> Will AI replace developers? The Stanford data shows a tilt toward seniority rather than mass displacement: entry for 22 to 25 year olds is suffering, older workers are growing. WEF expects a net gain of 78 million jobs by 2030.
As a 42 years old web developper, this situation gives me headaches. I feel both with new superpowers and dangerously threatened. I feel we will need much less developpers and yet I can imagine some years where a developper will be able to handle much larger projects.
I'm pretty convinced that manual coding will be done by late 2027 for at least webdev. At that point the optimal workflow will be coders controlling agents.
Not too far after it will be coders controlling agents who control agents who do the work.
The huge apps that my coworkers build out with minimal human input makes me scared to be a web dev in this environment. Just way too much can get automated.
Yeah completely unmaintainable code that cant handle more than 1k users before it collapses on itself. And quite probably so ugly its literally impossible to use.
That only matters if you need to support 1k users. Tons of software doesn't and if you can spit out a web app in a few hours a lot of companies are going to do that.
I'm sure there is still a pot for really high quality stuff but it's a tradeoff a lot of companies don't care. But the interfaces - they aren't ugly, they're very usable and complete, which is the main reason I think web dev is in trouble. It'd take me 10x the time to do it myself.
Simple web apps, the kind you're describing have never been difficult, they were never what kept SWEs employed.
I'm not even sure what you're talking about when you say "web apps", because the kind of "web apps" that have kept SWEs employed are created by large organizations that have large requirements. You sound like you don't know what you're talking about. Companies also aren't going to just stop growing, iterating and improving, which requires expertise. Im worth more than ever with AI, my value as a SWE has only been increasing.
You people seriously need to stop using LLMs because your brains are devolving like taxi cabs drivers who had parts of their brains responsible for navigation atrophy after the advent of GPS.
I don't get why you're being so rude. What's the point?
For the record, I hardly use these models. I've simply seen that they can replace existing web apps. That hurts the market for web devs, most orgs aren't building out large apps they are building out smallish ones and generative models do that today.
Sorry, Ive been hearing these exact same arguments for 3 years now and its always extermely divorced from reality. Its starting to get very tiring, its hard not to be rude at a certain point.
Don't go around telling people their careers are in peril when you clearly have no clue what you're talking about. What do you expect?
You've identified a very specific type of web dev, one that works on perf critical apps in a corp environment and said corp cares about perf and quality. Tons of people simply don't work in those environments and unfortunately this threatens their career.
I work in an environment where we can vibecode 4/5 of our web apps because they are either small enough or the perf doesn't matter. Or, I've looked under the hood and these are writing endpoints that would take me 3x the amount of time it would take it to use the same libraries. You say it's divorced from reality, I'm saying it's just what I've directly observed.
I'm glad it doesn't affect you, but you don't have any kind of statistical measurement (nor do I) to know how much of the market this impacts. I do know for a lot of the web dev I did pereviously, I could do it viea these models today.
Yeah I've been at that stage since Opus 4.6 pretty much. There's still a lot of back and forth in the conversation on complex tasks, but I have not written any code. We are working on a MASSIVE project for our multi-billion dollar client and the whole team is completely using AI. Even the UX guys and our project managers are making the prototypes they want us to implement by vibe coding it first in a static app. I haven't seen Figma usage in months.
You must have incredibly low standards for quality if you think manual coding will be finished any time soon, web dev or any.
secondly, using llms well still requires you to be a great programmer. Y'all are losing the plot.
Also, just because its easier to do doesnt mean everyone is suddenly going to want to do it. There's still always going to be only a small percentage of thr world that likes to build things.
Have you seen the code these things produce? Just because it compiles/ runs/ loads does not mean its durable. Its wild you people have these opinions, and it leads me to belive you have no idea what you're talking about.
There is mass dunning kruger due to llms, and people are speaking with way too much authority about things they do not understand. I hate this website.
> I feel both with new superpowers and dangerously threatened.
In some ways, having a superpower is not all that great because then you are forced by the prisoner's dilemma to use it. And if you think about it, it comes at a cost: a higher mental strain over a long period of time.
To use a physical analogy: a workout machine at the gym is very likely to assist you in targeting specific muscles compared to just lifting stuff you need to lift, but it also allows you to push that specific muscle group harder and thus increase the chances of injury to that muscle group precisely because it's so targeted. I think AI is like that with the brain.
(I'll add that it's perfectly possible to work out safely at the gym with lots of equipment. But I've spent years at the gym and I've seen a ton of injuries because the temptation to push a specific muscle group hard is there. And, AI is much worse because with AI, there is an economic incentive to push harder, whereas at the gym it's in your best interest to dial back...only sometimes people like to show off to their gym buddies.)
I wonder if LLM's might be closer to the weight belt in terms of lifting? It is a tool that lets you exceed your unassisted limits. If you kept doing the other things you would be no worse for using it, but if you only train with it I am not sure you would be a very well rounded lifter. The weight belt lets you do a few lifts better, at potentially the cost of being weaker than would be required if you did that left unassisted? Might be a bad metaphor honestly.
Aside (shower thought):
A little of a tangent, but I have always felt that the weight belt is kind of odd in that it has become completely accepted to use one for the deadlift and squat. Like, why did we decide that this one form of brace is acceptable but others enhancements are not "real" weightlifting? Why not an incline plane? Why not a pulley?
I'm not at all arguing for or against the usage of the weight belt. I just find it odd that it has become so accepted; seems curious how we decided to draw the line here and not there.
I don't get that experience at all. LLM's both relieve burden and improve speed. The only trouble I have with AI is when I run out.
That, and I miss coding... but... this new thing is its own puzzle, and I still do lots of pseudo-code and such, to talk to the AI, which kind of scratches the itch.
> AI Engineer. Number one on LinkedIn Jobs on the Rise in the US two years running, number one at Robert Half. About 1,550 new US postings a week, median 176,000 dollars. This is the engineer who puts LLMs into products: RAG, agents, fine-tuning, evaluation.
I'm doing all those things. The RAG work was interesting, but putting it on my resume has done nothing. Nobody is hiring. I have recent multi-year FAANG experience too. Before 2023 I could blast out 10 resumes and get at least 8 or so hits with multiple offers. Now I can send out 100 resumes and lucky if I get a single rejection letter. I am seeing the same thing reported by other white-collar working people like HR reps and project managers in all industries. There is definitely a stealth recession/depression scenario playing out right now and has been for at least the last two years.
Is it possible to jump into this with a background as a senior dev or is it too late? I don't know where to get started, would you just look like a junior or entry level candidate by doing that?
We hire “AI engineers,” and honestly the expectation most of the time is that they can demonstrate an actual interest and real work in or with AI. Mind you, we are not a frontier lab, but AI gaming company, and someone with senior dev knowledge is hugely valuable to the team if they actually have rolled up their sleeves and built with AI and push to improve our operations with AI. There isn’t an academic signal that would tell us you are really AI forward, and so much of the engineering job is super charged by having a real dev background for us.
I can’t say if this will change of course, and I only speak for a small gaming startup.
You can use it at different levels of granularity to fit your needs. I find a lot of the fun in designing the way the code interacts and class/module boundaries and responsibilities. So AI takes away the typing (code) part and makes it more fun at times. It give me more time to focus my thoughts.
It (focusing on the design) is about as fun as taking part in the 4 days specification/planning marathon with a team whose super intelligent members cannot comprehend trivialities and nitpicking is needed, as they really like to code and cv-boosting coding and engineering. and are willing to go large lengths to do any overengineered over-complicated contraption that slightly resembles the solution. Exhausting. At least I can yell at the AI without negative consequences. For now.
>Given the backlash amongst gamers against AI, how does being an “AI gaming” company work?
They are delusional or know nothing about how development works. I can promise you there's not a single major game studio left out there not leveraging AI in significant parts of their workflows at this point.
most eng roles at frontier companies don't require prior experience specifically with AI, there's a ton of regular engineering work required to train / serve models
Absolutely; I'd check out AI Engineering by Chip Huyen to get a fundamental understanding of the space and do a couple of side projects in the area to get familiar and be able to speak to the nuances of working with LLMs, and that will be enough to start working to transition and apply for relevant roles.
It’s for sure not too late, why would it be? The industry is young and evolving. It’s not even clear what an AI job is exactly, the definition and job titles have been changing a lot
Maybe - as with most ultra competitive industries you either know someone or have some extremely niche yet hyper productive knowledge others don’t have.
I have multiple Senior AI Engineer roles posted (really, senior / staff SWE who has reps and sets using AI for development, perhaps working with AI infra) right now. Right now it's a lot of getting Claude working well, model routing working well, and getting us off of commercial LLMs. Trying desperately to hire, but get a lot of "AI-pilled" non-developers in the same way I would open SWE roles ten years ago and get constant "Software Bootcamp" grads. These roles exist, in my anecdotal experience of having multiple of these roles open on my team lol. Perhaps this is just 1/100 ads, or 1/1000.
how do you interview for those? I think interviewing has not kept up with how things are, leetcode rounds seem to make even less sense nowadays and feel even more of a hazing ritual.
Personally seeing an "AI engineer" job posting I would not even apply, despite having many years experience in general, and experimenting/running models on my own hardware and being responsible at work for several skills, mcp and so on, it just feels that that kind of job is looking for some ex-openai/anthropic/ai-lab kind of skillset, which is not something a lot of people have; if this is what you are looking for of course it is what it is.
I think these days being "AI conversant" should be expected for senior/staff positions, it can also be listed a requirement, but not sure putting it as a job title, it feels more like having "IDE staff software engineer" which is not something anybody outside of jetbrains/vs.code would do.
sounds fun, I had fun fine-tuning some open weights models for local use, and I run a free public inference engine here: https://stateofutopia.com/services/ai/ it's fun.
What's the best way to get into any kind of frontier training or inference? Obviously OpenAI and Anthropic would be great, but I would love to work at inference literally anywhere. LLMs are super cool and I want more systems problems. Or is the boat sailed and it's too hard to get in?
My current issue is that solo learning and bootstrapping is hella expensive money wise and my current job doesn't pay that much either.
Mostly title inflation. Using GPT API does not make you an "AI engineer". I don't even know if anyone would seriously call themselves this. Usually they are either software developers that actually work at an AI company or a researcher. The others are mostly fake jobs.
I'm pretty sure people suddenly had BS like "Blockchain engineer" in their linkedin headline in 2022.
Edit: https://openai.com/careers/search/?q=Engineer
Most jobs involving engineering don't even have AI included in their title at OpenAI, though they do have Applied AI engineers and AI support engineers. I highly doubt these other companies suddenly have a huge growth of AI engineering jobs available.
I interview a lot of larping 'AI engineers' that either vibe coded a side project or were involved in low stakes pilots
Irl theres a short list of companies that actually build / apply ai at scale. for people trying to break in its usually not expected to have experience with it, just don't lie about it
> In AI Engineer postings the skill "prompt engineering" appears 8.9% of the time, RAG 13.6%. *The skill stayed, the job dissolved into other roles*.
That is Claude vibe. I thought it was an ok article when I was about half way through it, then hit this. And overall it seems like it could have been more focused, it listed various AI roles a bit carelessly.
> That is exactly what working with small and mid-size businesses looks like: find the process where AI pays off, build the agent or the integration, and get it to the people who will use it.
This is exactly right.
I’m making good side money from building and integrating “AI” with small business tools.
Custom CRMs and tracking solutions so they don’t have to buy salesforce etc…
Notably this doesn’t reliably work for juniors, which supports the article.
You need to already have experience making enterprise scale software, otherwise you’re going to miss some massive authentication or data exposure problem.
I anticipate the return of the software consulting company like initech, but instead of having Tom and Milton and Samir and Peter etc…, you only have Lumburgh and Michael Bolton.
I would like to have a resource, or maybe a coalition of companies that sign up and offer jobs that are guaranteed not to use AI. Although it might be true that there are some new jobs that didn't exist before, it would be great for people to band together to create a system that promotes the human element over using AI.
At this point that’s like asking to work at an artisanal hand made goods factory. Not common and I’m not even sure there’s particular demand on the consumer side for artisanal hand crafted code despite the negativity towards AI.
It's worse even than that. Although I personally wouldn't pay for a bespoke hand designed piece of furniture I can at least see that it is an artisan piece and understand the rationale.
Software is all shit anyway, it just helps me get a job done. How would artisanal software even distinguish itself. Leave the visual appearance aside. I mean the functionality.
Well for one, a lot of people now solo dev a project with AI and then often seem to abandon the project. There's not much sense of owndership in some of it. I try and use software that has as little AI influence as possible.
As a 42 years old web developper, this situation gives me headaches. I feel both with new superpowers and dangerously threatened. I feel we will need much less developpers and yet I can imagine some years where a developper will be able to handle much larger projects.
Not too far after it will be coders controlling agents who control agents who do the work.
We're in for a very tumultuous few years
I'm sure there is still a pot for really high quality stuff but it's a tradeoff a lot of companies don't care. But the interfaces - they aren't ugly, they're very usable and complete, which is the main reason I think web dev is in trouble. It'd take me 10x the time to do it myself.
I'm not even sure what you're talking about when you say "web apps", because the kind of "web apps" that have kept SWEs employed are created by large organizations that have large requirements. You sound like you don't know what you're talking about. Companies also aren't going to just stop growing, iterating and improving, which requires expertise. Im worth more than ever with AI, my value as a SWE has only been increasing.
You people seriously need to stop using LLMs because your brains are devolving like taxi cabs drivers who had parts of their brains responsible for navigation atrophy after the advent of GPS.
For the record, I hardly use these models. I've simply seen that they can replace existing web apps. That hurts the market for web devs, most orgs aren't building out large apps they are building out smallish ones and generative models do that today.
Don't go around telling people their careers are in peril when you clearly have no clue what you're talking about. What do you expect?
I work in an environment where we can vibecode 4/5 of our web apps because they are either small enough or the perf doesn't matter. Or, I've looked under the hood and these are writing endpoints that would take me 3x the amount of time it would take it to use the same libraries. You say it's divorced from reality, I'm saying it's just what I've directly observed.
I'm glad it doesn't affect you, but you don't have any kind of statistical measurement (nor do I) to know how much of the market this impacts. I do know for a lot of the web dev I did pereviously, I could do it viea these models today.
I'd say decades, not years.
How can I bet a shit load of money against this?
secondly, using llms well still requires you to be a great programmer. Y'all are losing the plot.
Also, just because its easier to do doesnt mean everyone is suddenly going to want to do it. There's still always going to be only a small percentage of thr world that likes to build things.
Have you seen the code these things produce? Just because it compiles/ runs/ loads does not mean its durable. Its wild you people have these opinions, and it leads me to belive you have no idea what you're talking about.
There is mass dunning kruger due to llms, and people are speaking with way too much authority about things they do not understand. I hate this website.
In some ways, having a superpower is not all that great because then you are forced by the prisoner's dilemma to use it. And if you think about it, it comes at a cost: a higher mental strain over a long period of time.
To use a physical analogy: a workout machine at the gym is very likely to assist you in targeting specific muscles compared to just lifting stuff you need to lift, but it also allows you to push that specific muscle group harder and thus increase the chances of injury to that muscle group precisely because it's so targeted. I think AI is like that with the brain.
(I'll add that it's perfectly possible to work out safely at the gym with lots of equipment. But I've spent years at the gym and I've seen a ton of injuries because the temptation to push a specific muscle group hard is there. And, AI is much worse because with AI, there is an economic incentive to push harder, whereas at the gym it's in your best interest to dial back...only sometimes people like to show off to their gym buddies.)
Aside (shower thought):
A little of a tangent, but I have always felt that the weight belt is kind of odd in that it has become completely accepted to use one for the deadlift and squat. Like, why did we decide that this one form of brace is acceptable but others enhancements are not "real" weightlifting? Why not an incline plane? Why not a pulley?
I'm not at all arguing for or against the usage of the weight belt. I just find it odd that it has become so accepted; seems curious how we decided to draw the line here and not there.
That, and I miss coding... but... this new thing is its own puzzle, and I still do lots of pseudo-code and such, to talk to the AI, which kind of scratches the itch.
> AI Engineer. Number one on LinkedIn Jobs on the Rise in the US two years running, number one at Robert Half. About 1,550 new US postings a week, median 176,000 dollars. This is the engineer who puts LLMs into products: RAG, agents, fine-tuning, evaluation.
I'm doing all those things. The RAG work was interesting, but putting it on my resume has done nothing. Nobody is hiring. I have recent multi-year FAANG experience too. Before 2023 I could blast out 10 resumes and get at least 8 or so hits with multiple offers. Now I can send out 100 resumes and lucky if I get a single rejection letter. I am seeing the same thing reported by other white-collar working people like HR reps and project managers in all industries. There is definitely a stealth recession/depression scenario playing out right now and has been for at least the last two years.
I can’t say if this will change of course, and I only speak for a small gaming startup.
You're just staring at a screen pushing pixels all day. Whether I do that with Claude or without doesn't matter to me.
Can you explain what you mean by this? Do you evaluate the design without reviewing the generated code?
They are delusional or know nothing about how development works. I can promise you there's not a single major game studio left out there not leveraging AI in significant parts of their workflows at this point.
most eng roles at frontier companies don't require prior experience specifically with AI, there's a ton of regular engineering work required to train / serve models
This stuff has only existed for 3 years. Everyone is faking it and no one really knows what's right or wrong yet. You're more than qualified.
Personally seeing an "AI engineer" job posting I would not even apply, despite having many years experience in general, and experimenting/running models on my own hardware and being responsible at work for several skills, mcp and so on, it just feels that that kind of job is looking for some ex-openai/anthropic/ai-lab kind of skillset, which is not something a lot of people have; if this is what you are looking for of course it is what it is.
I think these days being "AI conversant" should be expected for senior/staff positions, it can also be listed a requirement, but not sure putting it as a job title, it feels more like having "IDE staff software engineer" which is not something anybody outside of jetbrains/vs.code would do.
sounds fun, I had fun fine-tuning some open weights models for local use, and I run a free public inference engine here: https://stateofutopia.com/services/ai/ it's fun.
where are you planning to run your models?
My current issue is that solo learning and bootstrapping is hella expensive money wise and my current job doesn't pay that much either.
I'm pretty sure people suddenly had BS like "Blockchain engineer" in their linkedin headline in 2022.
Edit: https://openai.com/careers/search/?q=Engineer Most jobs involving engineering don't even have AI included in their title at OpenAI, though they do have Applied AI engineers and AI support engineers. I highly doubt these other companies suddenly have a huge growth of AI engineering jobs available.
Irl theres a short list of companies that actually build / apply ai at scale. for people trying to break in its usually not expected to have experience with it, just don't lie about it
That is Claude vibe. I thought it was an ok article when I was about half way through it, then hit this. And overall it seems like it could have been more focused, it listed various AI roles a bit carelessly.
> Here is the short version, with numbers and links.
> The one-paragraph summary:
This is exactly right.
I’m making good side money from building and integrating “AI” with small business tools.
Custom CRMs and tracking solutions so they don’t have to buy salesforce etc…
Notably this doesn’t reliably work for juniors, which supports the article.
You need to already have experience making enterprise scale software, otherwise you’re going to miss some massive authentication or data exposure problem.
I anticipate the return of the software consulting company like initech, but instead of having Tom and Milton and Samir and Peter etc…, you only have Lumburgh and Michael Bolton.
The numbers are stuck at 2018 levels. Accumulated inflation since then is over 20%. This bubble is not glamorous at all.
Software is all shit anyway, it just helps me get a job done. How would artisanal software even distinguish itself. Leave the visual appearance aside. I mean the functionality.
artisanal software is software that has a bound on the complexity. A human has to reason about it.