshoring is that good devs have options, even in poorer countries. Companies offshoring usually offshore crappy projects, crappy projects attract bad talent.
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.
Feels a bit chicken-and-egg - if I live in, say, Brazil, and I know that there isn't a great market for good devs here, will I put in the effort to become a good dev? If I really have a passion for dev work, will I try to move or work freelance?
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.
My point was, there ARE opportunities in Brazil that don't involve working in a slave outsourcing factory. Even individual remote dev contracts are not always the top tier of talent. The real good talent working remotely from Brazil is probably working for startups in silicon valley not your average mid US company.
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.
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.
It's quite simple market forces, from where I stand.
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.
I think many companies intend to do (2) but find themselves in (1) because they try to force their assumptions about the mythical cost savings, which aren't actually there.
I've seen companies blatantly pursue the lowest possible hourly rate, kinda disregarding anything else (time zone difference, quality etc). I believe it comes from a procurement mindset (works for other areas, and probably taught in business school):
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.
I can raise two points to disprove this - I'm East Euro, but have worked with people from all over the globe in all kinds of companies.
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.
Mid level companies have wide ranging talent since they almost by definition pay below market for talent, and don't even enjoy the brand value draw of the top companies. These companies benefit from the fact that there is some lag in labor discovering their true market value.
>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.
You don't get paid for talent, or education, or responsibility, or any other reason that could be plausibly presented as fair. You get paid for friction: how easy it would be to replace you, and how much harm it would cause if a good enough replacement cannot be found. On the average, as individual circumstances vary.
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.
> I’m sure if you’re paying more you can do better. But that’s how it’s always been hasn’t it?
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.
Interesting… 50% of US dev cost is like what if you don’t mind saying? 100k?
I found the biggest issue with these engagements is the trust with remote developers.
Until you’re over the many hundreds of thousands of dollars per year range, it works pretty well as a rule of thumb. If your company wants the cost of working in an office, they’re probably not paying high salaries to work in a rundown shack and the people making those high incomes also want better quality benefits (especially in the United States where healthcare is so expensive and cheaper plans mean more time having to argue with someone whose bonus is based on denying care).
Is that 90k the "lower average"? In Ukraine? I find that very hard to believe. Even here in Czechia that would put you firmly in the top 5 %. I don't think I personally know more than 2 people who make anything like that, and one of them isn't even in IT.
Top 5% overall or top 5% of dev jobs? Former sure, latter might be not? I live in Prague, and there are definitely multiple companies that pay more than 90k to middle positions. especially if it's before taxes.
Salary is not the same as paying a contractor. So, while salaries for senior engineers is 100k, as a contractor they might be charging 150K per year ($75/hr).
My workplace is mostly hiring LATAM and parts of Europe at this point. Occasionally NA but mostly for higher level engineering positions to backfill roles for people who left.
Hailing from a popular outsourcing destination I can attest to the fact that layoffs happened here as well, but now the same people are being rehired for 10-15% less. No growth whatsoever though.
I don't mean to spread FUD, but as an interviewer at a FAANG in the US, I have mostly been interviewing candidates in LATAM (Brazil and Mexico). Same for some of my other coworkers.
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.
This was unsurprising when American employees decided to use the significant power they gained over the pandemic to insist on WFH.
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.
That doesn't explain why many execs were keen on RTO. If it was reducing costs, you would thing they would've wanted to draw out WFH for as long as possible. But there a clear rush to end WFH as soon as possible and implement unilateral mandates to get everyone back to office.
The explanation is that they wanted the (perceived) productivity back. Alternatively, they may have wanted to push overall compensation down (e.g. having some employees quit and only luring back the ones they actually like). Of course, it varies a lot from market to market, depending on how hard it is to reduce headcount in the first place (e.g., in Germany, it is much harder than in the US).
It’s really hard to see what a team/people on it is actually doing if people are remote.
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?
> It’s really hard to see what a team/people on it is actually doing if people are remote.
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.
> It’s really hard to see what a team/people on it is actually doing if people are remote.
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.
> 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.
“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.
The work force abroad has caught up a lot though. Online learning has made it much easier to improve skills and their English has also improved. And since everyone has become more used to remote work, communication is also much better now. One reason a lot of outsourcing efforts in the early 90s failed was also because the Indian teams weren't properly kept in the loop.
It goes in cycles, often depending on the business focus at the time.
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.
Exactly. That’s around salary for senior engineer at big corp according union tables in Germany. If you’re really good or a manager, 110k is in there. More is doable at senior manager/director level or in key position in a company led by owner. US faang salaries sound here like science fiction.
Yes, 90 to 110k is pretty much TOP. A lot of seniority and experience. If you want to go higher, you have to go management path (even lower management is included in the union)
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.
Currently the going annual salary for a senior developer in Bulgaria (Eastern Europe) is somewhere between EUR 50k - 75 EUR net, with 25-30 days of paid time off. The gross will be about 25% more (income tax is 10%, social security around 40%, but there is a cap on the salary base, so effectively it is less). For freelance senior developer, the (B2B) daily rate is in the range EUR 350 to EUR 550, depending on the technology and project complexity. Of course, there are rare examples outside of these ranges (e.g. top talent in ML, BigData etc). Leadership roles are usually at 10-20% premium.
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.
everything provided by offshoring to india is just an expensive tech debt full of spaghetti code. it simply doesn't work and causes harm to the customers. but because cheap it keeps getting sold. working with indians is a nightmare bar none. and i'm saying this as someone who has been to india and is also personally intrigued by its culture. but it's conway's law all the way - visit places like mumbai or delhi and this is what you'll get ... chaos that kind of works.
I am not sure how things work in US but, living in an Eastern European country, I can say that salaries started growing, almost reaching western european levels, after which many foreign companies closed offices and moved to cheaper countries like India and Pakistan.
Where in Eastern Europe? At least from my perspective (I've done hiring at several companies, albeit just hiring individual remote workers rather than open a local office) developers in Poland and Czechia are still very competitively priced (less expensive than north/western Europe and A LOT less expensive than the US).
That's pretty weird. Usually, less companies mean less demand for labor. If supply is the same, that means price-for-labor (salaries) would go down. Unless, this was accompanied by people leaving (reduction in supply). Do you have an explanation for this?
It's a pretty common place to get outsourced devs from. But it doesn't have the conglomerate offices like India, so you have to do more to get a team from there. The people are basically the same so same level education, same-ish salaries, same English understanding, etc.
When I was hired during the pandemic, I was the ~140th hire. In 2 years, that number ballooned to 700 people. Billion dollar acquisition later, I have to fill-out paperwork, fight other teams, perform a ritual sacrifice, only to have everyone reject my candidates that are deemed too expensive.
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 don't disagree with you but a different view that could support the OP hypothesis could be that prior to the AI boom we could see that 90% of the capital went into "normal" software engineering jobs while maybe only 10% of the capital went into the ML market.
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.
It might burst but it may take years for that and until then it's going to be a struggle for most people in non-ML engineering jobs. But it might not burst at all? The trends in AI developments for the past few years are a bit concerning IMO.
> 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).
This analysis is more of an excuse for speculation about applications of LLMs in software engineering, than an analysis of the factors leading up to this.
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.
I apologize, and I'm embarrassed, I thoughtlessly repeated what my friend from Netflix said in a casual conversation. He specified that his larger org at netflix is hiring only engineers from Poland. That said, I believe the point I made stands.
Companies have moved away from tech innovation driving growth. Instead they’ve realized they’re defending mature, boring business models. They need to keep the lights on with some good-enough engineers.
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.
We are in a period analogous to the end of the Dotcom bubble. The Internet is obviously going to change the world, but right now all it can do is pets.com, etoys.com, kozmo.com. People are out of work, and those with other alternatives or preferences will leave the tech industry. In 10 years entire new platforms and modes will exist that generate even more opportunities than before. Some variation of this wave has been surfed up and down since the invention and industrialization of computation.
> We are in a period analogous to the end of the Dotcom bubble. The Internet is obviously going to change the world, but right now all it can do is pets.com, etoys.com, kozmo.com
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.
I don't think it's doomerism. I started my career during the dotcom boom. I think we are right now in a transition period that is little bit of the start of the boom (just add AI! is the new "just add Internet!") and also a bit of the scorched earth after the initial boom, when all the tourists (CS is easy money!) are rerouted to other careers if at all possible.
This also means it is a great time to start a small, profitable business because labor is undervalued.
One barrier in communication is that developers are not the ones able to post for jobs. I would love to have another coworker in software development. I have no say in hiring. Even a QA with a software development background would be gold.
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.
I got some great opportunities from that post but lately, it has been a nothing burger. I have been without a job for nearly a year and have applied to many positions on that thread with just one reply and that fizzled out despite being objectively qualified (a take home assignment, successfully completed within the allotted time of 24 hours).
If the analogy is LLMs are today's equivalent of the web, I'm skeptical. Internet-based communication replaced technology like telephones and fax machines for synchronous and asynchronous telecommunication. People like and need to talk to other people. It was clearly valuable from day 1. What is the equivalent value proposition for LLMs? They make for really cool and splashy demos. But practical usage doesn't seem to keep pace with the hype. Why is it such a safe assumption that this is just like the invention of the Internet (or, as Sundar Pichai put it, the invention of fire)?
I have personally written VBA macros (or hell - just a decent Excel spreadsheet...) that have put people out of their jobs. I didn't know this at the time, I was just a young gun able to see how to make some slow repetitive jobs, faster.
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.
"Hence there is immanent in capital an inclination and constant tendency, to heighten the productiveness of labour, in order to cheapen commodities, and by such cheapening to cheapen the labourer himself. "
"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.
there was a lot of skepticism, and from some heavy hitters.
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.
Like half of software jobs are Doing Something so that management feels like they are innovative thought leaders of the tech community, while the other half of the software jobs are keeping the unsexy business critical software working.
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.
It depends. If fed with high quality content (e.g. publications) results are decent and mostly on the positive side of time savings/value. Try to ask it a simple question, such as "pros and cons of these two smartphones" and it will suck up marketing bullshit from the internet and give you a convincing, but useless answer. I was really surprised to find this out, because I've been mostly living in the "scientific research" bubble, where it performs good to excellent and really saves time.
That really depends on the domain. In fields like manufacturing, resources extraction, and pharmaceuticals most of the valuable information is locked behind corporate firewalls.
Why do you think that? I collect books, mainly published from 1850s to 1980s, and it's not uncommon that they only occur in library listings and sometimes Goodreads when I search for information about their authors and history. This holds for some english language literature as well, but more so for books in other languages.
The Internet is basically a young, small portion of writings in english, as far as I can tell.
The internet created vast new opportunities for software engineers to create value.
Generative AI is about replacing software engineers in creating that value.
These are very different things.
Opportunity: you can buy $1500 in shingles, put them on someone's roof yourself in 4 days, and clear $5000 profit -- somewhere where rents/mortgage are under $1500 a month and a nice restaurant meal for two is under $50.
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.
You can replace software engineers with anything else here. Most VPs and higher leadership I see at FAANGs (who're arguably the ones recognizing opportunity) come from a PM or other non-software background. As much as I'd like to believe we're special, it doesn't seem like we are any special. We have no protection.
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.
Except now orders more magnitude of internet connected high knowledge third worlders are not only chomping at the bit for these jobs but qualified for them, and remote work is now largely accepted.
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.
Author of the article - and analyzed the likely impact of Section 174 in detail a year back [1], in Jan 2024, when it became clear that it was not being reversed like most assumed it would be.
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!
In all of my software career I have only worked for a company that actually produced software twice. One was as a contractor to Intuit and the other was Bank of America. Absolutely everybody else had these massive fleets of developers who struggled to put text on screen from a high centralized database to a web browser.
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.
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.
This happened to me quite recently. I didn't spot the hallucination, trusted the LLM output and introduced a problem into the application I'd never introduce if I didn't use LLMs in the first place. Then my co-workers did the same thing because the LLM guided them in the same broken way as it guided me.
That's a big, massive if. I think the more experienced you are, the less net productivity you gain. If you're more senior, is probably very low or even negative. I use ChatGPT every day pretty much, but it has wasted quite a bit of my time.
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 think experiences are highly variable based on the person and stack. I'd say I've gotten a ~50% productivity boost. I'm quite senior, make FAANG money.
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.
I disagree. The more senior you are, the easier it is to generate something and be convinced that "ok, this is good enough" or "aight, I see these things that need to be changed instantly". Frankly, compared to others, I don't have that much experience (about 10+ish years), but it saves quite a lot of time for me.
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.
>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.
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.
If there is a finite amount of software to be written in the world, and an increase in productivity of developers than that amount of software is done with less people. Also, if AI allows an non-software person say an marketing person, write a python script to query a few databases. Than the 10 person business data science team might not need a body as they no longer have to deal with the 10% lower level stuff.
I think if there is a finite amount (debatable), we're still nowhere near close to reaching that outer limit. An increase in developer productivity could mean the team does more with the same people, or builds more product that requires hiring more engineers to support.
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.
The term "hallucinations" has always frustrated me. The marketing there makes sense, but an LLM that hallucinates is an LLM doing exactly what it was designed for -predicting what a human might say in response.
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).
You'd think to sustain these ridiculous PE ratios, tech companies would be investing in product development more than ever before. Now is not a time to do less work.
Funny because I don't remember a downturn in software engineering jobs pre-COVID. My recollection is that it was easy street for STEM types until the inflation hit.
The graph is very interesting! The number of Indeed openings looks like it's basically leveled off at the initial early covid number over the last year; it's not rising, but at least it's not falling anymore. That seems like pretty optimistic news for developers! Total compensation is still pretty high, I hear. And it seems likely that if humans aren't replaced wholesale by AGI, the opportunities for building valuable new software are likely to increase a lot over the next few years.
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.
Are we still going through backlog at same rate but with fewer engineers with AI? Can we expect an "induced demand" for more nice to have features which require more engineers?
Assuming that AI coding is already helping the average dev with shipping more.
I don't think this means what they think it means: a _lot_ of job postings get filled before they get counted. Think about it, there have been a lot of layoffs. Many of these people have friends that can refer them to jobs before they are posted.
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...
So mostly, because a lot of companies are 'wait and see'?
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.
Look, this is harsh, but I’ve talked to a lot of people who run dev shops. None of them are hiring Juniors. And most of the Seniors they’re hiring are in Eastern Europe or Central & South America.
With Cursor and Copilot, the days of $190k entry-level SWE roles in the US are GONE.
Yet there is no let up in diluting software engineering labor pool with thousands of H1Bs amidst the already distressed labor market from American job seeker perspective.
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.
“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?”
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 believe the American public’s appetite to disincentivize and tax the American companies that ship jobs overseas while simultaneously benefiting from the protection, infrastructure and US government contracts directly or indirectly will only increase in the coming years.
Easy to tax physical goods, hard to tax "underpaid" service work.
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.
Sounds like maybe you’re from Europe? And have a frustration with foreign companies in the US. This further demonstrates that globalism as we’ve seen it with “free” trade is largely coming to a new era with strong headwinds, and it’s not just the US citizenry that wants that. The people that are pro H1B in the US, even given that more than half of the positions hired are foreign-born, are ignoring a stark reality that this will not likely continue. It’s gonna be a rocky road.
That description is painting me with primary colours, but yes, I'm from Europe. Chose to stay in the EU, post-Brexit, rather than stay in the UK or to move to the US.
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.
Everyone wants to say it will get better based on what happened in 2000 and 2008. I was around both times.
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.
I’ve been receiving emails from reps at companies lately. Some are from companies that didn’t hire me after getting laid off in 2022 for petty/unclear reasons. I had written some of them off as fake job postings. Now between that experience, this supposed decline, and ghost job postings I’m not sure if it’s a serious email or not.
"Programming languages are simpler than human languages"
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.
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?
Those occur when you've made an unambiguous statement that has no valid semantics. ("God should get a promotion."†) They don't occur when you've made an ambiguous statement ("We saw her duck.") As I noted above, it isn't possible to make an ambiguous statement.
† 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 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.
reply