The Case Against an Imminent Software Developer Apocalypse (zdnet.com) 54
ZipNada shares a report from ZDNet: Given the dour headlines as of late concerning the diminishing amounts of entry-level software development jobs, coupled with predictions of applications entirely AI-generated, one could be forgiven for assuming that software developers may soon be an endangered species. However, the data tells a different story. James Bessen, professor at Boston University, has been pushing back for some time against the talk of AI and automation displacing jobs on a mass scale, and lately has been arguing that the roles of software developers are nowhere near extinction.
AI is certainly not killing the software developer, Bessen said in a recent analysis (PDF). AI is taking over software development tasks and boosting productivity and output, but that is not translating into lost jobs, he argued. Instead, the types of software skills sought by companies are changing. "Surprisingly, however, after three years of AI use, software developer jobs have continued to grow robustly, reaching record levels of employment -- 2.5 million in February," Bessen said in the report, citing data from the US Bureau of Labor Statistics. The number of software developers in the US has grown by over 400,000, or 19%, since ChatGPT was introduced in 2022. At that time, the employed software developer population was just under 2.1 million. [...]
The productivity uptick developers are seeing may ultimately be a boost to their professional opportunities, however. "An important and possibly disruptive change is happening, but the common view misunderstands what is going on," Bessen pointed out in his report. "Careful case studies find that AI improves the productivity of software developers -- that is, the software produced per developer -- by 30%, 50%, or more. And the rate of productivity improvement in software development is improving." Tellingly, since 2022, when ChatGPT was introduced, developer productivity has increased noticeably, Bessen continued. "From 2003 to 2022, developer productivity grew at 3.9% per year; but from 2022 through 2025, it grew at 6% per year." [...] A coming flood of new software products, now more likely to be enhanced by AI, will continue to create jobs for developers, Bessen predicted. "Thus, mass unemployment of software developers seems unlikely to happen soon." This doesn't mean the job descriptions of developers or other computer occupations will remain static. AI is shifting and re-inventing these roles, Bessen added.
AI is certainly not killing the software developer, Bessen said in a recent analysis (PDF). AI is taking over software development tasks and boosting productivity and output, but that is not translating into lost jobs, he argued. Instead, the types of software skills sought by companies are changing. "Surprisingly, however, after three years of AI use, software developer jobs have continued to grow robustly, reaching record levels of employment -- 2.5 million in February," Bessen said in the report, citing data from the US Bureau of Labor Statistics. The number of software developers in the US has grown by over 400,000, or 19%, since ChatGPT was introduced in 2022. At that time, the employed software developer population was just under 2.1 million. [...]
The productivity uptick developers are seeing may ultimately be a boost to their professional opportunities, however. "An important and possibly disruptive change is happening, but the common view misunderstands what is going on," Bessen pointed out in his report. "Careful case studies find that AI improves the productivity of software developers -- that is, the software produced per developer -- by 30%, 50%, or more. And the rate of productivity improvement in software development is improving." Tellingly, since 2022, when ChatGPT was introduced, developer productivity has increased noticeably, Bessen continued. "From 2003 to 2022, developer productivity grew at 3.9% per year; but from 2022 through 2025, it grew at 6% per year." [...] A coming flood of new software products, now more likely to be enhanced by AI, will continue to create jobs for developers, Bessen predicted. "Thus, mass unemployment of software developers seems unlikely to happen soon." This doesn't mean the job descriptions of developers or other computer occupations will remain static. AI is shifting and re-inventing these roles, Bessen added.
Efficiency Boost (Score:5, Interesting)
This has had a great effect on my workflow. I am still able to do high-level architectural planning, determine use-cases and usability parameters, etc. When I have those pieces figured out, I can use an LLM (in this instance Claude Sonnet or Opus 4.6) to do the actual generation of code, which I can then review and correct as I see fit. I have not (and will never) used an LLM as a replacement for my "higher brain functioning", but when it comes to the "code-monkey" aspects of my work, it does them far faster than I can (and typically with a healthy respect for naming conventions, code patterns, etc.). I still get to enjoy the fun critical-thinking-laden aspects of my job, but the simple "regurgitation of learned code words" is offloaded to an all-too-willing counterpart.
Re:Efficiency Boost (Score:5, Interesting)
Same.
When AI was still young, I tried it and found the code quality to be unacceptable. I was at that time in the "it's a bubble that will pop" crowd.
It's better now. I use it more now. And it saves me time and makes me more productive.
It can't do my job without me. And other people on the team still come to me for help. My skills as a designer and knowledge of our legacy system still make me valuable. I can just do more in less time now.
The other consequences of AI (impact on electricity cost, pollution, etc.) are problematic. So are the legal issues with mass copyright infringement in the training data. That all needs to be properly hashed out. Probably the end will be the same: the super rich get richer and everyone else gets table scraps. That's just humans at work. But, apart from all that, AI is good.
I am officially in the pro-AI camp now.
Re: (Score:1)
Re:Efficiency Boost (Score:5, Insightful)
The amount of technical work a business needs to accomplish over time is not fixed. For a healthy business, there are always lots of things they would like to develop but can't due to limits in capacity. So, with this productivity boost, they can get even more features out the door and (hopefully) make even more money.
Why would they cut staff and keep a tiny throughput with a smaller profit margin? Their competitors, who capitalize on AI-assisted efficiency boosts, will eat them for lunch. And then hire the staff they laid off.
Re: (Score:2)
For a healthy business, there are always lots of things they would like to develop but can't due to limits in capacity.
AI rewards strict APIs. [dri.es]
Re: (Score:2)
For a healthy business, there are always lots of things they would like to develop but can't due to limits in capacity.
This sounds nice in theory, but for quite some time now, that hasn't panned out the way it seems like it should.
Let's use a great go-to example - the finance department. Back in the 1970's, it was mostly-manual. You might see a calculator in the back room, but the ledgers were written by hand, the credit card slips all came from a knucklebuster, and lots of people had full-time jobs doing calculations and data entry and inventory management.
*all of that* is automated away now. Scan a barcode, shipping manif
Re: (Score:2)
For a healthy business, there are always lots of things they would like to develop but can't due to limits in capacity.
This sounds nice in theory... Let's use a great go-to example... bookkeeping
There are few businesses whose development relies specifically on advances in bookkeeping. You still have to do it, but the less spent on it the better. The state of the art of bookkeeping is generally not a limit to the capacity of business development.
For some companies, this is true of software as well, but obviously there are many companies whose business development relies significantly on advances in technology. I'm afraid your example just doesn't apply to GP's point.
You'd be hard-pressed to grab a hundred people at random, have them think of the software they use regularly...and point to a time in the past decade where their software got an update and they were HAPPY
I don't know about a hundred p
Re: (Score:2, Interesting)
The key word here is "architecture". This is something most developers don't have the comprehension desire or ability to address. They just like writing code and understanding new libraries. It's also why most developers won't be successful vibe coders - you'll have devops folks and traditional systems people filling the role more capably.
The 'code monkey' function is more or less dead.
I am concerned for the future, though, as it's becoming increasingly possible to outsource thinking and architecture with A
Its just yet another tool writing boilerplate code (Score:1)
My own experience with leveraging LLMs has been one of an efficiency boost. As I have around 15 years of software architecture and development experience, I have yet to come across an instance where an AI is used to do something I can't do, and instead is used to do something that I could have otherwise done myself, albeit much more slowly.
It's a tool. It can help you do some of the "boilerplate" type coding. We have precedent, for example the graphical UI layout tools that arrived in the 90s. They allowed tools to generate large blocks of code for developers. AI assistants are doing something similar. They can be trusted (but some verification needed) to generate code for things that are well discussed in textbooks, academic journals, and other professions references. And they are often OK if a problem can be solved by gluing a few of such w
Re: (Score:2)
But for more complicated problems, they are pretty unreliable.
I agree, but I often wonder how many developers vastly overstate the complicated-ness of what they need to write.
Re: (Score:3)
AI is actually significantly better than a boilerplate code generator. It can create end-to-end code enhancements that actually work, and are well-written. It's also quite good a debugging, such as diagnosing obscure exceptions that are thrown. While I agree that AI isn't going to kill off programming, it's not correct to say that it's "just" biolerplate.
Re: (Score:1)
With respect to debugging, I'll repeat my previous points. It's an interaction on existing tools. For example it's an easier to use lint, and certainly can check for more things than lint. But fixing code. Except for simpler bugs that are more straighforwaore, I think again we hit a tru
Re: (Score:3)
Let me give you an example.
I was using Visual Studio / C# / SQL Server to import several files into a database, comprising several million records. All at once, the import threw an exception "Exception converting a string to an integer." There was no more information in the exception than that. I clicked the Copilot exception helper. It proceeded to inspect who knows how many thousands of watch variables, digested a bunch of related code files, found the actual files being imported and analyzed them, and fi
Re: (Score:1)
I've used AI agents for quite routine things too, and they are convenient. My point is that we're mostly seeing an evolution from existing toolsets. We're losing some of the grunt work. There is still plenty for humans to focus on.
Re: (Score:2)
Yes, Visual Studio knew exactly where in the code the exception occurred. But because it came from SQL Server, it didn't know which row of the imported data set caused the exception. The operation was simply a bulk insert of 2.3 million records, as far as the code was concerned. The issue occurred previously, in the compilation of the data, not in the spot where the exception occurred. I see you've never run into one of those obscure SQL errors, or you would recognize this pattern.
I think you haven't gotten
Re: (Score:1)
I think you haven't gotten to the point where you have seen AI do some "more than mundane" things. It's there.
I have, I just haven't seen it reliably do "more than mundane" things. In particular debugging often being a one step forward, one step back, process. Fixes the stated bug, creates new bug(s). Debugging ends up being an iterative process fixing the original bug and the new bugs created. Well some of them, not all are initially recognized.
I'm referring more to trust than capabilities.
Re: (Score:2)
It's true. When AI gets confused, it _really_ gets confused, and can't find its way out of the hole it created. On numerous occasions I've had to revert all the changes and start over, step by step, because AI couldn't figure out how to resolve the issues.
Re: (Score:2, Informative)
There are three kinds of people in this AI revolution.
1. The hyperventilators.
2. The nay-sayers.
3. The ones who actually use it like the power tool that it is.
Thus ... (Score:3)
Why not? All they do is to re-write perfectly good web applications into Android/iOS apps. And then cry pitifully when I just keep using the browser-based one.
Or over at Microsoft, all they seem to do is to find different places to hide their menus.
It's a time saver (Score:4, Interesting)
It's great to let AI debug something tedious, or write a script to make my day more protective, or bounce ideas off someone while I'm working at home.
AI is a game changer in our profession. I believe it is like the switch from using pencils and rulers for drafting to using AutoCAD. The drawings still had to be produced, but the people doing the work were able to get more done in the same amount of time.
Re: (Score:3)
It's fantastic for iterative testing and code observations.
It's great for 'bulking' up development plans, and saves a huge amount of time here. I can read many times faster than I can type, and I can type many times faster than I can type + convert ideas in my head into coherent English or code.
It's also substantially better than most developers at writing code: it mimics working projects, not your average enterprise "just get it done" project. Heck, depending on the model and the instruction, it produces b
Coding, Development (Score:3)
Re: (Score:2)
Well, yes and no. Agreed, AI can't develop. But it is way better than just coding unit tests. Maybe your experience is a few months old, I'm not sure. But today's AI coding tools, like GitHub Copilot, can handle end-to-end implementation, getting it right most of the time. It's also true that when AI gets stumped, it can get itself into a hold that it can't get out of. But most of the time, it's quite, quite capable.
Business Rules (Score:2)
Re: (Score:2)
Do you mean to say that those same Judges and Lawyers that now use AI to get their work done are needed to help us use AI to create our business applications? Something seems a bit off with that argument.
I strongly feel that red is better than blue. (Score:4, Insightful)
I've skimmed the article, and the article the article references. And this is meaningless, because "productivity" is not being measured in any rigorous way. It's all just vibes.
"Writing code is faster." What does that mean? Are you saying that more lines of code is more productive? (At this point, you sure as shit better not be, but.) Does it mean that the LLM can produce "good code" faster? How do you measure the quality of that code? (You probably aren't even bothering to.) Do you have a developer eyeballing the output code? (Liar.) Are you feeding the output into another LLM to test its quality? (You must have quite the token budget.) Are you counting bugs that crop up later, and strictly accounting for time it takes to fix them? (LIAR.)
Re: (Score:2)
Re: (Score:2)
I think the opposite is happening. Writing boilerplate for whatever is the current javascript framework de jure has contributed more to skill degradation than AI ever will.
Re: (Score:3)
Developer productivity is notoriously difficult to measure rigorously, and your list of concerns touch on some of the reasons why.
Sloppy measurements are the only ones available, for the most part.
There will be a subjective component to the assessments being made here. There is no escaping that. That doesn't mean that the conclusion is automatically false. You certainly have the option to refuse to adapt to a changing landscape while calling everyone else liars and/or idiots. At this point, I consider t
Manipulated bullshit (Score:2)
The 'official numbers' are manipulated bullshit. There aren't more software jobs, what there are is an ever increasing H1B replacement of domestic workers (in the US). They'll fire 6,000 US workers and replace them with 9,000 H1B. That's still a net job loss, even if there are 3k more workers.
As for the productivity of existing "skilled" developers, "that depends".
You absolutely can, with agentic frameworks with proper planning and orchestration produce far more good code, with more features. This is how th
Re: (Score:1)
Re: (Score:3)
I created an Indeed employer account and found that there are 18 candidates for every position available for a role I'm seeking. The majority (I'd wager 70%) all have 20+ years of experience.
It's shocking. I can speculate about the why but also... why bother.
Some Truth - Anecdotal evidence (Score:3)
I am using Claude.code to develop my novel cybersecurity architecture after multiple senior developers (25+ years) were unable or unwilling to build what I described. I kept getting "what they thought I meant to ask for" rather than what I actually specified.
Sadly, because I lack the coding background, it took me a long time to figure out that they were not following instructions. I burned through a lot of budget for not much results before I fired the developers.
Now, using AI, I am building what I spent years describing. I can code some but I can't write hundreds of thousands of lines of Rust and ever hope it works. In the end, I don't expect the AI results to be production ready. Instead, I am creating a proof of concept that will need to be optimized, improved or even rewritten by "real" developers. However, once I have a working system I can then point to it and say "Build me that but better."
AI has cost 2 senior developers their jobs by allowing me to actually create what I documented.
Re:Some Truth - Anecdotal evidence (Score:5, Insightful)
You have a "novel cybersecurity architecture", but at the same time "lack the coding background"? How does that work? Oh, wait, it does not.
Re: (Score:3)
There are some bosses at my company like you. They think the dev team is dragging their feet and building crap, when the real problem is that the executives refuse to listen to their people. You go right ahead and build your "novel cybersecurity architecture." Those 25+ year devs will break into it in a heartbeat, because they actually know the pitfalls and gotchas of real cybersecurity.
what's it all for anyway? (Score:2)
Look, we programmers work for corporations who make money selling access to the software we write. Billions of money. Smart companies know that more code faster doesn't mean software they sell is working. If software stops working a lot, customers go somewhere else. Nobody has a monopoly on software, but some companies have built the trust of customers or business partners in such a way that customers know they're getting the best the industry has to offer. LLMs can do the 80 part of the 80/20 split. You c
unions and an trades system can really help! (Score:1)
unions and an trades system can really help!
It lets me be a dogshit jack of all trades.... (Score:5, Interesting)
However, in contrast, there are many UI-related skills I don't have and AI makes it a lot faster for me to get ramped up. So it will help me write mediocre code I am clueless about, but for the stuff that I know really well (routine server-side business code), it produces dogshit mostly.
So put cynically, Claude 4.7 allows me to be a dogshit developer of all trades. Before, I'd just tell you "I don't know Angular well enough to do that...you should send that ticket to my teammate who is an expert on that." Now I can write dogshit code that might even work...in any language or technology of my choosing!!!!!
AND as a bonus, our total lines of code is astronomically expanding and quality rapidly going down because I have a few teammates who are much sloppier than I am about checking the output...and a few have stopped putting me on pull requests because I find many mistakes and tell them to correct them.
Re: (Score:1)
and updated all the unit tests to ensure the build passed.
Hey, these AI's learned to CYA - they're smarter than I thought!
Re: (Score:2)
AND as a bonus, our total lines of code is astronomically expanding and quality rapidly going down
If the quality of the codebase goes down and the number of incompetent programmers that need to be hired increases, that IS a good thing from a manager's perspective.
Managers are judged based on how many people they manage, not on the competency of those individuals, nor on the total salary of those people.
Principle of Endless Wish-List (Score:1)
Generally if more features (code) can be purchased for the same dollar, companies will add more features. They always want more shit, they just don't want to pay for it.
I'm also sure there will be a lot of cleanup work from the AI-slop that tends to creep in.
Braindead article (Score:3)
This braindead article is very out of touch with what is happening on the ground. Go and actually talk to developers, or look at Department of Labor statistics.
"AI is just making developers more efficient" - Yes, which means, we need less of them.
Speaking as the position of someone in charge of an engineering team - I don't plan on ever hiring another junior engineer, ever again. I only have jobs for senior engineers now. What does this mean for the industry? It is going to be dire - however my company is not a charity, I need to make decisions based on what is best for the company - and right now, that means hiring hardly anyone and those that I do hire are only seniors.
Re: (Score:2)
If every company had your idea to never hire junior devs again, then in 10 or 20 years there will be no senior devs. Then companies will be left with mountains of AI generated software that nobody has the skills to work on any more.
I conclude that you are planning for your company to only be around for a few years. I hope your share holders are fully informed of your plans for the future of their money.
Re: Braindead article (Score:2)
I agree 10000% with what you are saying
But again, my company is not a charity
I can't afford to hire junior developers and train them for funsies. That's what you are suggesting.
What does that mean 5-10 years down the road? We are all fucked. But again, I can't solve that for society. That is SUPPOSED to be the job of government, but they are all sitting on their hands eve though the industry has been screaming at them for 3 years now to start plannibg and do something
I think it depends on Managers. (Score:2)
AI is a power tool (Score:3)
Construction workers used to use hammers and hand saws. Now they ALL use power tools, and we still can't find enough construction workers.
AI is a power tool. I predict we will soon ALL be using AI, and we still won't be able to find enough competent programmers.
Translation (Score:2)
Did I get that about right?
Remember billionaires and CEOs see you as an input in their spreadsheets. You are not a human being and you have no agency of your ow
I have _never_ had colleagues ... (Score:3)
... this competent as I do now with AI. It's like a team of premium-grade senior API and software lib experts is just sitting by in a live IRC chat with full view of the current state of the project, ready and eager to discuss every detail of software decision we make as fast and efficient as possible. I have never had this level of insightful and helpful discussions of code with any human. Mostly because that would've costed hundreds of thousands if not millions in anual salary for the same effect achieved with humans. Here an example exchange from a few weeks back (this actually happened for real):
----
ME:>>>
> Could you explain the reasoning behind implementing a Baseline service? Is this a good practice and if so, why?
AI (Codex IIRC):>>>
> Short version: Yes, itâ(TM)s good practice here because you need the same Baseline metadata (file list + meta.name/type) in multiple places (Baseline routing + Mainview). A service is the cleanest way to share that data without duplicating fetch logic or tightly coupling components.
Why it makes sense:
â Single source of truth: one place to load and cache the list of Baselines.
â Decoupling: Mainview and Baseline donâ(TM)t need to know how to fetch or parse the index.
â Less duplication: avoids copy/paste HTTP logic in every component.
â Easier to evolve: if the index format changes, you update one place.
â Easier to test: you can stub the service.
Downsides:
â Slightly more code and indirection.
â If the data is only ever used in one component, a service would be unnecessary.
Given your setup (routing + entry generation), the service is a good fit. If you want, I can keep it simpler by moving the logic into Mainview and passing data down, but that couples the components more tightly.
----
I have 5 of these exchanges at this level or higher every day. It's like there's 10 of me, well rested and in distraction free speed typing mode, and I'm a seasoned and experienced senior webdev who still loves his job.
I'm currently developing a brand new software project from scratch, complete with CI/CD and live operations on the side. Entirely on my own, 90%+ remote and I get to finish early every day. Something like this was unthinkable just 18 months ago.
What's interesting though is that this amplifies that digital cultural distance between me and ordinary non IT people even further, so I guess the article has a point. For now.
Re: (Score:2)
I've had a similar experience! It's like there's a dozen SWEs working shoulder to shoulder.
Stop hiring for the sake of hiring (Score:2)
If you have one or two good engineers (not programmers, not developers... honest to goodness engineers) who with AI can keep 15-20 tasks running simultaneously (the cost is GPUs and screens), you don't need to hire someone right now.
Companies want the best they can get. They don't want to fill seats anymore. If they already have
Jevons Paradox? (Score:2)
During the industrial revolution when steam engines were getting more efficient, a nowadays largely forgotten genius William Stanley Jevons [stanford.edu] predicted that instead of reducing coal consumption as most thought, demand for it would rise, coining an economic law called the Jevons Paradox. Time will tell if AI does for software developers what power consumption did for coal.
I only recently stumbled on Jevons who for some reason isn't nearly as well known today as his peers George Boole and Charles Babbage. It t
Fundamental change of attitude... (Score:2)
There has been a way fundamental change of attitude in software development places over the years, in my experience:
In the 1990s, things were being "paved". So, most software was being focused on breaking new ground. One project, we had to roll our own communications platform, using two rounds of DES for speed, store shared secrets because of RSA patents. It wasn't great, but we did things nobody did.
After the 2000s, there was a commercialization, especially making stuff "secure" due to Sarbanes-Oxley [1