AI Will Create More Developers, Not Less
interconnect.substack.com
interconnect.substack.com
Take a Laptop with WinXP+Office 98 from the 2000 era and travel back in time to the 60s or 70s using a time machine. Show the power of Excel to any of those developers punching holes into a punch card, feeding room-sized "computers". Tell them that this "Excel" tool is used by business and sales people without any programming skills. Those developers would clearly tell their children to learn other professions, as no one will need developers in the future because this "Excel" and "Access" make it so easy. But the opposite become true, we needed more devs than ever at that time. Our field will change, the way we develop software did change and will continue to do so - maybe using AI all day.
These new generative AI algorithms make things easier, but they produce worse results. E.g. you can put in 50% of the effort and get 80% of the result. By lowering the amount of knowledge required to build something, you open up the labor pool to more unskilled labor. So while technically the "number of developers" will increase, the ranks will be increased with unskilled labor and the profession will cease to be desirable knowledge work.
This is very similar to how craftsmen and tradespeople were replaced with automation. For example, "Luddites" are known as people who dislike technological progress, but they were really textile workers who opposed having their high-skill jobs replaced by low-skill laborers with the factory owners pocketing the difference between the high and low skill pay.
Excel is actually a good comparison. It's easy to learn, it works just fine for small stuff, it's easy to misuse, it's hard to catch mistakes, and more often than not at a certain point the grown-ups have to take over and rewrite the project using actually working technology.
> the ranks will be increased with unskilled labor and the profession will cease to be desirable knowledge work.
The real skill in software development has never been coding, but rather formally specifying problems and turning them into a complete set of instruction. If people could do that, we'd be all out of a job already. Java syntax isn't rocket science, folks...
There's nothing you can do in Python that you can't do in C. But while Python is easier to learn and use than C, it's less efficient, in some cases hundreds of times slower than C, let alone expertly-optimized assembly.
But the ease of application development in languages like Python has expanded the scope of what can be realistically done with software, causing developer employment and salaries to rise while lowering the required computer science knowledge.
As for the luddites, they lacked imagination. Low-skilled laborers might have been able to make textiles, but there were plenty of lucrative opportunities for talented people able to design, build and maintain machinery instead of merely operating it. It was not uncommon for talented engineers to become factory owners themselves, some of whom started as unskilled laborers (Samuel Slater, for example).
That's the problem with comparing this to previous tech advances. Past advances made people much more efficient at some task, but it still required the people to work the new thing. But now we have a new thing that might be able to work itself.
There's no time in human history that I'd rather live in than the present day.
While this is a great argument that there will be more software in the future (something that I don't think anyone disputes), it's not an argument in favor of the proposition that there will be more developers, because the whole point of AI-driven development is to automate what a developer does.
One reason there are a lot of businesses still using antiquated systems is that 1) the systems work for their purposes, so their need is not acute; and 2) custom software development is expensive. Your local gym can't afford to pay someone thousands of dollars for a custom app (or even hundreds of dollars for a customized version of an existing app). But AI-driven development promises an order of magnitude drop in cost. So yes, we should see a lot more digitization as it becomes available.
I do agree in the short term that it won't drive the number of developers down, since for the foreseeable future skilled people will be needed to oversee the AI product. But it will vastly increase productivity. Unfortunately, if history is any guide developers will not reap any of the rewards for the increased productivity, and their real wages may actually decline.
In my experience this isn’t such a problem. When technology makes it easier/cheaper to solve problems, the response is usually to try and solve more problems rather than spending less.
Your local gym might be able to afford a small app now, but I bet the big gyms start adding personalized nutrition/fitness routines into their products.
If we’re going by the history of software engineering, a more likely scenario is that developers start using generative AI to replace a lot of the people/management roles and capture even more of the profit. Of course, things aren’t the same as they were in the 70’s and 80’s, so maybe things won’t work out the same way, but I don’t think it’s a foregone conclusion.
One example: restaurant menus. Second example: car audio volume buttons/knobs.
The number of useful digital products that don’t yet exist is enormous. We can’t fathom the vast array of creations that future generations will dream up.
P.s. very strong agree about digital menus… so annoying to use!
Many software developers seem so accustomed to a disproportionately high labor value that they: a) incorrectly assume they can easily pivot their capability into something else, and b) have entirely lost touch with the realities of most labor markets. Even if its not tomorrow, I think this is going to be a rude awakening for many people in software.
The function approximation that AI offers is good enough to open software development to entire new aspects that have long resisted it. Even something seemingly so simple as order taking from a drive thru window has required algorithms that we have simply been unable to effectively create. We still haven't created them, but we can now approximate them.
These function approximations can now write creative stories and draw pictures of an astounding variety that would have simply been considered impossible even a decade ago. Soon that gym you mentioned will have 100 wide ranging automated app designs to pick from, or even a thousand automatically categorized by style. There will still be countless developers in the mix but they will be working on much more interesting tasks than writing a gym app that we currently allocate Human resources to.
Perhaps wages will decline, but the work will be a lot more interesting :)
They'll be using tools that utilize AI to build software. The AI will be writing the code. In the future there will be a lot more jobs watching and assisting and using robots in factories as well. The net count of factory workers will be lower than whatever the peak was (whether we're referring to the peak in eg the US or China).
This is how corporations will boost margins even further in the tech field. AI writing code will be considerably cheaper than what big tech has been paying (which isn't to say big tech will eliminate all of their code writing jobs, just a lot of them). Peak code writing employment has occurred in the US, the latest tech layoffs are the end of that era (and will be recognized as such in hindsight).
Software developer will typically no longer be understood to mean a person that writes code, the terms will change (and most likely that'll happen this decade). Only older developers will still think of it in those terms.
Oh, you mean, like, you write the code directly? Dude, what century are you from?
Anyway, I think advancement of technologies which generate text can make an individual developer’s task more efficient, e.g. the compiler. But, because of this efficiency on the individual, then more individuals will now be able to take part which means more jobs will be created.
To restate, I don’t think “AI” will make people more productive, it just makes doing the task more productive. This seems like a ridiculous statement, but consider the cotton-gin. Because of the tool, we could throw more and more people at the problem of gathering cotton.
That’s what the development of new tools gets us, more work with the tool that does the job.
Learning about it mid through that process I told her to try asking chatgpt to write a python script for it (she knows a bit of python, knows how to run a script etc.). Needed a pip install of the qr code lib, but other than that worked right away.
ChatGPT turned the next 4 hours of menial work into 10 minutes. This is the true incoming disruption.
People who just vaguely know how to program, getting a powerful tool that can assist them 95% of the way for simple automations like this.
Problem is how do you know if it's correct? If you have to verify it all it may take even longer. And when it's wrong it's back to square 1.
> People who just vaguely know how to program, getting a powerful tool that can assist them 95% of the way for simple automations like this.
Is assuming it's correct.
The notion that only a human can effectively check code, will be viewed as a joke before this decade is out. Humans will be very inferior. Emphasis on very. Anybody that doesn't realize that yet, is still thinking in the terms of today, right now - and not understanding how fast this is all about to change, much less how much it'll change in a decade of time.
There will be button pushing monkeys, that are called ... code confirmation engineers, that sit at a computer and sign off on the AI systems doing code checks. They'll go to school for it, they'll almost never directly review code, most of them won't be able to even if they cared to (which they won't).
This is a very absurdly small minority of people.
Basics of Python courses are very common everywhere, now.
AI can reduce the barriers between languages and cultural customs as well.
Languages, maybe, but cultural customs? I highly doubt that. Perhaps the perfunctory sort (e.g.: "vous"-ing and "usted"-ing), but there are real differences between cultures in both the work place and and personally that AI can't help you with. Nothing but compassion and a willingness to learn will help you with that one.
There really is no shortcut to compassion and understanding... so much of it is simply taking the time to inform yourself.
You can get an LLM to summarize a wiki page on some culture's customs, but there is better content written, edited, and curated by humans on the topic.
We're simply saying that AI can give you more cultural understanding through deep diving into language. Not that opening chatgpt makes you understand everything about a culture or a group of people.
No one here was saying LLM's are a panacea for cultural misunderstanding and it's bizarre you think anyone on HN would believe that. You've created an imaginary argument for the sake of positing your opinion. You did however bring up willingness, which is an important consideration.
This is getting rather pointed and I ask you to consider whether it's just to be questioning my motives for posting. It's clear you think I've not addressed your argument, the civil thing is to attribute that to a perceived misunderstanding and not some desire to speak past you with my own opinion.
But if you compare two experienced developers - one of them uses AI, the other doesn't - I don't see what's there to be reviled, one of them just found a tool that makes them more productive.
The median software developer in the US is solidly earning six figures. That's going away, and hundreds of thousands of people are going to be angry about it. Their previously solid path up and out of the middle class is about to vanish. They were betting on that. A large number of people are in school today thinking that median software developer job, writing code, paying $110k per year + great benefits, is still going to be there in 2033.
I was an experienced polyglot developer and I occasionally used PHP for rapid prototyping, but the hate knew no nuance on HN and elsewhere.
A software developer in the future will be someone who only interfaces with AI centric tools, with the AI writing essentially all the code. It'll be rare to directly tweak code for the typical software developer. It will no longer mean someone who writes code to develop software, it will mean someone who uses AI tools to develop software.
What you think of as a software developer today, isn't what people will know them as in another 15 years. It'll happen fast. The code is going to get abstracted away. An 18 year old in 2040 that is looking to become a software developer will go to school for how to build software, which will mean using AI tools to do it; they won't go to school to learn how to program. In fact they won't know much about writing code at all.
In several decades it will be a far more rare skill, that will be viewed as a bit archaic, to still directly write software code. Stories will abound about the old times when millions of people used to directly write code, and they got paid ridiculously well to do it. That problem got solved.
One of the great benefits of AI will be abstracting away complexity (complexity as far as the human mind is concerned), and writing software code is a big juicy target that won't be difficult to abstract away (for something like GPT version 13).
AI tools assistance tools currently help with code-writing at function-level, which is analogous to copy-pasting PHP functions from SO. There are a lot of opportunities to shoot oneself in the foot with both. Without an understanding of the larger (sub)system, AI-generated code will have integration bugs and superfluous code. A good/experienced developer can use the same tooling and emd up with excellent code, but that won't stop the ecosystem from being tainted as awful.
With the growth in token sizes, AIs will be able to ingest/understand larger codebases and generate more consistent code, but that will happen over time, just as PHP got Laravel as well as base-language improvements.
Only reason FANG employees are not paid McDonalds salary is because of the barrier to entry.
The barrier will go down yes, but so will wages.
But I also think that if, like the author, your optimistic scenario is that things will be better "software-ized," you're in for a disappointing century. The incentives that lead to most software being crap are not going to change in the slightest.
Instead, more things will get software. Some of it will be good, most of it will not.
Also, I think that software engineers in general will get more productive because there will be almost no reason to spend hours pouring over docs anymore. That's a big productivity enhancement. Also, if want to work in a framework or language that you're not an expert in? That's easy peasy now. AI will be a boon to companies that have legacy systems that they were previously struggling to find engineers to work on.
I think the big issue for software is going to be that engineers are no longer the bottleneck and the focus is going to shift from _how_ we build things to _what_ we're building and even more importantly _why_. Engineers should be rejoicing over that because it means that software development is going to be a lot more focused on building and creativity than the mechanical details of how do you get framework X to work with data Y.
What about the ones that are already digitized and despite their unavoidable convenience for being so, are less enjoyable than they used to be and more loaded with surveillance and advertising?
One real issue I hope we resolve is to then see workers benefit from their increased efficiency in regards to their compensation.
Pay has not kept pace with increased workers efficiency historically. The efficiency gains technology produces tend to go to the top (C Suite and shareholders). While I, like many, am a shareholder of many companies (so I do benefit as equity markets go up) - the vast majority of Americans own very little if any equity in any equities and this is what leads to absurd wealth disparity.
It'd be nice if this time were different and we could create a system that avoids all the extra cash this generates from being funneled up like regulatory capture - I suspect it won't be.
I think opening doors for more people to enter the career is good, and making it easier for developers to do their jobs is good. But I don't think quantity over quality is the right model to apply to software development, because history is full of examples of how just shipping something that works can mean massive expensive consequences down the road.
Regardless of the “random person thinking out loud with nothing but their own opinion as evidence” aspect to this article, they really are on the money with this one little fragment of their post.
Lots and lots of developers, and no meaningful difference between someone who took a two week “how do I write prompts” course and someone who spent 4 years at university. Learning programming. Or law. Or business. Or teaching. Or writing.
That’s the future I see coming, in probably the next 3 years.
However, as great that might be for companies who like cheap ass labour, it may not be great for individuals…
I think it’s fair to say it’ll be disruptive, that’s for sure.
Knowing how to use an LLM, having access to a good one, using it regularly anywhere you can, you would still get, what - 10%, 20% increase in productivity? It's significant but it's not "your experience and education don't matter" level of significant.
And on the other side of the room some people are still arguing whether these things are useful for anything at all.
Yet.
(c) 1960–
A large share of developers making six figures today, that are in the middle ~15-25% group (in terms of skill), will often not be good enough to cut it in the future as the field narrows for writing code, the gravy train for the median and below will end and the pay will drop below six figures (in today's dollar). The entire bottom 1/3 or more of the code writing software field (people which are earning $75k-$85k+ today in the US) will largely be obliterated.
There are around 1.1-1.2 million software developers in the US today, earning roughly a median of $110k-$115k before benefits. Until recent developments, those people were almost all writing code directly (with some standard copy-pasting from Stack of course). At least half of those jobs are going away and quickly. Most will transition over to creating software using AI, but rarely or never writing code directly, as few people write machine code today. AI will similarly propogate a new layer, with fewer writing code directly (and that field may perpetually shrink for decades as it gets squeezed ever tighter by ever better AI approaches).
The number of people directly writing code has or is close to peaking. This is the end of that era.
People in the emerging software world (eg parts of Eastern Europe or Latin America, etc) will have a potent economic ladder largely pulled away from them (this part sucks big time, but nothing can stop what's coming). The opportunities to use AI to create software will still be there for those people (it won't pay nearly as well as writing code does today); it'll become a lot more rare for rich corporations to pay well to write code directly off-shore and the job numbers will dwindle substantially.
It didn't kill painting, it just changed it.
It also opened up a new field of the professional photographer.
I can't stress this enough this is exactly what's going on.
AI is a tool for simple boring repetitive tedium that takes too long.
Abstract art came because after photography, painting the exact same painting of yet another rich person didn't pay the bills anymore. And frankly, thank goodness for that.
Work in the future is not going to look like it does now. It's going to mutate. Ok, that's fine, it always does.
If this article is not right, the consequences for those of us without independent means could be dire. The time to create a safety net for those displaced by AI is now.
> look around you and count how many products or services you use regularly that you wish are better digitized or “software-ized.” Unless and until that number is 0, there will be more developers, not less.
Even before LLMs, fully digitized markets are regularly disrupted by better iterations of software.
More developers will be needed because AI is increasing the addressable market of software.
Everyone is a 'developer' without needing the expensive bootcamps and 24 hour scam courses and there will be less highly paid ones, including juniors and seniors.
So in reality it is less developers needed (and being hired and being highly paid) and an abundance of 'developers' which is effectively everyone using AI and LLMs for almost for free to drive it all the way down to zero.
For me, it's a great starting point that reduces carpal tunnel, but not much more.
Essentially professional translations are now much more accessible to people that would've not been able to afford them in the first place. Yes, it pays less per page, depends on the type of translation actually, but it is also much faster to do and there's more business.
Does it pays less on per page basis? It mostly does (but it highly depends on the type of translations: legal, medical, etc have the same fees) but it is more than compensated by being able to churn much more work by focusing on reviewing and interpreting the author's meaning and context.
What? How does this logic work out?
Is this not considering how AIs can software-ize products and services in place of a software developer?
How about talking about how AI can empower a single developer to do the job of 10 people?
These logical implications obviously point to "less developers".
I mean it seems really really obvious to me. Whether you agree with the potential of AI or not you need your "lens" to include this angle.
The problem I'm seeing is a lot of "simplification" to the point of blindness. Certain Software engineers don't want to face a world where a machine completely trivializes the skills they spent to much time cultivating so they self induce blindness. They lie to themselves the same way tobacco execs and big oil execs had to lie to themselves before lying to the world.
The main problem is this: While we made significant progress with LLMs in producing an intelligent AI, We don't know if the current problems with LLMs (hallucinations, occasional logic errors, etc..) are technological dead ends... That's it.
That's the main crux everyone is debating about. Even the big experts in AI like Hinton are divided on this fact so there is actual controversy here.
I think you missed some of the nuance of the article.
You’re talking about how many jobs there will be, and they’re (also) talking about how many people will qualify as “developers”.
Will the number of jobs go up or down or stay the same? Who knows? Probably down.
…but that’s only half the story.
Will the number of “developers” able to do what only “developers” can do now go up down or stay the same?
Well, since my mom told me she put some php on her Wordpress site from chat-gpt last week, I’m gonna sit in the “a considerable number more developers will materialise” camp. I.e. a huge glut in the number of people competing for that (probably shrinking) pool of jobs.
Downvote me as you wish, plesbians.
That's the story of life itself: whole generations of people die, ideally while planting ever-bigger trees for future generations to enjoy the shade under. This happens over and over, drips of tragedy in an ocean of comedy. :D