Did you know that Wednesday Wisdom is also a podcast! Find it on Apple Podcasts or on Spotify. Did you know that there is also a custom GPT called Midweek Muse that has access to all of Wednesday Wisdom? Did you know that all Wednesday Wisdom videos are also available on YouTube?
Completely unrelated, over in our sister publication Thursday Thoughts, there is a new article about how to customize your own ChatGPT to act as a (mock) Designated Pilot Examiner during the oral part of the pilot checkride.
Two weeks ago, I discussed how the history of software engineering desperately sought to make programming easier, culminating in the latest and greatest invention in the field: AI coding tools. Unfortunately, as was discussed last week, coding with AI tools is still hard. So what does all of this mean for software engineering as a profession? Will there still be software engineers or will people build their own $100k SaaS without knowing how to code in five easy steps?
Software engineering is a tight loop that encompasses thinking, researching, writing documents, reviewing documents, meetings, coding, reviewing code, testing, debugging, and manual operations. 99% of the things I do as a software engineer fall into one of these buckets. AI tools are hugely helpful during most of these activities, significantly cutting the time required. Really, the only things it does not greatly help with are thinking and writing good and original documents (the only ones I write, of course). In other words, AI tools make me more efficient and effective by shrinking the time required for most parts of my job. The question now is: What are we going to do with the time saved?
One potential answer is: “Nothing”; if three people can do now what we needed five people for until recently, we are just going to lay off two people and keep output constant. This is surely what many people are thinking is going to happen and some CEOs are adding to this by pointing to AI as a main driver for recent layoffs.
Another potential answer is: “We are just going to write more software!”
Every organization I know has a stack rank of things that they would like to do and a “red line” that marks the most important thing that they are not doing. If AI tools make the existing software engineers more productive, the red line could move and the things that we just did not have the people for might now suddenly become possible. For instance, I know for a fact that the government is sitting on a pile of legacy software that they would love to rewrite in Java and maybe now they can…
So which of the two is it going to be: The same amount of software with fewer people or more software with the same amount of people?
The answer, of course, is: But not at the same time and not at the same place.
“Wait, what now?”, I can hear you thinking, “how is that even possible?”. And what will the impact of that be on the price of software engineering (in other words: On software engineers’ salaries)? To answer that question we first need to look at some other trends in the market because, apart from the current US administration, nothing lives in a vacuum.
One of the things that often amazes me, is how many great software engineers there are these days. I regularly come across extremely well written articles or great pieces of software, authored by people I have literally never heard of before. When I started studying computer science back in 1984 there were not very many of us and when I joined the Dutch Unix User Group (NLUUG), I could say with some confidence that I knew almost everyone in the country who ran Unix computers. Since then, millions of people have taken up computer science and, brains being distributed the way they are, quite a lot of them are either brilliant or at least very, very smart. Apparently, there are now about 47 million software engineers in the world, which represents approximately a doubling since 2018. Consider that for a second, in the last eight years, the number of software engineers has doubled. What have all these people been doing?
Throughout the history of software engineering, we have always had a shortage of software engineers and shortages typically means rising prices (hello war on Iran). To the joy of all people in the field, software engineering (until very recently) offered the highest overall earning potential and career ceiling, out-earning traditional engineering fields like civil or mechanical engineering. In fact, looking at the job market as a whole, only top medical specialists, finance executives, and top legal and business professionals earn more than the highest paid software engineers.
High prices for a good or service typically means that you consume less of it, though how much less is influenced by a concept called “price elasticity”. All major business endeavours these days need to be supported by software, so every company and every startup that is trying to carve out a unique market has had to hire software engineers. Given the shortage of qualified software engineers, prices have been going up and companies had to bid against each other to attract talent.
Short history lesson: In the Netherlands, a lot of sectors of the economy are ruled by collective bargaining agreements that are negotiated between the unions of employers and employees. If the government thinks that a recently negotiated agreement is a good one (they usually do), the government minister for labor will declare these agreements binding upon the whole sector; even for employers and employees who are not members of the unions that negotiated the agreement. One of the things contained in most agreements is pay scales, which are typically drawn up so that people with approximately the same education and experience, earn about the same.
My first job was at a bank and my labor contract was governed by one of these collective bargaining agreements. Banks had problems hiring IT talent because they couldn’t pay software engineers more than accountants with a similar level of degree and experience. To address this, the bank employers had negotiated a provision into the agreement that they could pay a “labor market surcharge” for certain specialties because otherwise they would never be able to hire a software engineer ever again.
For software and Internet companies, the availability of software engineers and related IT talent was so crucial that a real “battle for talent” ensued. When I got hired by Google in 2006, it was made clear to me that my most crucial contribution to the company was hiring other people. Consequently, we hired everyone we could find (provided that they jumped over the hiring bar), wherever we could find them. I was working for Google in Zürich, which is lovely and amazing, but it was made clear to me that the only reason that that office existed was to function as a hiring bridgehead in Europe, hiring talent that could not or would not move to the US.
As a manager, I regularly got spreadsheets of “H-1b losers”, people who we hired and who wanted to move to the US, but who didn’t get an H-1b visa (because the number is limited and there are lotteries involved). We would offer these people a position in Zürich instead and tell them they could move to the US after a year when they would qualify for an L-1 visa (intra-company transfer).
Some of those did, but some came to their senses and stayed in Zürich.
We would hire people without a headcount plan or without a vision of what these people were going to work on. When they accepted the offer, their resumes and interview packets would go to an allocation meeting where managers competed to select people for their teams. At Facebook, new people started in a “bootcamp” to learn about Facebook’s internal technology and development environment. Part of that bootcamp was “team selection”, where managers would present their teams and try to woo the scarce talent to their teams. News abounded about big tech’s high salaries and it is not strange that many young people decided to study computer science, leading to the aforementioned doubling of the number of software engineers from 2018 to 2026.
Obviously, these good times had to stop rolling at some point. A lot of tech companies overhired enormously and branched out in endeavors that did not make a lot of business sense. Non-big tech companies had to pay inflated salaries to hire “the best of the rest”. Lots of companies couldn’t hire at all and had to make do. A reckoning had to come and guess what, it did! Starting in 2022 and picking up steam in 2023, tech companies started laying off people by the thousands, flooding markets with capable software engineers who had mostly become accustomed to cushy jobs with great compensation. In my “home market” of Zürich, the layoffs hit really hard because, Switzerland’s awesomeness notwithstanding, in terms of good tech employment, Google was pretty much the only show in town and tenure at Google Zürich had been stellar because of the simple fact that there was nowhere else to go.
That said, I am happy to report that since then, and probably fueled by the availability of tech talent, Zürich now has a small but thriving and (as far as I can see high quality) startup scene.
Economics prescribe that if supply grows relative to demand, prices will drop and this is exactly what has been happening. Some of that has been overt (just offering lower salaries) and some of that has been covert (in the form of downleveling in the offer). This price effect has not been so drastic that polite society crumbled, but the downward trend is easy to spot. People moved, people moved on, people retired early, and people took jobs with a lower salary.
It is in this market that AI tools made their entry…
Suddenly, we have a technology available that makes software engineers more effective, at a time there seem to be more than enough software engineers on the market already. This makes many people rightfully very nervous, because quite a lot of people who have been in the market since about 2018 have racked up massive student debt that is not paid off yet, because they haven’t been riding the gravy train long enough. Life is even worse for the young people studying computer science right now because they are graduating into a market that is going through a huge reset, where there is a lot of competition and a brand new toolset that will have an as-yet-unknown impact (though we can probably say with some confidence that it won’t be positive for the number of software engineers required).
So, looking at all of that, what is the future of software engineering?
First of all, AI coding tools will not make software engineers unnecessary. As I argued last week, coding with AI tools is still hard and so we will continue to need people who are experts at solving problems with computers. But, and here is the crux, most of these people will probably be of the more “senior” variety who have a good understanding of how to build complicated working software with AI coding tools. There will probably be a lower demand for more junior people because a lot of what they used to do can now be done efficiently by senior engineers with the help of AI coding tools. In fact, my recruiting friends tell me that this is already happening; the job market for senior people with AI expertise (to the extent that there is some to go around already) is quite healthy. Not even the AI frontier labs believe that AI coding tools will completely replace software engineers, as evidenced by the fact that both Anthropic and OpenAI (my own employer) are still hiring software engineers. But, we are not hiring new grads or people with limited industry experience, which might give you an idea of where we think the industry is going.
This of course poses the question where new senior engineers will grow if fewer companies are hiring junior engineers. Great question, to which I don’t have an answer yet. For now I assume that everyone assumes that is someone else’s problem 🙂
By the way, it is not that these junior engineers will do nothing. A lot of them already have jobs where they will get experience and eventually become more senior. One thing that the AI coding tool boom might fix is level inflation where new grads got hired as “senior engineers” and people with five years of experience “demanded” promotions to staff level. Compare this to the construction world where you typically need 10-15 years of experience to become “senior”.
What helps the software engineers is that AI itself represents an “expansion of the disc” (see last week’s article for an explanation of that concept), meaning that AI will in itself cause more software to be written because really, all these artificially intelligent customer service representatives are not going to write themselves!
All of these things are happening at the same time: Some companies are still letting people go, some companies are still hiring (but potentially hiring fewer and different people), some companies can suddenly afford to hire people because there is ample supply, some people will leave the industry and the bottom, where the people reside who only had a job because we were in a situation where everybody with a pulse could one, will fall out and they will have to find something else to do.
I know this sounds terrible and harsh, but it is unfortunately the reality.
One thing we can learn from Wall Street is that, regardless of whether the market goes up or down, there is always money to be made 🙂. These are times of great change and it might easily be quite a few years before the situation stabilizes again. For the people who worry a lot I have got this to say: This is the fourth crisis of sorts that I live through and every time people told me that now things were going to be different. They were right, things always turned out to become different (hint: they always do), but over time I have always been able to ride whatever wave there was to be ridden. Not because I am brilliant, but quite simply because I have continuously invested in my knowledge and skills. If you are a software engineer today and you are not playing around with running your own models and writing your own agents, you are doing yourself an enormous disservice.
What I think is really different this time is that the times where there was a great shortage of software engineers might be behind us. Software engineering is difficult and so I think there is always decent money to be made, but the times where a new grad could get $200k per year and a nice equity package might be behind us. Top engineering students will probably still do very well, just like top law students and top medical students (post-residency), but a computer science degree from <insert your favorite party school here> might not cut it anymore.
In itself I do not find this problematic. The great shortage of software engineers from 2010-2020 and related price increases were historically an anomaly. When I was hired by Google in Zürich in 2006, with 18 years of experience already, I was offered $175k (in 2026 money, give or take, inflation and USD/CHF exchange rates being what they are) plus a modest equity package. Levels.fyi tells me that that is what a new grad SWE gets these days in Zürich. Nothing wrong with that, but it does show a significant price increase and I think we are going to see these prices will stabilize (at best) or maybe even drop a bit.
Then finally I have to say something about the overexcited news reports that non-coders will now use AI coding tools to develop their own applications? Won’t that cut into the need for software engineers? Will experienced software engineers not simply be replaced by business majors with Codex?
Answer: No.
It is indeed true that with today’s tools, you can create quite decent simple applications without knowing a lot about coding. As an example, I “coded” “Chordcraft” (an application for showing how to make certain chords on the piano) using Codex and without any technical intervention. But really, I think that this is about as far as it will go, because anything more complicated needs an understanding of how computers work. It’s a bit like arguing that the availability of “Technic Lego” will decrease the need for mechanical engineers building trucks and planes. I think that most applications that non-engineers (will) create are ones that people would have simply done without if AI coding tools weren’t there. If it will have a measurable impact at all, it will be small and at the bottom of the employment pyramid (a place where I think not many Wednesday Wisdom readers have their professional abode).
That’s it for this herniated three-episode exposé on AI and software engineering. I promise that next week I will return to lighter subjects, such as that we are all mushrooms: Kept in the dark and fed a lot of bullshit.
