Let’s assume that in 10 years, AI has advanced absurdly, insanely fast, and is now capable of doing everything a Senior SWE can do. It can program in 15 different languages, 95% accuracy with almost no mistakes, can create entire applications in minutes, and no more engineers or SWEs are needed… What will all the devs do? Do they just become homeless? Transition to medical field, nursing? Become tradespeople like plumbers, HVAC?
Honestly people are getting distracted here. Now lets say A.I makes developers 50% more productive thats a huge boost for smaller companies with only handful of developers.
Many companies are only thinking about reducing costs for themselves but at the same time they’re freeing up a lot of talent for new and old competitors.
Here’s some food for thought:
- Open source developers may use A.I to develop better software to close gap between paid alternatives. (Blender, Gimp, Krita, Linux distributions, mastodon, lemmy, pixelfed)
- Many LLM’s can already be ran freely and locally. These will only get better as technology progresses. This can make selling/profiting from A.I services a lot harder
- A.I may be used to block ads or obfuscate (create bunch of fake data) user data that is sold to advertisers.
- Some media sites are already using A.I to write articles. Whats the point when users may just use chatbot to get all the information without ever engaging with the source.
These are just few that come to mind. but the unkowns with this are quite terrifying.
Now lets say A.I makes developers 50% more productive
That’s wildly optimistic. If I recall correctly, early studies are showing the 51% of participants who saw any improvement, reported an average of a 20% improvement.
Even granting that optimism, since 5% of all software projects are on time and within budget, we may look forward to a whopping leap to 7.5 out of every hundred software projects arriving on time and under budget, in a best case scenario.
The hard truth no one wants to talk about is that the average software development team is awful.
This glorified parrot tool of LLMs is one of the coolest we have seen in awhile, but it’s not going to materially fix the awful state of the field of software development.
The average software development team doesn’t understand how to deliver high quality maintainable solitions on a reasonable timeline.
AI may mildly improve the delivery timelines of the still very incorrect and over-budget solutions delivered by the average development team.
That’s wildly optimistic. If I recall correctly, early studies are showing the 51% of participants who saw any improvement, reported an average of a 20% improvement.
Yes the value is wildly optimistic to match the expectations driven by all the hype from these companies pushing their LLM services.
Even granting that optimism, since 5% of all software projects are on time and within budget, we may look forward to a whopping leap to 7.5 out of every hundred software projects arriving on time and under budget, in a best case scenario.
The hard truth no one wants to talk about is that the average software development team is awful. The average software development team doesn’t understand how to deliver high quality maintainable solitions on a reasonable timeline.
You’re oversimplifying things here there are a lot more variables that influence success in software projects. The company you work for might have oversold the project, the client might only have vague understanding of what they really want, project management may fail to keep the costs, developers or timeline in check, client or the company you work for might have high employee turnover causing delays as new employees need proper induction to the project, the initial tech stack may become deprecated or obsolete mid-way the project, etc
You’re oversimplifying things here there are a lot
I think… we’re agreeing?
My point is that what is currently possible with AI doesn’t solve any of that.
People in this thread keep discussing growth in programmer productivity as if programmer typing speed and number of languages known are the limiting factors of programmer productivity. They are not. It’s all the other bullshit that makes (the vast majority of) programming projects fail.
My source: I know so many programming languages and I type insanely fast. My team is also productive beyond all reason. These two tidbits are only related in that I tried and failed with the first before succeeding with the second.
You seem like someone who hasn’t really worked in software development.
Software engineering does not simply mean coding. A production grade software application goes through analysis, design, implementation (where coding happens), testing (several phases), release and maintenance. Not to mention infrastructure concerns (storage, databases, microservices, service orchestration, middleware, etc). The whole process is too nuanced and complex to conclude that AI would make the whole career obsolete. It might shake up some areas of software engineering but only a small part of it.
You’ll still need people to verify that the AI generated application actually behaves as per the business logic, runs optimally with the hardware you have and scales as your business grows. Which means engineers for testing and reviewing the generated code plus engineers to setup the infrastructure where the application will run.
engineers for testing and reviewing the generated code plus engineers to setup the infrastructure where the application will run.
That’s still a lot of software engineers displaced in the hypothetical scenario. That means you only need the devops and qa engineers, and a solution architect or principal engineer or whatever your company calls that sort of role for the analysis and design part.
That will never happen, or at least with how ai currently works. It’s basically a glorified autocorrect, it uses the same technology underneath.
But presuming it does, yes. We will have to go to another industry, like AI prompting. Coding is a tiny part of professional software development.
Yes, exactly this.
When compilers came along, some people honestly thought it would dumb down programming so much that anyone could do it.
When high level programming languages came along, they rejoiced again - now finally anyone can make software.
When Intellisense meat you no longer had to remember variable names, write your own imports and could guess how most libraries work, the bells rang out once again in celebration.
And now we have AI, it’s cool but really just another step like all those steps before. For me, it’s a replacement for the documentation I never read anyway. I can ask an AI a stupid question rather than bothering a human developer.
These days it’s my job to manage a small team of developers - when I ask them why they wrote a stupid thing that makes no sense, 90% of the time, the answer is that an AI wrote it for them.
Glorified autocorrect… YES! It’s a really good analogy that i will use to temper the expectation of my boss. Also: AI hallucination is just a fancy way to say ’it’s a wrong answer’.
And if it’s going to be full-blown AGI then we’ll become AI psychologists.
deleted by creator
I even used Claude AI to write an entire C# application, I did ZERO coding, yes, literally nothing! I have NEVER coded in C# before, I gave it all requirements, worked with it like a project manager… it created a full blown working application that was beyond my expectations.
I achieved the same in 2000 with a home grown framework, and again in 2006 with Ruby on Rails.
Astonishingly fast prototyping is a quarter of a centrury old.
- How are you enjoying maintaining this app in production? (Or is it not there yet, because it’s just very nice for a prototype?)
- How did Claude AI do at deploying it?
- Are you satisfied with Claude AIs answers to your boss’ traffic analytics and load balancing questions?
- When will Claude AI let you know how the A/B tests proved out for optimizing sales?
- Or doesn’t it do those things yet?
Computers are replacing us. They’ve been at it since their inception.
Keep learning the trade and you’ll find there’s a metric ton more that computers cannot help with, than that they can help with. That will get better. I’m working at making it get better.
I figure that my learning how to train the computers is job security. I didn’t count on it being a harsh lesson in how long it’s going to be before computers get not stupid.
I do have a plan for when I automate myself out of a job. It’s just not a plan I’m really counting on, because I’ve been trying for decades and I only have so many decades left of doing this.
I’ve been constantly advised to have an exit plan, for when the computers replaced me, for the entirety of those same decades.
Most often by the same people who want me to charge less.
Funny thing, that. Take care who you listen to on this topic, and what their motives are.
My motive is to (continue to) charge the rest of you a shit ton of money before the AI finally replace us.
It does help me if you all don’t buy into the bullshit that CEOs have been spouting about replacing us all.
We’ve all been undercharging for about 3 years due to it.
AI hasn’t accomplished jack shit, but a lot of you have accepted lower pay than you probably should.
I make very good money, but I can’t help but notice that it would be a bit more, if the rest of you would wise to the scam and raise your own prices.
Some days, all I do is code.
So your instincts are correct. You need to learn the rest of the job, before the part you are doing is replaced by robots.
Lots of my colleagues in SWE use full blown AI development tools.
We all use full blown AI development tools. Before that we had other tricks that did the same thing.
We must beware mistaking the instrument for the musician, or we get sold a broken old instrument that doesn’t perform miracles outside it’s master’s hand.
What are you using to get Claude to write for you? I’ve been using it to write a full stack Go/javascript app but it needs a lot of handholding.
There will always need to be a human operator to guide the machine.
I’m using Cursor. It’s done some very impressive things for me. https://www.cursor.com/features
deleted by creator
You have to understand what software can do, how to design it, and how it should interact with other systems in order to write software and not just code, and AI can’t do that. If you tell it to make you A, and what you really want is B, you’ll never get what you want.
Only about 10-20 percent of my job as a software engineer is writing code. AI can be really amazing at writing code, but unless it can do the other 80-90% of my job without me, I’ll be safe.
Now, whether middle and upper management will know this is an entirely different question. A lot of them think that lines of code written is a good measure of productivity, when in fact it’s often the opposite.
I foresee there being a big struggle for management to come to grips with the fact that AI is better suited at their job than ours.
My best days as a software dev are negative line days.
NegaSLOCS are the best SLOCS.
Hear, hear!
If it is able to replace software devs, it’s probably able to replace 95% of the jobs that require mainly using your brain.
Yeah it’s being applied to software devs right now but it’s already capable of replacing nearly every manager/supervisor in existence.
It can make schedules, direct tasks based on inventory, and balance a budget. Have a human backup available on call to fix hallucinations and you’re golden.
They’re just gonna sit around and wait a few months until they are begged to come back and can demand more compensation. The current generative AI, which is not general AI, will not be able to replace high functioning jobs. Eventually, a lot of those software engineers will be asked back and get much more for their services.
The plan is to rehire them back temporarily to babysit the AI and fix all the AI generated crap. Then realize it was cheaper to actually just have the devs make code. Then hire them back at a reduced rate on a more permanent basis with the understanding that they believe the code will still be partially generated by AI and cleaned up by the same people and they aren’t paying top tier for third hand AI slop.
They’ve been doing the same thing in IT for decades, just replace AI with outsourcing.
Same in a lot of other industries too. This is literally how capitalism functions. This is how they reduce costs when they can’t find any other way.
Except it is often more costly to do this in the long run, so it’s a fiscally stupid move that corporations seem to make over and over again.
I think part of what perpetuates it is, the people making the decisions don’t stay there long term, so they never really face the repercussions.
Some more stable places seem like they may have realized this though and keep things all or mostly in house.
Well if it can replace senior software engineers… Wouldn’t it also be able to do almost all of the other jobs? Or are you referring to some specific future where AI advances massively, but robotics does not and handymen are still safe?
I’d say if all humans are unemployed, society would change massively. We can’t really tell how that’d work. But if machines / AI do all jobs, get food on the table… I don’t really know what other people would be doing. I think I’d relax and pursue a few hobbies and interests. Or it’d be some dystopia where humankind is oppressed by the machines and I’d fight for the resistance.
But regardless… In a world like that, money wouldn’t work the way it does now. Neither would salaries for labor mean anything.
Yeah. In this wild scenario, only the people who can crack the robots security protocols, and reprogram them, will have any influence over society.
I promise to be a benevolent ruler.
Except Michael Bay will have to return to making Transformers movies full time. Sorry about that, in advance.
Or the people who own the robots and dictate their programming (/control them). That would be my concern. Unless they’re sentient and make decisions completely on their own, they can be used to oppress people to other people’s wishes. As it’s the case with all (modern) technology. And currently AI isn’t shaped by the people, but by a rich minority and big tech companies. And I see some issues with that, specifically, in the near future.
Agreed on all points.
That said, I am smarter than the asshole CEOs, and the current state of computer security is abysmal.
So there’s still some hope that we are barreling toward my (mostly) benevolent reign over endless Michael Bay blockbuster summers.
Hopefully, for everyone’s sake, reality will fall somewhere in between.
But joking aside, money isn’t the only form of power. There aren’t that many billionaires (compared to he rest of us) and a billionaire’s influence is limited by what the rest of us will or won’t do.
Lol. Yeah I get it. Though I still think the rich companies dictate a lot of things. They do a lot of lobbying and paying people to make sure it’s not them who funds the majority of the country, they choose how much you pay for medication and everyday items, they choose to spy on everyone on the internet. Make you buy things you don’t need, make housing prices subject to speculation. Make everyone addicted to their phone and spend like several hours a day with it. Separate society into filter bubbles. I think a lot of these things aren’t liked by the people. Or are extremely unhealthy. Yet, they are a thing and never change. I think because some people will this into existance. Sure, they’re far from being almighty. But it’s enough control they have over everyone already.
And I think as they can use the internet as a tool for their interests (which had ultimately been invented to connect people), they could as well do the same with AI. I mean they train those models and choose in which ways they’re biased. What the can and can not talk about. If that’s paired with the surveillance tech, that’s already inside of each smart TV, smart appliance or Alexa… It’ll be kind of a dystopian scifi movie where someone else watches your steps all day, uses that to manipulate people… some kind of puppet master whom the bots really work for.
I’m really unsure. Sure, almost everything can be hacked. But does that really have an effect on the broader picture? Everytime I see some major hack, the next day it’s business as usual and everything keeps working as it used to.
There are a lot of dumb takes here in the comments
Developer displacement works the same way it does for any other technology
The problem is not that the job is eliminated but that fewer are needed per unit of output
My startup only has 4 engineers because we don’t need 5
This trend will continue until the SV hiring bubble bursts
Even if we stipulate that, I’m not convinced it’s a big deal. The software field continues to grow like crazy and we can never find enough people to hire. If ai gets good enough to take the place of some of that hiring, fantastic!
Coding is just a part of the overall “programming” problem. Most problematic areas are in translating what the customer wants into code (requirements analysis), modifying code to overcome specific constraints, integration, etc and etc
Don’t forget testing the code to make sure what is delivered actually matches what the customer wants.
They’ll either move up the food chain to higher-touch work where AI can’t compete, or they’ll do other things.
Keep in mind that most devs aren’t really all that good at their jobs, so it will probably be economically beneficial for them to do something else. I say this as a long-time hiring manager with many decades of experience in the field.
It can program in 15 different languages, 95% accuracy with almost no mistakes, can create entire applications in minutes
Only if you believe the hype. It can do that in best-case scenarios when the requirements are written as rigorously as code, or where it’s replicating a common pattern.
Do they just become homeless?
During previous layoffs, a lot of them left the field, and some of the rest founded startups. It wasn’t always the case that firms were founded by teenaged sociopaths with strong family connections to VC funding. There was a time when they were founded by people who knew how to do things.
Last time I used it the code it gave me wouldn’t actually run. After 6 iterations and fixing the rest it kind of worked. In theory that should only get better but I’m not sold yet.
I would never have expected it to run, be shocked if it did. You use AI to get over humps, get new ideas and approaches. It’s excellent for time saving in those cases.
AI isn’t ready to replace coders, but it’s quickly going to make a dent on the numbers needed.
AI isn’t ready to replace coders, but it’s quickly going to make a dent on the numbers needed.
Let me push back on this a bit - this belief comes from the assumption that I, as a hiring manager, need more team members because they can only type so fast.
My actual need for separate development team members is to achieve a bench depth of two people in each of the seven specializations necessary to keep my employer un-bankrupt. (My annual bonus is better if I somehow miraculously cover the 14 specializations necessary to make us never look like idiots. But these are wishes, not miracles.)
I don’t currently see any sign that AI will ever materially affect the number of people I need to hire.
In contrast, the specific individuals I hire have massive impact on how many others I need to hire. One person with three specializations brings me massive savings.
But I pay my people to understand our organizational domains of expertise. LLMs don’t bring any new understanding whatsoever into the organization.
I see your thought process there. But I’m not sure modern IDEs led to less devs. Time will tell but I just few most of this as vapor ware atm. Let’s also look at the fact that chatgpt is hemmoriging money even with high price tiers. It is possible this just burns itself out.
Retire. All I ever wanted to be was a programmer. If I can’t do that anymore I’ll just retire. I’m saving/investing every penny I can just in case.
Same. If I can retire before my job is irrelevant, I’ll work on my own projects on my own terms. If I don’t, at least I have a nice pile of assets and can coast with another job.
That said, I don’t think people like you and I will have problems, because we’ll adapt. It’s the “programming is just a job” crowd that would have a lot of issues.
I was going to learn how to give a really good handjob but the AI robots will probably take over that too.
Spend their days (and some nights) tweaking and refining AI prompts to get the stupid thing to generate the software that the dumbass product manager wants and the user does not.
You know…
Pretty much the same thing they do now.
Yeah. The whole job is figuring out just the right away to say “pretty please” to the computer. The ways it’s done changes every decade or so. The fact that it’s a huge pain in the ass has yet to change, in spite of decades of marketing promises.
This thread is full of people comparing OPs hypothetical about 10 years from now with last year’s capability.
Will AI progress that fast? ¯\_(ツ)_/¯ It probably won’t get that good, but it doesn’t matter. If it gets as good as your average junior that’s going to mean something like 100% increase in productivity, which means 50% as many jobs and that’s going to be a BIG FUCKING DEAL.
Especially when it’s going to be replacing a lot of other types of office workers. What kind of job is your average software dev going to transition to? Tech support? Not anymore. UI Designer? LOL. Manager? And who are you going to be managing?
If the US doesn’t hit 15-20% unemployment in the next 10 years I’ll eat my hat. I’ll be eating it either way because I’ll be starving to death.
There is a hard limitation on LLM, it doesn’t and by definition can not have a criteria for truth, and unless something completely new emerges, it will never replace a junior, really. Some managers can be convinced that it did, but that will be a lie and the company that believes it will suffer.
It can transform some junior jobs for sure, some people might need to relearn some practices, there will probably be some shift in some methods, but unless something fundamentally new will appear, there is no way LLM will meaningfully replace meaningful amount of people
