All this depends critically on one premise: that sometime in near future AI coders will become fully automated and produce senior level code. If not we are wholly fucked because currently they are employing less and less junior coders which means that we will be running very low on number of senior coders in a decade or so. If LLMs still need supervision by then there won’t be enough senior coders to do so.
I’ve used it to explore some avenues without having to write a complete implementation. If the approach shows promise, then I go through the code and mostly rewrite it because the code it generates is terrible. I also use it if I don’t care about the project I’m on. They want to “do test-driven development” while having poorly-defined requirements that constantly change on a whim while also setting unreasonable unit test coverage thresholds? Cool, I’ll let the AI shit out a bunch of unit tests and waffle stomp it to satisfy your poorly thought out project requirements.
as a programmer I feel like this would be pretty cool. but this isn’t really how it is at all. I’m usually asking Claude code to do something very specific and then I’m throwing whatever it does away because it’s not correct. if I could have a little baby that I had to babysit I think that would be better
I was “there” with Claude as you describe about 3 months ago. Since then, Claude has stepped up to being able to create fully functional microservices. It helps if you completely specify what you want, it helps if you don’t specify funky libraries or other tech that has poor support on the internet, it helps if your total ask amounts to 1000 lines of code or less - but I have gotten up around 3000 lines before Sonnet 4 choked a few times.
Before this, my AI queries were mostly limited to specific API function call syntax, and they would only be right about 2/3 of the time, which beats randomly trying things myself until I eventually guess the right variation… Yes, it’s better to consult the documentation - when it’s available - it’s not always available.
Conversely, I’d imagine there are babysitters out there who at times wish they could just throw the baby away.
I’ve thrown the baby away many times
Eyoooo
As a senior dev, I say it’s not worth it. Our junior devs rely on it too much and I spent most of yesterday trying to figure out for my junior dev why their code didn’t work. Eventually came to the conclusion that they just have to redo most of it because it’s utter garbage and invents new code to do what the architecture already has.
As a senior dev, I agree but am impressed that you’re dealing with a functional architecture to begin with
Interesting how this article is contradicted by hundreds of others.
Because if we say anything positive about AI programming we get downvoted to hell…
I’m not a supporter of the companies making LLMs and how they profit off others’ intellect, I’m not a supporter of their use of the technology for fascism, genocide, and pure evil. I’m not a moron that thinks LLMs are intelligent.
But I recognize it as a very useful technological advancement, it’s a very useful tool and to pretend otherwise is foolish. LLMs are an amazing coding aid, and when used correctly and fed the right context, they can save hours of frustration and research dead-ends.
Edit: see? I just called them a useful tool and got downvoted…
I share the same opinion as yourself and gave you an upvote. LLM are a truly fascinating technology and I’m amazed by how little we understand on how it even does what it does. That said the process that got us here and what they are currently being used for is amoral at best
I am generally a sceptic myself, especially in my own area, which is software development. But recently in a board game community, someone was scolded for asking ChatGPT about a rule dispute (and it was wrong). All upvotes to unhelpful “AI bad” comments. I pointed out that while this was true 3 months ago, ChatGPT 5 (and only that one) can very accurately answer such questions when asked the right way, showed how to ask the user question and the (now correct) response, and mentioned my 35 board game test questions and results with major LLM flagship models. (Almost all LLMs did horribly, under 70% even in yes/no questions, but ChatGPT 5 with specific instructions or “Thinking” model got 100%.)
Even as a sceptic, I can acknowledge that LLMs just jumped from completely useless to perfect in the past few months when it comes to this specific niche.
That picture accompanying the article is backwards. Why is it portraying the AI as the babysitter and not the baby that needs to be supervised by a human?
You’re absolutely right!
I completely messed up the picture. It should be the other way around. Do you want me to correct my mistake and generate a new picture?
/s
<<proceeds to produce a derivative of the same picture>>
Carla Rover once spent 30 minutes sobbing after having to restart a project she vibe coded. Rover has been in the industry for 15 years, mainly working as a web developer. She’s now building a startup, alongside her son, that creates custom machine learning models for marketplaces.
Using AI to sell AI, infinite money glitch! /s
“Using a coding co-pilot is kind of like giving a coffee pot to a smart six-year-old and saying, ‘Please take this into the dining room and pour coffee for the family,’” Rover said. Can they do it? Possibly. Could they fail? Definitely. And most likely, if they do fail, they aren’t going to tell you.
No, a kid will learn if s/he fucks up and, if pressed, will spill the beans. AI is, despite being called “intelligent”, not learning anything from its mistakes and often forgetting things because of limitations - consistency is still one of the key problems for all LLM and image generators
AI is, despite being called “intelligent”, not learning anything from its mistakes
Don’t they also train new models on past user conversations?
Considering how many AI models still can’t correctly count how many ‘r’ there are in “strawberry”, I doubt it. There’s also the seahorse emoji doing the rounds at the moment, you’d think that the models would get “smart” after repeatedly failing and realize it’s an emoji that has never existed in the first place.
Chatgpt5 can count the number of 'r’s, but that’s probably because it has been specifically trained to do so.
I would argue that the models do learn, but only over generations. So slowly and specifically.
They definitely don’t learn intelligently.
That’s the P in ChatGPT: Pre-trained. It has “learned” based on the set of data it has been trained on, but prompts will not have it learn anything. Your past prompts are kept to use as “memory” and to influence output for your future prompts, but it does not actually learn from them.
The next generation of GPT will include everyone’s past prompts (ever been A/B tested on openAI?). That’s what I mean by generational learning.
Maybe. It’s probably not high quality training data for the most part, though.
If you bring a 6yo into office and tell them to do your work for you, you should be locked up. For multiple reasons.
Not sure why they thought that was a positive comparison.
This feels like one of those paid fluff pieces companies put out so that smaller ones feel like they’re “missing out”
Gotta love how devs and engineers are supposed to be on the front lines of innovation and progression. But most of the it’s just moaning and calling the next gen dumb. 15 years ago the current devs would be called dumb for using Frameworks amd how it’s cheating since it’s not self written. Do your part and educate and guide the next gen instead of complaining about tech evolving and being used.
If you are wondering how it could possibly be “worth it” the end of the article has this.
The Fastly survey found that senior developers were twice as likely to put AI-generated code into production compared to junior developers, saying that the technology helped them work faster.
So vibes. Vibe coding is “worth it” because people got good vibes.
The research shows that - while engineers think AI makes them more about 20% more productive - it actually causes an approximate 20% slow-down.
AI cannot use logic or reason. Everything it outputs is a hallucination, even if it’s sometimes accurate. You cannot trust anything it outputs.
I’m a senior dev and I want nothing to do with AI. By the time I understand what I want well enough to describe it in a complete sentence or paragraph, I can just write the fucking code myself. I figure it out as I go.
The whole point of having devs under you that is to be able to trust them to get the job done and do it right. You want to be able to delegate tasks to them and not have to peek over their shoulder every five fucking minutes to be certain they’re not making a mess of things.
I seriously doubt AI will ever be able to replace that. Not until they figure out how to make it afraid of fucking up.
As a senior dev I have found AI useful for auto completion (where you see beforehand what it wants to write directly in Visual Studio) and code analysis (as it does find some bugs and can give good hints for code structure). I would never trust it with anything even remotely complex though.
It kinda scares me that people trust it enough for “agent mode”, as giving it direct access to change stuff directly has simply put never worked.
Yes. It’s extremely helpful when I’m doing a refactor and can just go TAB TAB TAB TAB Oops not that TAB TAB done. Saves me a lot of time with the boilerplate, but is very bad at the logic portions.
You do refactoring with auto complete?
As someone right there in the trenches getting hired specifically to clean the slop up, I don’t buy this survey at all and I’d be very suspicious of any “senior dev” that participated in it cause…where are they? I’m not seeing them when I go in to my clients offices because they all got axed. I do see a lot of junior prompt monkeys though.
If I try to get it to do more than predict the next two lines of code it’s gonna fuck something up. A nervously laughable thing I saw at work was someone using a long spec file to generate a series of other files and getting high praise for it. It was the equivalent of mustache templates but slower and with a 30% chance of spitting out garbage. There was also no way to verify if you were in that 30% zone without looking through the dozens of files it made.
senior developers were twice as likely to put AI-generated code into production compared to junior developers, saying that the technology helped them work faster
Perhaps senior devs are more likely to use more granular, step-by-step, controlled prompting. Asking it do write specific functions in specific ways and following specific approaches and conventions instead of just “do me an app, robot bro”.
That’s actually how I am using AI for my work (web dev, pls don’t hate me). If I am stuck or have some tiny function missing for a task I ask AI, check their output - if it’s garbage I continue on my own again or if it’s usable I review the output and continue from there. Also, I think AI can be neat for „rubberducking“ when I am debugging some stupid shit and point me in directions I haven’t looked before.
Similar to how I have found success with it. Is it revolutionary? No, not at all. But it’s a variable sized (big for some use and nonexistent for other use) incremental tool that requires a new skill set to use effectively.
Mix in all of the hype and its easy to see why people are confused and why some get different results.
But surely you test the code and review it, right? That’s how you reinstate trust in what it outputs?
Disclaimer: I’ve never used AI to code, not even copilot.
Based on my coworkers… no.
They get the Ai to write the code, and the tests.
Then hand it over to me to review and test.
Its all overly verbose, does things that are not required or desirable, and insists on re-writing existing code to match its own style.
I hate it passionately.
Damn. 😢
Sounds awful. I would just reject these PRs, dude. Tell them that AI is good for scaffolding and creating a draft, but you gotta maintain the human quality assurance, and that’s not your job, it’s theirs.
You mean rewrite it all from scratch? If you have any kind of standards that is what you end up doing. If you know what you’re doing you do it right the first time and move on. Using AI for coding it like trying to babysit the most inept, inexperienced intern to ever walk the earth. It wastes time and the end result is far worse.
That’s what I’m afraid of, and it doesn’t seem like employers are aware of this in general. Irks me especially as a consultant.
It’ll sometimes do dumb and/or redundant or too complicated shit. Pile up a couple of those and your codebase can get unmaintainable fast.
I find if you give it small chunks and keep an eye on it, it’s great.
I think one of my recent prompts was “Create a procedure that creates an example configuration file with placeholder values. If a config file doesn’t exist on start, give a warning and create the example config.”
It also works great as a replacement for an ORM.
The research shows that - while engineers think AI makes them more about 20% more productive - it actually causes an approximate 20% slow-down.
AI cannot use logic or reason. Everything it outputs is a hallucination, even if it’s sometimes accurate. You cannot trust anything it outputs.
Research shows that - while people think having more people in the household gets the housework done faster - babies actually cause an approximate 100% increase in time spent on housework.
Children cannot use logic or reason. Everything they output is brabbling, even if it sometimes resembles actual works. You cannot trust anything they say. Parents are stupid for having them. (/s)
Developers see AI as a “child” that might need many years to grow up, but it’s still worth all the trouble they go through. It’s an emotional choice, not a rational one.
I keep seeing “vibe coding.” WTF is vibe coding? ELI5
A security nightmare waiting to happen.
Also performance, maintenance, and regression.
Vibe coding is asking gpt for code, copying it into your code environment, then telling gpt about any errors or issues. The problem is that it actually works a significant amount of the time, let’s be generous and say 80%. Another 15% of the time it cannot solve a problem itself. And finally the worst possible outcome is the last 5%, where it creates a seemingly working solution that actually breaks on edge cases or has potential security issues.
One important aspect of vibe coding that I always see that is missing in explanations is the part that vibe coding is the is the generation of code through AI, but without understanding what the code is doing, the effect of this is you are totally dependant on the AI to keep generating the code, so if any error happens you don’t have fucking idea in what to do. If you generate the code using AI and you understood what the AI did, is not vibe coding.
Currently, I write all production code at work without any AI assistance. But to keep up with things, I do my own projects.
Main observation: When I use it (Claude Code + IDE-assistant) like a fancy code completion, it can save a lot of time. But: It must be in my own area of expertise, so I could do it myself just as well, only slower. It makes a mistake about 10 - 20 % of the time, most of them not obvious like compile errors, so it would turn the project into disaster over time. Still, seems like a senior developer could be about 50% - 100% more productive in the heat of the implementation phase. Most important job is to say “STOP” when it’s about to do nonsense. The resulting code is pretty much exactly how I would have done it, and it saved time.
I also tried “vibe coding” by using languages and technologies that I have no experience with. It resulted in seemingly working programs, e. g. to extract and sort photos from an outdated data file format, or to parse a nice statistics out of 1000 lines of annual private bank statements. Especially the latter resulted in 500 lines of unmaintainable Python-spaghetticode. Still nice for my private application, but nobody in the world can guarantee that there aren’t pennies missing, or income and outcome switched in the calculation. So unusable for the accounting of a company or anything like that.
I think it will remain code completion for the next 5 years. The bubble of trying more than next-gen code completion for seniors will burst. What happens then is hard to say, but it takes significant breakthroughs to replace a senior and work independently.
In real code, so after the first week of development, typing really isn’t what I spend most of my time on. Fancy autocomplete can sometimes be right and then it saves a few seconds, but not nearly 50-100% added productivity. Maybe more like 1-2%.
If I get a single unnecessary failed compile from the autocomplete code, it loses me more time than it saved.
But it does feel nice not having to type out stuff.
That’s why all research on this topic says that AI assistance feels like a 20-30% productivity boost (when the developers are asked to estimate how much time they saved) while the actual time spent on the task actually goes up by 20-30% (so productivity gets lost).
I find it also saves a certain “mental energy”.
E. g. when I worked on a program to recover data from the old discontinued Windows photo app: I started 2 years ago and quickly had a proof-of-concept: Found out it’s just sqlite format, checked out the table structure, made a query to list the files from one album. So at that point, it was clear that it was doable, but the remaining 90 % would be boring.
So after 2 years on pause, I just gave Gemini 2.5Pro the general problem and the two queries I had. It 1-shot a working powershell script, no changes required. It reads directly from the sqlite (imagine the annoyance to research that when you never ever use powershell!) and put the files to folders named by the former albums. My solution would have been worse, would probably have gone with just hacking together some copy-commands from SELECT and run them all once.
That was pretty nice: I got to do the interesting part of building the SQL queries, and it did the boring, tiring things for me.
Overall, I remain sceptical as you do. There is definitely a massive bullshit-bubble, and it’s not clear yet where it ends. I keep it out of production code for now, but will keep experimenting on the side with an “it’s just code completion” approach, which I think might be viable.
Yours is pretty much the best-case scenario for AI:
- Super small project, maybe a few dozen lines at most
- Greenfield: no dependencies, no old code, nothing to consider apart from the problem at hand
- Disposable: once the job is done you discard it and won’t need to maintain it
- Someone most likely already did the same thing or did something very similar and the LLM can draw on that, modify it slightly and serve it as innovation
- It’s a subject where you are good enough that you can verify what the LLM spits out, but where you’d have to spend hours and hours to read into how to do it
For that kind of stuff it’s totally OK to use an LLM. It’s like googleing, finding a ready-made solution on Stackexchange, running that once and discarding it, just in a more modern wrapping. I’ve done something similar too.
But for real work on real projects, LLM is more often than not a time waster and not a productivity gain.
That’s completely true; it’s hard for me to judge on a small scale when I won’t (for good reasons) let it touch my customer’s production code.
It makes a mistake about 10 - 20 %
Anecdotally, Copilot does the reverse for me.
Copilot leads me on flights of fanciful code that is absolutely not possible, and the joy turns to tragedy when I find out it lied insidiously about a particular niche function the entire time.
A day will come when I get to know what vibecoding is. Or maybe this word will die out sooner. You never know.
Vibe coding is when you ask a chatbot to code for you, then ask it to fix the errors it generated, and repeat until you can’t find any more errors. Later, someone notices that your application was coded by a chatbot, exploits one of the many security flaws, and steals all your data and credentials.
I thought this was a normal coding. Then how do you call those who heavily rely on google and SO?
That’s regular programming. You still have to fit everything together, so you end up reading the code much more closely. Chatbot enjoyers don’t read it at all.
Copy pasta chefs.
Imposters.
I’m well aware the plural of “anecdote” isn’t “data”, but literally no dev I know (senior or otherwise) thinks this. Give me a junior work with - most of them at least actually learn.
Amen. I’ve tried the vibe coding thing but it’s frustrating because a) too often the AI output has some profound problems and it gets annoying ‘babysitting’ it; and b) I usually prefer the challenge of figuring out syntax and implementation issues myself.
If something is taking too long I’ll ask the LLM. But I feel like if I do this too much my skill set will atrophy and I’ll lose my sharpness. So it’s a balancing act.
But this brings up another wider question: where is the line between “occasionally getting AI help” and “vibe coding”? Surely it’s subjective.
The definition may have changed but I feel like originally it was only vibe coding when the “dev” did not know what they are doing. When some one with little to no programming background is able to build and app on “vibes” alone.
Also applies when the dev could know what they’re doing, but just doesn’t care to.
Well then, I’ve been doing this all wrong.
My first interpretation of vibe coding was to code for fun and personal enjoyment without worrying about industry standards or deployability. More often see with self-taught youth.
I felt like i have been doing this for years before ai became a thing.
I don’t think the two cross, really. A vibe coder asks for a bunch of features and then starts refining the output, fixing bugs and adding features. A developer knows the specific architecture and from years of writing tasks knows how to break work into manageable chunks and uses AI to implement something they have already defined and know where it fits in. The skill to write a good story isn’t far off from writing a good prompt.
I use AI all the time, and every time I hear someone describing vibe coding it makes my skin crawl.
I’d say the use cases of: mundane but time consuming, pointed inquiries or interactive rubber ducking, are all getting AI help. Offloading a design where you don’t have a clear understanding of how it should be done is vibing.
Yeah it’s the same skillset I use with Junior devs except I don’t have the hope AI will grow out of its bad habits
Senior devs love vibe coding because they have the knowledge and skills to recognize and fix errors. They hate it because it makes morons think they don’t need the knowledge and skills to recognize and fix errors.
As a senior dev I hate vibe coding. I can write code an order of magnitude faster than I can review it, because reviewing code forces you to piece together a mental model for something made by someone else, whereas when I write the code myself I get to start with the mental model already in my head.
Writing code is never the bottleneck for me. If I understand the problem well enough to write a prompt for an LLM, then I understand the problem well enough to write the code for it.
I’m a junior and even I feel the same way, reading and understanding someone else’s code not only takes me longer but is far less rewarding than just writing it myself. There’s also the issue as a junior that if I read AI code with issues that maybe I don’t notice or recognise, but it compiles fine, it could teach or reinforce poor practices that I may then put into my own work.
I understand how to turn the results of a select statement into an update statement, but the AI does it a hell of a lot faster.
I find if you give it small enough chunks, it’s easy enough to review. And even if you do have to correct, it’s generally easier to correct than it would be to write it all by hand.
Outside of my own specialty I can people in the software industry bogged down by managing excessive boilerplate. I think this happens most often in web dev and data science.
In my opinion this is an indication that the software tools for those ecosystems need improvement, but rather than putting in the design effort to improve the tools in the ecosystem, these Big Data companies see an opportunity to just throw LLMs at it and call it a commercial product.
Yep this
