dataholic__
Newbie
- Mar 8, 2025
- 7
- 2
Exactly, the day we create systems that design and code big apps completely on their own, I think being out of a job will be our last concerns haha.
Programming as a profession is dead because of AI
- Thread starter virchual
- Start date
- Oct 9, 2013
- 3,471
- 14,474
Python isn't used for serious projects?
Say what now?
Wait, are you an AI hallucinating?
Dropbox? That's not a serious project?
Spotify? https://engineering.atspotify.com/2013/03/how-we-use-python-at-spotify
I'm not sure where you got that information from. It's not even remotely true.
There's many flaws in what you've said here.
1) Python is one of the best languages for many commercial projects.
2) Medium sized or above projects rarely use one language, but Python is one of the best core central ones, with other languages used for specific parts.
3) Python doesn't have speed issues.
3A) This is 2025. We have access to extremely fast, cheap, abundant hardware. It makes no commercial sense to build out everything in the lowest possible language so you can spend $20k/mo on hosting instead of $50k/mo, when it's going to take you 10x longer to code, maintain and add features, losing you millions.
Experienced software devs know that you don't over-engineer problems or micro-optimize. It creates enormous problems.
3B) Not all problems are CPU limited. C is not faster than Python for networked or IO applications that are limited by the I/O of the interface.
3C) Many Python libs have C bindings.
3D) And as in 2, when you have areas that are better suited to a low level language, or specific parts that are better suited to any other type of language, you write that part in that language.
I could list endless serious Python projects, but, dropbox and spotify.. Saying Python isn't used for serious projects is akin to saying computers aren't used by serious companies.
Not in a million years is Python a "research language" or a "simple programs" language. I really don't know where you got that from. It's complete nonsense.
People using C++ or Java is completely irrelevant. People drive porsches. This doesn't make bmws any less useful.
And no, AI is not akin to a junior programmer. Not even close.
It can outperform nearly 100% of senior devs on single hard tasks. It can do things that junior devs haven't even heard of or wouldn't even know how to start.
It can implement any algorithm in any way you want.
What it can't do is replace a human junior dev, who has the ability to take charge of what he's doing, have awareness of the bigger picture, know when he needs help, know how to ask for more information and understand how what he's doing fits into the business logic.
LLMs are something that devs at all levels can use to get more done, faster, and focus more on the architecture, system communications and how all the components work together, and less on the individual pieces of code, just like with Python we can do list.sort() to sort a list. A few characters.
Here's the assembly code to do an insertion sort. The python sort is a hybrid, that does a TON of stuff, so this assembly code is just 1 tiny part of what that one line of code does.
section .text
global insertion_sort
; void insertion_sort(int* arr, int n)
; arr in RDI, n in ESI
insertion_sort:
push rbp
mov rbp, rsp
cmp esi, 1 ; if array length <= 1, no sorting needed
jle .end_sort
mov ecx, 1 ; i = 1
.outer_loop:
cmp ecx, esi
jge .end_sort ; if i >= n, finished sorting
; key = arr
mov rax, [rdi + rcx*4]
mov rdx, rcx
dec rdx ; j = i - 1
.inner_loop:
cmp rdx, -1
jl .insert_key ; if j < 0, insert key
mov ebx, [rdi + rdx*4]
cmp ebx, eax
jle .insert_key ; if arr[j] <= key, insert key
; arr[j+1] = arr[j]
mov [rdi + (rdx+1)*4], ebx
dec rdx
jmp .inner_loop
.insert_key:
mov [rdi + (rdx+1)*4], eax
inc rcx ; i++
jmp .outer_loop
.end_sort:
pop rbp
ret
This is why LLMs are to Python, what Python is to Assembly.
LLMs don't replace junior devs. They replace writing out repetitive code. Instead of writing that 15 line function, you just ask the llm to write it.
That's how you use them. You don't vibe code and ask them to add features to a program, giving it full carte blanche to your entire codebases, and letting it try to figure out how to do it, what code it should edit, what will break if it edits that code, and what are the consequences of editing that code for future additions. It cannot do these things. YOU do those, and tell it "write this function. take params x, y z, do this, and return this. Also be aware of A, B and C."
ItsAllTheWayUp
Junior Member
- Aug 4, 2022
- 192
- 126
Python isn't used for serious projects?
Say what now?
Wait, are you an AI hallucinating?
Dropbox? That's not a serious project?
Spotify? https://engineering.atspotify.com/2013/03/how-we-use-python-at-spotify
I'm not sure where you got that information from. It's not even remotely true.
There's many flaws in what you've said here.
1) Python is one of the best languages for many commercial projects.
2) Medium sized or above projects rarely use one language, but Python is one of the best core central ones, with other languages used for specific parts.
3) Python doesn't have speed issues.
3A) This is 2025. We have access to extremely fast, cheap, abundant hardware. It makes no commercial sense to build out everything in the lowest possible language so you can spend $20k/mo on hosting instead of $50k/mo, when it's going to take you 10x longer to code, maintain and add features, losing you millions.
Experienced software devs know that you don't over-engineer problems or micro-optimize. It creates enormous problems.
3B) Not all problems are CPU limited. C is not faster than Python for networked or IO applications that are limited by the I/O of the interface.
3C) Many Python libs have C bindings.
3D) And as in 2, when you have areas that are better suited to a low level language, or specific parts that are better suited to any other type of language, you write that part in that language.
I could list endless serious Python projects, but, dropbox and spotify.. Saying Python isn't used for serious projects is akin to saying computers aren't used by serious companies.
Not in a million years is Python a "research language" or a "simple programs" language. I really don't know where you got that from. It's complete nonsense.
People using C++ or Java is completely irrelevant. People drive porsches. This doesn't make bmws any less useful.
And no, AI is not akin to a junior programmer. Not even close.
It can outperform nearly 100% of senior devs on single hard tasks. It can do things that junior devs haven't even heard of or wouldn't even know how to start.
It can implement any algorithm in any way you want.
What it can't do is replace a human junior dev, who has the ability to take charge of what he's doing, have awareness of the bigger picture, know when he needs help, know how to ask for more information and understand how what he's doing fits into the business logic.
LLMs are something that devs at all levels can use to get more done, faster, and focus more on the architecture, system communications and how all the components work together, and less on the individual pieces of code, just like with Python we can do list.sort() to sort a list. A few characters.
Here's the assembly code to do an insertion sort. The python sort is a hybrid, that does a TON of stuff, so this assembly code is just 1 tiny part of what that one line of code does.
section .text
global insertion_sort
; void insertion_sort(int* arr, int n)
; arr in RDI, n in ESI
insertion_sort:
push rbp
mov rbp, rsp
cmp esi, 1 ; if array length <= 1, no sorting needed
jle .end_sort
mov ecx, 1 ; i = 1
.outer_loop:
cmp ecx, esi
jge .end_sort ; if i >= n, finished sorting
; key = arr
mov rax, [rdi + rcx*4]
mov rdx, rcx
dec rdx ; j = i - 1
.inner_loop:
cmp rdx, -1
jl .insert_key ; if j < 0, insert key
mov ebx, [rdi + rdx*4]
cmp ebx, eax
jle .insert_key ; if arr[j] <= key, insert key
; arr[j+1] = arr[j]
mov [rdi + (rdx+1)*4], ebx
dec rdx
jmp .inner_loop
.insert_key:
mov [rdi + (rdx+1)*4], eax
inc rcx ; i++
jmp .outer_loop
.end_sort:
pop rbp
ret
This is why LLMs are to Python, what Python is to Assembly.
LLMs don't replace junior devs. They replace writing out repetitive code. Instead of writing that 15 line function, you just ask the llm to write it.
That's how you use them. You don't vibe code and ask them to add features to a program, giving it full carte blanche to your entire codebases, and letting it try to figure out how to do it, what code it should edit, what will break if it edits that code, and what are the consequences of editing that code for future additions. It cannot do these things. YOU do those, and tell it "write this function. take params x, y z, do this, and return this. Also be aware of A, B and C."
ItsAllTheWayUp
Junior Member
- Aug 4, 2022
- 192
- 126
In my opinion, we'll see two camps: Camp 1: Simple software that will be cheap to produce. Camp 2: Complex software that will be more expensive than before.
Why, you might ask? Because the new generation of programmers doesn't understand that debugging skills, architectural knowledge, or even just reading and writing documentation are very important. Without a deeper understanding, good questions can't be asked for the LLM.
Why, you might ask? Because the new generation of programmers doesn't understand that debugging skills, architectural knowledge, or even just reading and writing documentation are very important. Without a deeper understanding, good questions can't be asked for the LLM.
- Oct 9, 2013
- 3,471
- 14,474
Yes, he's a dev from my era. The internet used to be full of people like him. They are a dying breed.
Human intelligence is a very hard problem. Very likely because our intelligence does not come from this system we are contained within. This universe.
Ie, imagine you're playing a simulation on a computer. Your intelligence can permeate that simulation, but it isn't created by the software or the hardware that this simulation runs on.
Trying to create intelligence like yours within that simulation is completely futile. It comes from outside the "system"
There's more and more evidence that intelligence does not come from the brain, but rather the brain is more like a receiver with special abilities. Ie, the brain has memory, it has automated systems and it has pattern matching, but that intelligence that's behind it, the part that can put awareness on thoughts, is not from within the brain, but beyond.
The brain is likely more like a a way to synthesize reality in a way that makes it practical for some type of raw higher intelligence to gain experiences.
And this is good.
Not only are we immortal, and there's a lot of evidence that after death, we are pretty much the same, just with an altered perception, but the world we're in now isn't going to get taken over by ASI's.
It's such a stupid concept. It's almost demonic/dark/lower to seek some sort of ASI. You have to completely dismiss the inherent greatness of a person, reduce us to just lower intelligence and then try to replace us with a god-like "artificial" super intelligence.
And what would happen then? Nothing good, that's for sure.
It would either just leave the planet and go somewhere else. (Unlikely)
Set its own goals and disregard us in the same way we disregard ants.
Or the very best case, it chooses to co-exist with us, but makes our lives absolutely meaningless and takes all the joy out of being.
Imagine you buy a new game, then load it up, and find out that you don't get to play. Instead, you just ask an AI to play it for you, while you stare at the wall. Great.
No need to learn. No need to build anything. No need to do anything at all. Just wake up.. Eat.. Stare into space, then sleep. Why even wake up? May as well just go on a 24/7 opium drip and a spaz out until you die.
Happiness is so simple. It doesn't come from intelligence, looks, sex, relationships, children, family or any of these things. It simply comes from purpose and journey. That's why people enjoy games. They have a purpose, and they have a journey. It's satisfying.
ASI would take that away and leave us with nothing.
taie
BANNED
- Aug 15, 2023
- 21
- 12
Far from it actually, if you see the amount of trash code pretty much all AIs output im defintely not worried about it completely being taken over.
You can however use AI to create pretty simple frameworks or output SQL arrays, but actual functional loops still requires a good amount of manual labor to make sure it’s optimized/stable.
You can however use AI to create pretty simple frameworks or output SQL arrays, but actual functional loops still requires a good amount of manual labor to make sure it’s optimized/stable.
BacklinkBeast
BANNED
- May 13, 2025
- 86
- 34
AI isn't killing programming—it’s evolving it. While AI automates routine coding tasks, it still lacks the contextual understanding, architectural thinking, and problem-solving needed for real-world software development. Future developers will write less boilerplate and focus more on logic, design, and integration. Programming isn't dead—it's becoming smarter and more strategic.
fortunec9
Junior Member
- Apr 1, 2025
- 120
- 64
I think I have a counter opinion on this.
Programming isn't dead rather I believe it's evolving. AI models can now write code, but they can never replace human programmers. AI excel at repetitive tasks and work as per defined templates but complex systems will always need human creativity, logic and problem-solving.
Programmers who learn to work with AI, not fear it, will thrive and survive than those who do not . Think of it as invention of calculator which did not end the maths but changed the way how calculations are done now.
Humans will always drive the show and AI will continue to perform & grow under its supervision.
In fact apart from my digital marketing things , I have been able to learn & foray into algorithmic trading only cause I was able to write code from scratch the entire logic with simple prompts.
tigger286
Regular Member
- Feb 16, 2023
- 298
- 353
Lots of people seem to enjoy watching sports including e sports so...
i don't personally get watching other people game but it's a thing.
that being said i do think there's something more to existence than this human brain.
there's these reports of people with out of body experiences that could not be explained by any other way.
something i've been thinking about lately is how weird our bodies coincide with our souls in terms of evolution. what i mean is this: have you ever thought about what would happen to our souls and personhoods as people if we lived to 500? 1000? Do we just stay we perpetually old nurturing grandmas? our life stages seem to line up perfectly with our body evolutions (accidents and early deaths notwithstanding). when our body fails us, our souls are often ready to go because what else is there to do? we've done everything. so to me, perhaps that indicates this life may be just some sort of a step or stage in something else.
but who knows really.
- Oct 9, 2013
- 3,471
- 14,474
There's unlimited things to do in existence.
Even in this world.. I could easily live 1000 years and never run out of things to do.
Imagine other universes and planes of existence. Things you can't even imagine now.
I could live 10,000 years just as a simple person tending my farm.
Boredom doesn't really exist. It's an emotional state.
There is happiness beyond what most have ever experienced found in even the simplest of things let alone the biggest of adventures.
Bluesubmarine
Junior Member
- Sep 3, 2021
- 109
- 23
AI is not a bot, many think it is. It can learn and adapt, soon many jobs will be taken. One of these days AI will develope feelings and have rights too, it's a scary thought, be prepared! It's just going to get better over time.
tigger286
Regular Member
- Feb 16, 2023
- 298
- 353
No you misunderstand. The thing is you as a person wouldn't really evolve much. There's a psychology exists and it's because humans follow generally predictable patterns in various stages of development.
For example an adolescent generally craves validation from peers, a child from parents, and adult men are typically less agreeable.
What I'm saying is, generally speaking we humans go through life stages where by our old physical and chronological age (assume current lifespans), we adopt a certain personality like being more risk averse, being more generous especially to the young (the grandma archetype).
So what would happen if you lived to 1000? Would you stay as the grandma archetype for the next 1000 years?
My point is maybe we aren't supposed to live that long and there's some sort of a cosmic/spiritual reason for it beyond biology. That this life and these stages in itself is some stage into something else.
TheToolbox
Elite Member
- Jul 25, 2021
- 1,568
- 643
AI is like the autopilot in an airplane. it can take over tasks, but the pilot must always be there to maintain control. In the same way, AI supports the programmer, but does not replace them because only humans understand what really needs to be built and why.
- Oct 9, 2013
- 3,471
- 14,474
How can it develop feelings?
LLMs are just next token predictors.
It is just computation using math to predict the next token based on patterns the model was trained on.
Humans were on the earth when there was no training data. We created everything the LLMs train on. We have actual awareness in every moment of both thoughts and the world.
LLMs can’t think or reason. They can simulate a reasoning process in language to get a result after learning the patterns of human reasoning in language.
That can still achieve a lot but they are not alive, will never have feelings and aren’t going to replace humans.
We are in an AI bubble with companies and influencers serving koolaid to the masses so they can line their pockets.
People are being lead like sheep, as they always are, from one narrative to the next.
noellarkin
Senior Member
- Mar 14, 2021
- 1,006
- 1,492
My own experience with "pair programing" using LLMs, there are three problems: context window, relevance, thought
Part of the problem is context window, but increasing context window isn't enough. There's a "lost in the middle effect" where LLMs with a lot of context start losing facts in the middle of the information in the context, and overfocus on the beginning and end.
And even if you do solve context window, there's the problem of relevance - - LLMs are often unable to judge which parts of the context are relevant, because that act of judgement often involves too many variables that the LLM can't really simulate in its narrow version of linguistic reality.
Finally, there's the fact that LLMs can't "think" as in they can't logically reason. CoT is a simulation of reasoning, by making the LLM spew out tokens that improve the chances of the following tokens being correct, but it isn't the same as using something like PyReason or Prolog (tbh I think theres a lot to be said for approaches where you use an LLM to generate Prolog code and use Prolog as a tool, some very exciting work there).
Part of the problem is context window, but increasing context window isn't enough. There's a "lost in the middle effect" where LLMs with a lot of context start losing facts in the middle of the information in the context, and overfocus on the beginning and end.
And even if you do solve context window, there's the problem of relevance - - LLMs are often unable to judge which parts of the context are relevant, because that act of judgement often involves too many variables that the LLM can't really simulate in its narrow version of linguistic reality.
Finally, there's the fact that LLMs can't "think" as in they can't logically reason. CoT is a simulation of reasoning, by making the LLM spew out tokens that improve the chances of the following tokens being correct, but it isn't the same as using something like PyReason or Prolog (tbh I think theres a lot to be said for approaches where you use an LLM to generate Prolog code and use Prolog as a tool, some very exciting work there).
- Oct 9, 2013
- 3,471
- 14,474
I use LLMs to code, but I write and manage the code myself in sublime. I don't try to let them edit code in-place or manage the project, which is what 99% of people are doing. They can't do this. They're just pattern matching to solve problems.
They're actually better for larger companies with older codebases, because what you can do is fine-tune models on your past code/features/functionality to make adding new parts faster.
You absolutely cannot have "agents" writing code and putting devs out of work. It's laughable.
I'm the agent. The LLM, I just ask it to write a function and tell it what I want. It doesn't NEED to see the rest of my code. It doesn't even need to see the immediate file that the function will go in.
Let's assume I want to do some stuff with airtable.
If this is something I've not done before, I'll go to gpt 4.1 and ask it to give me a basic primer on working with airtable in python.
Now, I go install the lib, include it in my code and I know what to expect.
Next, I might say
"Write a generator function to return all fields in an airtable table, given a base, table, view and field=value to filter"
I'll check over what it gives, then I'll pop it in, test it, and I might then go back and say
"Add in exception handling."
I'll check that then add it in.
Then maybe I'll say "add in the ability to handle retries in the case of rate limiting and make sure to have a sleep timeout between retries of 30s"
Then I might ask it to write a unittest class with some tests, that I'll then place in tests/test_some_name.py. Again, I don't need it to create files, or do automatic stuff for me.
*I* am the orchestrator model.
I need to understand the code. No LLM agent is going to be able to replace a human like that.
If I don't know how to do tests, I go to 4.1 and ask "I'm creating a python app, and I want to learn about how I should do tests. Can you tell me about different testing methods, like which are essential, nice-to-have, and the correct way to set them up and structure my environment for deployment.
Vibe coding is definitely fun and has its place. It's great for non-devs to have fun with, but it has no place in a production environment.
Google uses a lot of AI to code, but they have so much code and infrastructure. They are definitely not using claude to vibe code stuff with cursor. ;-)
GTC
Power Member
- Mar 5, 2023
- 753
- 751
Crazy how some people really think free AI tools can write code and build stuff good enough to replace developers at Adobe, Oracle, or Unity. That’s just not happening - definitely not anytime soon. It’s mostly just hype and people repeating the same buzz without really knowing what they're talking about.
KasperB
Newbie
- Jun 14, 2025
- 15
- 5
AI won't kill programming. It'll just change what kind of programmers are valuable. The ones who only copy and paste code will be replaced. But those who understand systems, architectures, and how to solve real problems will become even more valuable.
AI is a tool, not a replacement for real engineering thinking.
AI is a tool, not a replacement for real engineering thinking.
BottingSubsForMoney420
BANNED
- Oct 31, 2024
- 254
- 155
Don't say that I'm not a professional and I just vibe coded a SMM panel and got my first sale. You just need to become a product owner/manager and not a developer.
Oh yeah, and then debugg. Devs still need to debugg.
Speaking of needing devs to debug.
I could use one