..What about people who love to code?

[-]

ivancea@reddit

"But what about people that love to knit??!! If everybody now works on a factory making clothes 100x faster, what will they do!!!"

I'll be blunt: when a skill has no value, you do it at home, like your grandma does. Welcome to engineering, evolution and progress, where your last 20 years of life are important to you, but not to the world, and the world keeps advancing like it has been doing for the last millennia.

But, seriously. Ask your grandmother/father about how the world changed before you were born, and how humans adapt. Because seriously, this is post #652 this month with this topic, and this is the thought #267733676 about automating this century. Maybe it's time to learn from the past, don't you think so? That's what books are about

[-]

boring_pants@reddit

Maybe it's time to learn from the past,

Turns out that's easier said than done.

Because you can't just say "knitting went away so therefore coding is going to go away". That is not "learning from the past".

You're right that if and when a skill has no (commercial) value, it'll be confined to the home and to hobby projects.

But there doesn't seem to be much evidence that coding has, or will ever have, no value. That's just the outcome you want. Wanting something does not make it true.

[-]

ivancea@reddit

Turns out that's easier said than done.

Well, nobody is paying you to do easy things, right? I'm sure you, as an engineer, work in solving hard problems. It's good to apply the same to you personal life!

But there doesn't seem to be much evidence that coding has, or will ever have, no value

First, agents started to do good code just a few months ago. Second, there is evidence already: agents have been doing quality code since a few months ago. Plus, a hundred other dumb tasks.

That said, nothing of that matters. We're engineers, we work with the unknowns, and prepare to all the potential outcomes. Right? That's our trying, our job, and for many of us, our passion. If something is broken, fix it, instead of doom-talking. It's never constructive

[-]

boring_pants@reddit

First, agents started to do good code just a few months ago. Second, there is evidence already: agents have been doing quality code since a few months ago. Plus, a hundred other dumb tasks.

I am starting to notice an interesting trend among AI evangelists. You guys are terrible at data-driven analyses.

"I saw an LLM write some code, therefore coding has no value" is not a valid argument. The conclusion does not follow from that premise.

We know agents can write some code to solve some problems. We don't know that agents can write any code to solve every problem. We also don't know that they will ever get to that point.

We're engineers, we work with the unknowns, and prepare to all the potential outcomes.

Is that why you've chosen one outcome that you regard as unconditionally true, no matter the evidence or lack thereof?

[-]

ivancea@reddit

I guess you would start this discussion, and it will lead nowhere. So I'll be quick: I've been working with AI, like all my company, and other companies I know, for months. It works. You won't trust it, but whatever. You want data-driven, but you simply don't want to read about real data about the topic, and fall into survivor bias (the people you read shitting on AI are the few that didn't get the expected results).

Anyway, calling somebody "evangelist" because they use a tool is already far under the level of technical discussions I want to have.

We know agents can write some code to solve some problems. We don't know that agents can write any code to solve every problem. We also don't know that they will ever get to that point.

Agents are proven to be able to do most work. +80%™ of the code, for sure. And add you want "data", I have 2 jobs, one in a team tech company using it, and another running model evals. Where I get to get actual data on what models can do, both bad and good models. And even see bad models code currently in most cases. And they're not easy tasks. (Feel free to ignore the facts of you don't like them, I'm good)

Is that why you've chosen one outcome that you regard as unconditionally true, no matter the evidence or lack thereof?

For somebody so data-driven, you're so eager to fall into missing-data fallacies. Nothing of what you said there is correct. But you already know that, but I suppose trolling is easier than debating

[-]

boring_pants@reddit

You want data-driven, but you simply don't want to read about real data about the topic

If you can point me to some reading material on how AI can handle literally all coding problems that humans are currently responsible for then throw me a link.

If not, then I don't think the problem is my unwillingness to read.

(the people you read shitting on AI are the few that didn't get the expected results).

This entire thread is about AI making ALL human coding redundant. So if there are people who don't get "the expected results" then that disproves the entire premise. Then we can move into a different discussion of how much coding will be made redundant. But we've proven it is not "all".

Agents are proven to be able to do most work. +80%™ of the code, for sure.

"Proven", eh? Is this like, a rigorous mathematical proof? Because it sounds anecdotal to me.

Disregarding the word "proven" there, I'm not sure if you are aware but 80 is actually a different number than 100. It's a little known fact, but an important one.

If AI can only replace 80% of coding then coding still has value.

I'm sorry that pointing out this flaw in your argument so clearly hurt your arguments.

If you want to make the claim that humans writing code has no value then you need to show that there are zero use cases for humans writing code. You didn't.

Perhaps you should talk to chatgpt about it. I hear it's an excellent therapist.

[-]

ivancea@reddit

If you can point me to some reading material on how AI can handle literally all coding problems that humans are currently responsible for then throw me a link.

A shame nobody is talking about that.

If AI can only replace 80% of coding then coding still has value.

Aand nobody is denying that either. Keep going.

This entire thread is about AI making ALL human coding redundant

It's not, and you're free to read the first comment if you want.

But we've proven it is not "all".

You should read the definition of "prove". And we're talking about the future. So, again, your data-driveness falls down the cliff.

"Proven", eh? Is this like, a rigorous mathematical proof? Because it sounds anecdotal to me.

It's world-level empirical evidence. You should check that too. And for God's sake, stop reading reddit and go work to different companies. You're talking as somebody that tried AI once and now you're not only outdated, but unwilling to learn.

I'm sorry that pointing out this flaw in your argument

Insert meme: Which argument??? Because nobody was talking about that. Are you reading?

If you want to make the claim that humans writing code

So, all your comment was based on your assumption of that I was saying that LLMs will replace 100% of the coding, huh? All that comment to talk about... Nothing real.

Mate. Stop there, and talk with engineers from different tech companies. You'll find a lot

[-]

newEnglander17@reddit

I don’t really care either way as if I have to adapt so be it, but for three years now people have been posting on here “started to do good code just a few months ago” so apparently we’re eternally always just a few months after things got good and we’re always six months away from 95% of software engineers won’t be needed. Comments like these make it hard to believe anything.

[-]

ivancea@reddit

for three years now people have been posting on here “started to do good code just a few months ago”

I don't really care about posts, I try it. 2 years ago... You had copilot.. You had an autocomplete with knowledge for questions.. But not a valid agents. And this was known. Nobody said 2 years ago that they weren't touching code anymore. Not even 1 year ago.. Not a senior at least.

Comments like these make it hard to believe anything.

Comments on LinkedIn are for non-technical people, not for you.

Anyway, this isn't about what people say; nobody cares about that. This is about the reality of the current technology. You can try yourself, that's all that matters

[-]

big-papito@reddit

Wait until you've done this for 20 years. People at my mark start getting really jaded. I've seen those posts even before LLMs charged in.

I just want to solve problems and look like a rockstar. I KNOW I will figure it out eventually, but my time is the price of gold in the second leg of my life. I want to do a lot but I do not have the time I had in my 20s and 30s.

Also, engineers these days are being saddles with more and more tasks, it can qualify as abuse. Companies keep squeezing more out of us while the pay has been the same for effectively a decade. It's hard to break the $200K ceiling if you do not migrate to Big Tech.

Now we are full stack, we are security experts, and we also need to know every goddamned cloud toy AWS has to offer and know how to use it. There are not enough seconds in a day for all this.

That said, when you are young, code away, struggle, learn. It will pay off later. Those who take the easy way out will suffer. I am blessed in a way that I did not have a choice but to do it the hard way, like those students in Kung-Fu movies.

At first, I freaked out as well. "Do I just review code for a living now"? Yes and no. You can just do that if you want to, but trust me - slaving over every line of code, constantly looking up documentation and testing each change will get old once you have done it 15 billion times in your life.

[-]

Heavy_Discussion3518@reddit

22 years in and I started to feel this way about 5 years ago. AI coding harnesses flipped the script for me. I can get so much done, so quickly, at about 95% the quality. I don't write nearly as much code, and that allows me to focus on the bits I've come to enjoy far more in my mid career - abstractions, architecture, and just plain solving problems.

[-]

w4nd3rlu5t@reddit

Can you describe your workflow? I’m still pretty much just using Claude code and haven’t really started on the agentic stuff yet.

[-]

big-papito@reddit

Watch this video and one hour will save you months of experimentation. Matt did the hard work: https://www.youtube.com/watch?v=-QFHIoCo-Ko

[-]

kchamplin@reddit

Excellent video, thank you for posting

[-]

w4nd3rlu5t@reddit

Thank you!

[-]

arbitrarycivilian@reddit

Genuine question: do you think AI is incapable of abstractions and architecture?

[-]

Heavy_Discussion3518@reddit

Not at all. I think LLMs can get there. Trouble is doing it at scale, context windows start to get big, and with the quadratic memory requirements today, it's a hard ceiling for now.

[-]

PmMeCuteDogsThanks_@reddit

Yes, I’m about 20 years in and I feel the same. You really need to work on your setup, AI harness, code structure and so on. But I can accomplish so much with so little effort it’s laughable. I’ve done more in 3 months than what I did last 2 years.

And it’s not only big systems, I can finally do all those personal projects, ”when I get a few weeks of downtime”. Code is so cheap now. I’ve literally developed more or less a personal DE/WM for Linux that works just the way I want it. Completely unusable for someone else, I’ll never bother creating a public repository. But it just fits how I think about interacting with a computer.

[-]

yolobastard1337@reddit

I find it has also breathed some life into problems that I am competent at, but dread:

Fixing up spaghetti code I inherit. I can pick between not thinking too much about it OR refactoring nicely. I can think of old code where I would both need to strain my little brain too much, AND still end doing exhausting little surgical incremental changes.
Problems I have solved before, in different languages, frameworks or contexts.

[-]

ksceriath@reddit

That is some useful advice. But orgs seem to think that SWE are folks that love to deliver (probably right from their pov).. whereas I believe most folks like the appeal of hand-crafting something logically coherent. So the juniors aren't really getting a choice here, the tools are getting shoved down their throats. I sometimes wonder that the mental muscle I built with failures over the years.. there would be an entire generation of engineers who don't go through that process.

[-]

big-papito@reddit

A new idea that is being surfaced now is to have AI in Tutor mode. You do the work and it's basically helping you like a supercharged StackOverflow search.

If done right, I can see how one can speed-run the learning process. Then again, we humans love to pop a pill to take the easy way out and avoid the diet.

[-]

ksceriath@reddit

That is definitely very satisfying way to use it. And you are in control of the quality of the code as well. But I can't say how many orgs will take that approach.. most won't prioritize your learning at the cost of churning out fewer LOCs..

[-]

big-papito@reddit

I am afraid to say, you won't get to do this on the job - on your own and in school.

[-]

GuyWithLag@reddit

Also, engineers these days are being saddled with more and more tasks

Oh man, I remember the time when Sun Microsystems recommended half a dozen distinct roles for a JavaEE application...

[-]

vitek6@reddit

In my opinion we are FAR, FAR, FAR from a state where no coding is needed. Don't believe in marketing bullshit.

[-]

Blueson@reddit

I am honestly at the point where I doubt people who are getting sold on this were very strong developers to begin with. I keep hearing the models are improving, but all the code I see has the same shitty quality as previously.

I have at most been able to produce an acceptable QT dialog by vibe coding. Everything else I try is just unoptimized, unorganised or tons of fluff that does nothing?

Somebody will tell me my prompts suck and it's a skill issue. It's honestly a skill issue to think prompting is hard, it's literally just writing. Competent developers were doing this before and will keep doing it after AI.

[-]

cscqtwy@reddit

It's hard to diagnose what's going wrong from afar, but it is doing most of my coding at this point. Are there some parts I still do myself? Of course, but they're getting rarer.

The turning point, fwiw, was something like late last fall. It hasn't been good for very long. Also, tooling matters a lot. My setup allows it to see compiler error messages, run tests, inspect all of my dependencies, etc. It's much worse if it doesn't have a reasonable way to see whether the thing its doing is working, and to understand why it isn't.

Somebody will tell me my prompts suck and it's a skill issue. It's honestly a skill issue to think prompting is hard, it's literally just writing.

Ah, I see the problem. Writing effectively is hard. The LLMs are far less valuable to people who are strong developers but weak at communicating.

Nobody reasonable trusts it enough to let it pass without review.

Yes, definitely. We're doing far more code review than before.

Only way you get developers competent enough to review this slop to begin with is to have them write code by themselves. There is still need and room to manually code.

I'm worried this might be the case, but not actually sure. You can get good at most things by doing them directly. It feels like this might be different, but I can't explain why it would be.

[-]

it_happened_lol@reddit

My man, you have your blinders on. I don't know what you're doing wrong, but this isn't 2025. AI is not a mass hallucinating slop machine. My advice to everyone upvoting this is to do some self reflection, if not for any other reason than self preservation.

[-]

Blueson@reddit

Send me a repository that's at least 80% vibe coded and solves a complex problem larger than a landing page or another "plan your day activities" app and let me review it.

[-]

it_happened_lol@reddit

I am not suggesting "vibe coding" is an effective method of creating software, I am disagreeing with sentiment like this.

I keep hearing the models are improving, but all the code I see has the same shitty quality as previously.

I have at most been able to produce an acceptable QT dialog by vibe coding. Everything else I try is just unoptimized, unorganised or tons of fluff that does nothing?

This shows an extreme lack of knowledge and experience with the tools. I am not suggesting vibe coding production software.

[-]

Blueson@reddit

Then what experience do you suggest we are missing?

[-]

it_happened_lol@reddit

Are you using the newest models (Opus 4.7 or GPT 5.5)? Are you using Codex or Claude Code? Is your SDLC process well defined? Is your CI/CD process well defined? Do your tickets have well articulated requirements? Does your codebase have good test coverage? Do you have coding standards, linting, etc?

I have not seen bad code produced in those conditions. Maybe that's where the disconnect is? The models literally converge on the best code they can produce with the guardrails you provide to them. The code produced should look and work like the rest of the codebase. Adding new features are typically flawless. Adding multi-team features is just a matter of exposing the correct context (I.e. we'll use GitLab MCP + Jira MCP + Slack MCP) to literally give it every applicable cross-domain data it can) and it will usually do a better job than most humans in what it proposes. Some things it will get wrong because not all context can be found digitally.

If you are doing something novel, that the A.I. doesn't have the context for, it's not going to excel.

I will say that I've some teams at our company struggle and here's why:

- Some teams have really poor requirements gathering and the teams are mostly mid-level developers. The code in their codebases was also already poor quality. What is the output? More garbage.

- Our C suite has some proficient technical leaders who are on the opposite end of the spectrum. They are incredibly component in AWS, CI/CD, product development, basically everything but coding. They have actually done pretty well, but unfortunately they find themselves backed into corners where, since they don't know how to code, the app meets the requirements of a complicated domain but the A.I. has not really created the right abstractions. It's hard to know what needs to be re-written (my team is not taking over this project, but some other dev team likely will, so I'm interested to see how it pans out)

[-]

MiniGiantSpaceHams@reddit

You're the only one bringing up vibe coding. I don't think that anyone who works on real projects thinks you can straight vibe code them.

[-]

Jiuholar@reddit

This subreddit confuses the shit out of me on this topic. I've written like 100 lines of code by hand since February. My workplace is in the middle of an enormous shift in the way we build software... What planet are these people on.

[-]

vitek6@reddit

Good luck with that in the future.

[-]

ItSeemedSoEasy@reddit

Are you actually reading the code it produces? Line by line?

Have you seriously not seen it, day-to-day, regularly produce:

- Tons of pointless error handling

- Ginormous, do everything, functions

- Completely miss really obvious code reducers like recursions or sub-classes, etc.

- Duplicate functions that have subtle differences

- Make silly assumptions about sorting, filters, timezones, default sort field, etc., all the little minutiae decisions you don't even realize you're making, and hardcoding some truly bizarre stuff

- Completely cheat at tests by making an "everything passes" function

- Use incredibly out-dated techniques and code, unless you constantly badger it or keep context short

And all the coding agents do this in different degree. Like last time I used Codex (we have CC at work), it made an absolute ton of assumptions.

And that's putting aside the incredibly variable quality even day-to-day, one minute it's fine, the next minute it's like it's had a lobotomy.

[-]

ithinkilefttheovenon@reddit

In my experience these types of slop come from poorly scoped requests. Just like we break down Jira stories into smaller, more manageable chunks, we can do the same with asking LLMs to create code.

- error handling: define specific requirements around error handling and the tell llm not to write code not specifically asked for

- ginormous functions: ask the llm to analyze the ginormous function and break it up in a sensible way

- duplicate functions: ask the llm to call a unified function and move the subtle differences outside the function

- silly assumptions: define the requirements, don’t leave them for the llm to assume

- cheating at tests: give the llm instructions on how to define the tests and what to test

I feel like the problem is one of expectations. We’ve seen examples of simple prompts like “Write a tetris game” turn into complete projects. But when you dig into that code is going to have a lot of problems. The same is true if you gave an engineer the same task. What really needs to happen is to define the process using skills, define the requirements of the project, and be willing to iterate. The LLM needs direction just like a programmer does.

[-]

Jiuholar@reddit

Are you actually reading the code it produces? Line by line?

yes, for now. we're looking for ways to change this but haven't found anything yet.

Have you seriously not seen it, day-to-day, regularly produce:

yes to all of these, but not day to day. just once. then we handle the scenario with a tweak in system prompt, context or static analysis/linting tooling.

And that's putting aside the incredibly variable quality even day-to-day, one minute it's fine, the next minute it's like it's had a lobotomy.

less of a factor when you are managing your own models. can lock versions, weights, compute, system prompts etc. so you're not at the mercy of whatever anthropic/openai feels like doing that day.

[-]

boring_pants@reddit

The planet known as "a different employer than yours". Not every company makes the same decisions. Not every company makes decisions at the same speed either.

[-]

Jiuholar@reddit

I mean, sure, but the knowledge that you can create non-slop code with AI at speed and scale can and has been freely discovered by people regardless of what their employer is doing.

[-]

boring_pants@reddit

Has it? How do you know it's not slop until you've had to maintain it? It'll take years to properly evaluate the health of code bases written by AI. We know AI can write code that solves a subset of problems and which looks fine at a glance.

We do not know that all that code that needs to be written can be written by AI. And we don't know that the code being written by AI will still be maintainable three years from now.

You're jumping to conclusions. (And perhaps engineers who jump to conclusions before they have all the relevant data are also more easily replaced by AI?)

[-]

Jiuholar@reddit

How do you know it's not slop until you've had to maintain it?

i do maintain it.

It'll take years to properly evaluate the health of code bases written by AI

this is a true statement of code written by humans, too.

We know AI can write code that solves a subset of problems

yes. thats exactly how we're using it.

which looks fine at a glance

humans are still reviewing code, we have a growing set of custom static analysis tooling to support the volume

We do not know that all that code that needs to be written can be written by AI. And we don't know that the code being written by AI will still be maintainable three years from now.

You're jumping to conclusions. (And perhaps engineers who jump to conclusions before they have all the relevant data are also more easily replaced by AI?)

would you like to try again and respond to what i've actually written?

[-]

boring_pants@reddit

i do maintain it.

I'm sure. But you haven't been maintaining AI-written code for long.

this is a true statement of code written by humans, too.

Of course. The difference is we have 50+ years of experience with the maintainability of human-written code.

We know AI can write code that solves a subset of problems

yes. thats exactly how we're using it.

Cool. Good for you. But if your claim is that AI will replace all human coding then that's not sufficient.

would you like to try again and respond to what i've actually written?

Okay. You wrote that AI will replace all human coding.

You also wrote that we only know AI can solve a subset of the problems we try to solve through coding.

So once again, you have not presented sufficient evidence for your claim. Once again you are jumping to conclusions.

I have responded to what you wrote.

[-]

Jiuholar@reddit

You wrote that AI will replace all human coding

where did i say that?

[-]

boring_pants@reddit

Fair, you didn't. But you replied to a discussion about whether AI will replace all human coding with some strong opinions on the power and ability of AI, and it is pretty natural to connect the two and presume that you agreed with OP's premise.

If you didn't, then cool, I apologize for reading too much into your comment.

[-]

Jiuholar@reddit

i mean, fully replacing the physical act of typing code on the keyboard, character by character, does seem like a real possibility to me. ive personally seen a significant reduction in manual code writing since february. i interpreted OP's post to be literally "no physically typing code" not "programming as a profession no longer exists". i certainly don't believe the latter.

but the fact of the matter is, the technology is a significant force multiplier. used wisely, a single engineer can produce the same output as five, with no penalty to quality.

the key is the process, education and tooling around it. more often than not, the example people point to in order to support the claim its "just hype" amounts to basically just dumping a jira ticket into the prompt and hoping for the best, and then being shocked when the outcome is slop. its like caller Kubernetes trash and useless after trying to scale your app with a docker run bash script and it failing.

[-]

ptrnyc@reddit

In my experience AI is actually much better at reviewing code than humans.

[-]

itdoesntmatterrly@reddit

Either those people are writing code only for hobby/personal project or they are in denial idk.

[-]

SarmackaOpowiesc@reddit

Github just announced that they are going start limiting tokens and all the juniors and mid levels on my team collectively started freaking out.

I just laughed and kept producing more & better deliverables without relying on an over hyped tool.

The bottleneck was always testing the code, reviewing it, and figuring out what the requirements actually were. Writing the code was never the problem in the first place.

[-]

Jiuholar@reddit

The bottleneck was always testing the code, reviewing it, and figuring out what the requirements actually were

Yeah, these are the main components of the enormous shift. We can now produce more code than either side of the pipe can handle, so we're working on increasing the flow in both directions.

It was overhyped. It still is. But it is much better than it was a year ago. If you genuinely still believe it's only hype, I implore you to give it another ago, and approach it with a high degree of engineering rigor and intellectual curiosity.

My workplace is achieving measurable, significant product and quality improvements with this technology.

[-]

SarmackaOpowiesc@reddit

It's not an enormous shift in my opinion. It just lets shitty engineers fake it longer.

[-]

Jiuholar@reddit

There are multiple significant contributors to Linux and other widely used open source software at my workplace. I'm inclined to trust their experience of 30+ years on this.

[-]

SarmackaOpowiesc@reddit

Lol

[-]

boring_pants@reddit

When you say "measureable" you know the next obvious question is "how did you measure it?"

[-]

Jiuholar@reddit

Primary metrics have been features delivered (not tickets - actual features in production, receiving positive feedback from users), bugs raised (by customers), build pipeline speed and questions asked in slack.

[-]

Venthe@reddit

you have your blinders on (...) There's a chasm between what this poster thinks is the current state of affairs and reality.

I'm regularily reviewing models for their quality. I'd say that the subop is on point. Quality is bad. Abstractions non-existent. It is better, but far from the "standalone" solution - and worse - it is "good enough" that developers forget how to code.

Without fail, "mostly" vibe coded applications and services I've seen across the companies are struggling to achieve reliability and time-to-delivery comparable to pre-LLM, even if the initial gain was theoretically massive.

but this isn't 2025

This isn't ~~2021~~ ~~2022~~ ~~2023~~ ~~2024~~ 2025 - yeah, I've heard this song for a while now. And we'll have AGI in half a year, and ~~GPT-2~~ ~~GPT3~~ ~~GPT5~~ Mythos are too dangerous to be public.

Don't get me wrong, it is an amazing tool - and the wonders that spatial fidelity can bring are amazing especially given the speed of token generation. But even putting aside how heavily subsidized are current models; you need very experienced developers to keep the quality in check and very mature developers so that they will not allow for their brains to turn off.

And since "fresh" coders can't code shit due to prevalence of LLM's, we are indeed heading for the bright future, lit up by the fires on prod that is.

[-]

Deaths_Intern@reddit

I think a lot of folks who still carry this sentiment probably never enjoyed and/or really disliked reviewing others code, and maybe have struggles with turning on their mental compilers unless they "experienced" the development of the software by writing it themselves.

Cause yeah, it writes pretty good code these days. You just have to actually read through it and follow the logic paths to understand it. Sometimes that's a maze, just like stuff we might write ourselves, but the mental difference between looking at the "finished" maze vs carving the path yourself can be quite different.

[-]

developerknight91@reddit

I’m in the process of learning how to code with AI. I have 12.5 YOE as a Software Developer/Engineer and from my studies you are very right, things have changed.

My company, mostly my fellow co-workers are in denial I believe. I have one co-worker who is a staunch AI coder denier, they keep saying “it’s a financial bubble and it’s about to pop” I agree to an extent there is a bubble, but in my personal opinion from the research I have done AI is akin to what the Internet was in the 90s.

We all need to pivot NOW. Either integrate AI into your workflow or in my personal opinion pivot to a job that has a bit of cushion against AI, like possibly Systems Engineering.

They are going to need I believe 3 types of tech professionals when it comes to software - Systems Engineers(to maintain and run the systems that the AI will be working on and building) Software Architects(to have the vision to prompt the AI agents to build and scaffold the system architecture) and MAYBE Senior Software Engineers(to validate the code works before sending it off to QA) I say MAYBE because a Architect could easily verify the output looks good so Engineering roles are the most vulnerable positions right now as we have all seen due to these layoffs in the tech world.

It’s time to adapt and it’s time to pivot. They will need less of us going forward and those of us that are still working need to make ourselves as valuable as possible if you want to stay in the tech world.

We are getting automated out our jobs, it sucks but it’s also reality. And I’ve got a kid on the way I don’t have time to take a chance the bottom falls out on this stuff when so much money is going into it to make sure it doesn’t fail.

[-]

ngfdsa@reddit

I work at a FAANG and have not written a single line of code since December. My whole team is AI first at this point and the entire company is trending that way.

AI can produce garbage code but it is your job to make sure that doesn’t get shipped. The SDLC has changed, implementation has been delegated to agents. Our time and energy needs to shift to planning and review.

First you need to make sure as much of your business domain is documented and available for the AI on demand, not all loaded into context. Similarly, you need to document your build/deploy tools, coding standards, etc. That takes time and effort but it pays massive dividends for your whole organization. You can collaborate with your tools during the design process but it should mainly be driven by you and this is where most of your time and energy should go. Clear requirements, HLD, and especially LLD are critical to success. You can separate LLD from specs if you want but I just make the LLD detailed enough to not need separate specs. Now you should have your business domain knowledge, all the context about your project, and specific implementation instructions. This is when you let the AI loose, not as a one shot, rather you use multi agent orchestration to implement small tasks one at a time. This keeps the review load low for you and your team. You use multiple agents to go through an antagonist review with an agent specialized to challenge the assumptions of the implementation, and a specialized code review agent. Allow them to work in a loop until they are satisfied, and then at that point you come back and review the output.

With this approach our team’s output has increased roughly 500% and code quality has not dipped, if anything our quality has increased because we have significantly better documentation and test coverage.

[-]

vitek6@reddit

you still code. You just don't write lines of code. Anyway I would love to see that code in few years...

[-]

boring_pants@reddit

I am honestly at the point where I doubt people who are getting sold on this were very strong developers to begin with. I keep hearing the models are improving, but all the code I see has the same shitty quality as previously.

I'm starting to form similar opinions based on the quality of the arguments presented by AI-pilled developers. If you can't spot the logical gap between "I have seen an AI agent write some code to solve the kind of problems I see in my job" and "therefore AI agents can write any code to solve any problem" then yes, your work probably could be outsourced to an LLM without much loss of quality.

[-]

Venthe@reddit

Plus there is a massive difference between code that works and code that can be maintained. LLM's, especially with tests work well enough to create something that works. Hell, given enough context they will continue to maintain the output.

But the context is finite; and the output is far removed from how the human developers work; so when something fails - and something will fail because it is not a matter of "if", just "when" - then you have a code that is foreign not only to developers, but to humans in general.

[-]

another_dudeman@reddit

I think these developers were somehow doing repetitive stuff most of the time. I'm typically doing vastly different things each day

[-]

DrShocker@reddit

imo it depends on what problems you're solving. If I need to update a dependency then AI can often take care of that reasonably well. If I am trying to tune a performance sensitive thing then it might act as a glorified template generate to fill in whatever the benchmark framework needs but it won't usually have great "ideas" or "predictions" about what will actually make the code faster. So in general, there's formulaic stuff I think it'll usually get close enough that it saves time, but I agree with you about letting it run rampant.

[-]

Trevor_GoodchiId@reddit

I'm seeing those "we did a month worth of work in 3 days" posts and it's always a Tailwind crud with 5-7 screens. It's a 3 days job to begin with, what were y'all doing all this time?

[-]

xeric@reddit

Where I work it’s very team dependent. Some teams have automated nearly 90% of their tasks to agentic orchestrators, and have fundamentally changed the way the work to be more focused on ideation, ticket creation, code review and CI devops bottlenecks.

Other teams that are more reliant on legacy platforms and third party integrations are seeing AI struggle to deal with their codebase.

But in a year or two I think more and more teams will be in the former workflow, as the need/pressure to adapt our codebases to support agentic workflows will keep increasing.

[-]

vitek6@reddit

Maybe or maybe it will be other way around and former teams will just get back to normal coding as in a long term agentic workflow will be unmaintainable.

[-]

xeric@reddit

I think the more likely way we get more of the latter is if token costs skyrocket due to either Chip shortages (geopolitical conflict, especially with Taiwan), profitability pressure, energy shortages or some mix of all 3

[-]

officerblues@reddit

Yeah, the cost structures also changed, and people will slowly realize that the full rewrite vs refactor non-debate we used to have has shifted, and sometimes full rewrites are actually a valid idea, rarely, now. AI being much better in green fields and small projects can be leveraged. Running legacy systems is a much higher cost nowadays, and legacy teams will be under very high pressure to move out of legacy.

[-]

SarmackaOpowiesc@reddit

The risk with rewrites was never the act of writing the code.

If was always the specter of thousands of undocumented requirements in the code base that no one even knows about.

You can't vibe code a replacement of you can't even formulate the requirements the the slop machine.

[-]

officerblues@reddit

The slop machine can formulate the reqs for you, friend. It's also a much smaller investment, this time around.

[-]

SarmackaOpowiesc@reddit

No, it cannot do that with any amount of precision.

[-]

a-cloud-castle@reddit

It's a tool, and it is possible to wield it effectively. We are definitely in this transitory era where there are trade secrets. Those that know, know. Those that don't, whine and complain.

[-]

vitek6@reddit

of course it's a tool. It's even a great tool in my opinion. But it won't replace all the coding anytime soon.

[-]

BobbaGanush87@reddit

What do you mean marketing bullshit? I haven't coded in 6 months.

[-]

vitek6@reddit

sure buddy.

[-]

big-papito@reddit

Correct. Reviewing code is harder than writing it. This is why we all hate it. This will separate the experts from the vibers.

[-]

zzzthelastuser@reddit

I dislike reviewing code as much as anyone else, but I don't think the reason for that is that reviewing is hard/harder than writing code.

I think we dislike reviewing code that doesn't fit our inner model of how things should be done.

I genuinely like reviewing my own code and I think others do as well. Because, when you xou think about it, that's where the curse of over engineering comes from. You enjoy reading your own code, because it makes sense to you and you always spot something that you could do different or better.

Reviewing other people's code CAN be fun if it's apparent that they have put some thoughts into their design and the code looks coherent.

What I dislike about reviewing vibe code is that there is no mental model in the architecture that aligns with me. It's difficult to describe, but I'm sure everyone here knows what I mean.

It's like someone asks you to draw a triangle, a square and a circle on a sheet of paper. And for some inexplicable reason you draw the triangle with a pencil and the other objects completely filled with a text marker. Sure, the objective is done, but it makes no sense to do it that way.

[-]

big-papito@reddit

Right. However, in an organic AI-first codebase, where the style is defined well, I see a world where the code being written is fairly consistent, regardless of who is directing it.

[-]

criloz@reddit

but if Ai is so good why we need to review the code, I don't understand.

[-]

Which-World-6533@reddit

Yep. The only places I've seen this work effectively are where the problem is a well known problem with a tightly bounded space for a solution.

There is so much hype at the moment it's painful.

[-]

VanillaCandid3466@reddit

Honestly?

All these AI companies are running at massive losses that simply cannot run like that indefinitely. Which means the subscriptions are either going to massively reduce in available tokens or skyrocket in cost.

The average $200 subscription is actually burning through $4000-$5000+ worth of tokens.

So what are the management types going to do when the actual costs are reflected in subscription prices?

I think that will bring balance back.

[-]

overzealous_dentist@reddit

They're running at losses only because they're in growth mode reinvesting into capex. Otherwise they're massively profitable.

[-]

boringfantasy@reddit

Source?

[-]

overzealous_dentist@reddit

Reuters: Anthropic is generating $559 million operating profit a quarter

[-]

boringfantasy@reddit

Numbers look a bit dodgy, recommend Ed Zitron for a concrete analysis.

They have to constantly keep the models on the rails and replace GPUs. So even if they stopped training new ones, there’s immense ongoing cost.

[-]

overzealous_dentist@reddit

that's... capex.

[-]

boringfantasy@reddit

???

That is CAPEX that is never going away. Jt's not as simple as "we can stop spending it on new models".

There is little concrete evidence that Anthropic/OpenAI are turning a profit.

[-]

overzealous_dentist@reddit

It's completely fine to keep taking on capex every year for growth as long as you're good (on average) on operating profit. the objective is to create something incredibly valuable and pay dividends out of that, with rounds of funding as necessary, not to operate the same as a hardware store, which can't continually create massive amounts of new value.

[-]

Buttleston@reddit

Well, good news, they won't have positive operating profit next quarter and will be back to not talking about profit and I guess you'll be somewhere explaining why actually it's good that they are losing billions of dollars a year

[-]

Buttleston@reddit

!Remindme 3 months

[-]

overzealous_dentist@reddit

!Remindme 3 months

[-]

RemindMeBot@reddit

I will be messaging you in 3 months on 2026-08-25 14:47:46 UTC to remind you of this link

CLICK THIS LINK to send a PM to also be reminded and to reduce spam.

^(Parent commenter can ) ^(delete this message to hide from others.)

^(Info)	^(Custom)	^(Your Reminders)	^(Feedback)

[-]

Buttleston@reddit

There is much evidence that they are not, also

[-]

Buttleston@reddit

This is a fake number generated by some grade A manipulation

a) this is "operating profit" i.e. it excludes model training and other capex

b) this is still only achievable because they got a (very temporary) incredible deal from spacex on hosting

https://www.reddit.com/media?url=https%3A%2F%2Fpreview.redd.it%2F0hq7fdnoue2h1.jpeg%3Fwidth%3D1206%26format%3Dpjpg%26auto%3Dwebp%26s%3D2aa307f5f683ddbc8da3c96ee5fd3d2b45114829

This profit is temporary, likely only for this one month

[-]

PepegaQuen@reddit

Yes it excludes model training. that's the point of talking about net operating profit.

No, the musk deal thing was only just done, and is not online yet.

[-]

Buttleston@reddit

Treating model training as capex is pretty non-sensical - it pretends that you won't be doing this training essentially forever, which you will. It's a classic trick companies do, shuffle numbers around on the books to claim that a permanent expense is actually just a temporary one

And that is what the meaning of "ramping up" is - the point is that they have a sweetheart deal this month and next (which they *are* paying) but the cost will go up after that.

That WSJ article is a travesty of financial reporting

[-]

boringfantasy@reddit

Last line of the WSJ article also summarises it perfectly:

"It is unclear what accounting methods Anthropic has used to book revenue and costs, as the company isn’t yet required to follow the financial-reporting requirements of a public company."

[-]

overzealous_dentist@reddit

did you miss that this was a thread about being profitable except for capex?

[-]

boringfantasy@reddit

It's CAPEX that will never go away so what's the point of removing it? I feel like I'm going insane bruh

[-]

Buttleston@reddit

As far as financial trickery goes it is not new at all. A similar trick I have seen played over time and time again is firing permanent employees (operating expenses) and replacing them with more expensive contractors because you can claim that it's a one time expense, not a permanent one. A one-time expense that somehow never goes away

[-]

UnderstandingDry1256@reddit

This is BS everyone repeats nowadays.

How did you estimate the costs to AI companies? API pricing tells nothing.

Other sources?

[-]

m00shi_dev@reddit

Copilot recently revealed the real cost to their subscription users as they’re moving to token based billing to soften the sticker shock. The costs were anywhere from $60 - $2k depending on usage, and users were paying like $35/mo.

This has been known for awhile now.

[-]

xdyldo@reddit

You can buy a $5000 pc and run a custom model better than the best models from literally a year ago. Qwen 3.5 is equivalent to like Sonnet 3.5.

It will of course get better with time. The best models may get more expensive but all the local models will get a lot better and older models will be cheap as well.

[-]

PepegaQuen@reddit

even something like Kimi 2.6 or Deepseek V4 pro are pretty cheap at API pricing, and very useable unless you expect to just vibe everything brainlessly

[-]

1One2Twenty2Two@reddit

It will of course get better with time. The best models may get more expensive but all the local models will get a lot better and older models will be cheap as well.

It's not the models that are expensive. It's the infrastructure on which the models are running.

[-]

xdyldo@reddit

The models getting better implies less infrastructure to run it.

The local models these days are amazing.

[-]

Frequent_Bag9260@reddit

Genie is out of the bottle though. AI is literally used everywhere now.

What are companies going to do? Pretend like AI never existed and tell everyone to do it manually again?

[-]

ghost_of_erdogan@reddit

Two things can be true at the same time;

LLMs are here to stay and tokens are subsidised.

[-]

djnattyp@reddit

Two more things can be true.

LLMs are here to stay as timewaster bullshitter chatbots and slop generators.

Slop isn't a replacement when you need something that's not bullshit.

We already had template libraries, unit test generators and search engines that were just deterministic algorithms. Having Dr. Sbaitso crap out a "maybe this is decent, after I fix it up" solution isn't worth being a gold plated rubber duck.

[-]

Frequent_Bag9260@reddit

If LLMs exist, manual coding will be phased out. That’s inevitable given what we’ve seen so far

[-]

rbnd@reddit

It will become yet another hobby. Like wood working.

[-]

Hot-Profession4091@reddit

There’s a lot of parallels to how hand tool woodworking went away post-Industrial Revolution. We’ll make more, a lot more, of a slightly lower quality. Folks won’t reach for a bench plane or chisel, even when it’s an objectively better choice than a machine. Anyone who is left writing code by hand will do it out of some idea of craftsmanship or preservation of the old ways. “It becomes a hobby” is exactly right.

[-]

petrol_gas@reddit

What???

Computer Science is a monumental branch of mathematics. It’s most simple formulation IS code. You sound like a guy lamenting that all the horse buggy makers losing their jobs is the end of engineering proper.

[-]

Pleasant-Cellist-927@reddit

Except nobody here is talking about Computer Science, we're talking about Software Engineering. There is a reason these have always remained distinct.

In CS the hard part is actually writing the code and figuring the science behind it, while SE has never been about writing the code, because most companies don't actually have difficult requirements, and more about playing the corporate kayfabe that has made simple changes nearly impossible because 5 people who aren't even doing the ticket all have different political interests in you not doing the ticket or only doing the ticket their way or some other petty shit.

[-]

petrol_gas@reddit

So okay. There’s just a ton to unpack with this but I’m going to skip to the end.

If software engineering is solely an artifact of business and has nothing to do with CS in practice then no one will be doing it as a hobby.

So guy still sounds a bit like someone lamenting the end of a horse buggy which is exactly why I said it.

Also if I had all day I’d fight you on that whole SE and CS are unrelated bit. (Modern business is a mess but SE has always been about CS applied to practical problems).

[-]

Pleasant-Cellist-927@reddit

I feel like your horse buggy analogy is inherently flawed because horse buggies stopped being made when something better came along in cars.

The better analogy is everyday household furniture made of wood. In the old days, I'd go to a woodworker, ask for a chair and table and wait a week, maybe more, come back and collect. Now, IKEA just factory cuts a million copies of the same generic chair and table and I just go in, buy the pieces, assemble it and put it together.

The whole process has been optimized at the cost of 'craftsmanship', but the average joe doesn't care about that. He just wants a table and chair in his home for a reasonable price. If it breaks in a year's time, oh well. He only paid $50 for the set. Now if average Joe wants a table and chair that will last him a lifetime, sure, there's still some skilled woodworkers around who can create bespoke pieces but those guys have knowledge that Bobby working at IKEA for a summer job won't and will charge accordingly, upwards of $5000.

As for the difference between CS and SE, I don't agree there either. Sure, in theory SE is just applied CS but the problems it is being applied to are so simple that, if the corporate aspect were taken away, you could teach the problems in a CS101 course. Obvious exceptions for y'know, medical tech, military tech, etc but most people are just making static websites, internal CRUD tools or integrating SalesForce into yet another company.

[-]

petrol_gas@reddit

I think your example is a more accurate representation of the problem as you said.

Truth be told though, cabinet makers and specialty shops are standard practice for making wooden things that are meant to last. The cabinets in your home are made of wood by a cabinet maker- except in one case, when you’re poor or you’re building a rental property- you’re buying the cheapo ikea style furniture.

You can call this optimized and in the sense that optimizing for least cost / value is optimal- sure.

My horse buggy analogy is not meant to be accurate or correct. It’s meant to call out the hopeless negativity of the original guy who is 100% misunderstanding how AI affects the industry.

[-]

Hot-Profession4091@reddit

lol I’m not lamenting shit. I haven’t written any code by hand in months and I couldn’t be happier for it.

[-]

den_eimai_apo_edo@reddit

🤔

[-]

lurco_purgo@reddit

I definitely see the parallels, but there's also a lot of differences. For instance, physical production relied on the Industrial Revolution because of quantity. In software engineering this is not the issue.

[-]

robertbieber@reddit

Pretty much any custom furniture maker is going to have and use hand tools when appropriate though. Sure the really mass produced stuff is all coming off power feeders and CNCs, but I'd wager most one off pieces are seeing at least some chisel/plane/handsaw usage even if they're mostly built with power tools

[-]

Blecki@reddit

This guy has never touched wood... you can't build anything without picking up a chisel at some point.

[-]

robertbieber@reddit

Probably never been a better time in history to be in the market for a new hand plane, either

[-]

demosthenesss@reddit

While true in niche cases it’s generally not true. Woodworkers use way fewer hand tools now either way.

Which I think will be how coding goes too.

[-]

big-papito@reddit

Woodworking is not knowledge work, it's a hard skill. Yes, you can work to improve yourself, but woodworking does not reset itself every few years. "Hey, gang, new tools just dropped, forget everything you know".

It's like saying everyone will be just vibing science now for fun.

That's a no.

[-]

Zombie_Bait_56@reddit

Given the way that LLMs "learn" to code we've seen the birth of the last programming language. Heck, we've probably seen the last library as well.

Assuming this isn't just another bubble.

[-]

yxhuvud@reddit

Nah. LLMs do pretty well with Crystal despite there not being a lot of training material. As long as the language is not too foreign the LLMs will be able to handle it.

[-]

Zombie_Bait_56@reddit

There are thousands of repositories tagged with "crystal" on GitHub alone. Undoubtedly there are a lot of forks in that number, but I don't think the LLMs care.

[-]

yxhuvud@reddit

Much more important is that the code look a lot like Ruby (to the point where fails usually is in the way of ruby-isms).

[-]

ithinkilefttheovenon@reddit

I think we could see new LLM-friendly programming languages come about. Most popular languages today are abstraction layers sitting on top of machine code, for the purpose of making it easier for humans to write the code. As LLMs become the developers, the need for that abstraction goes away.

Actually maybe you are right — if LLMs can just write machine code then the programming language is no longer needed.

[-]

yxhuvud@reddit

It is fairly stable now, but there was a time where stuff like lathes and other simpler tools were still being invented where it was in plenty of flux.

[-]

Hot-Profession4091@reddit

Lathes are literally ancient. I hear what you’re trying to say, but it misses the mark with that example.

[-]

yxhuvud@reddit

Powered lathes as exist in a workshop are definitely not ancient.

[-]

oVtcovOgwUP0j5sMQx2F@reddit

joinery is absolutely knowledge work

[-]

newEnglander17@reddit

It is but it’s also a skill. I have all the knowledge and tips and tricks about sharpening plane blades and I am still 100% terrible at it after several years.

[-]

hibikir_40k@reddit

Have you never seen a programming videogame? Sometimes they simulate systems that are not all that different from the real world, others they go with something completely surreal, which is still turing complete, but with different computing models, strange multiprocessing or what have you.

People buy them, and play them, for fun. You don't have to if you don't want to... but, fortunately for the world, you aren't the only person in it.

[-]

gingimli@reddit

Eh, I think they are more similar. Programmers don’t have to learn totally new tools every few years, they have to learn a new abstraction that is hopefully easier but ultimate does the same thing as the old tool.

[-]

SlapNuts007@reddit

Yeah these takes make no sense to me. We haven't even seen the development of entire language ecosystems built around AI yet, so the idea that there will even be a "craftsmanship" upside to hand-coding is debatable.

[-]

thedifferenceisnt@reddit

Artisan hand crafted applications

[-]

pydry@reddit

Doesn't really work like that. Software is infinitely replicable.

Do you want one weather app on your phone that actually works or 25,000 half baked buggy vibe coded pieces of shit?

If you secretly really wanted the latter, then everything has changed. *everything*.

[-]

thedifferenceisnt@reddit

You are assuming vibe coded will always mean worse. Considering the improvements I've seen in Claude and other models recently I doubt that that is the case

[-]

pydry@reddit

Some people lack the experience to be able to tell the difference.

[-]

Oldmanbabydog@reddit

Ethically sourced fair trade organic code written by free range developers

[-]

dronz3r@reddit

Zero emissions, sustainable and green code. Zero GPUs used.

[-]

anonyuser415@reddit

Sure AI can unroll loops these days but there's just something about those Cherry Red-typed characters man

You have to get the single origin executables, you'll see what I mean

[-]

tripsafe@reddit

Relevant Krazam video

[-]

_ak@reddit

It always has been a hobby. A lot of late 90s, early 2000s open source was just hobbyists doing stuff for the pure joy of it.

[-]

dryiceboy@reddit

Full circle

[-]

Which-World-6533@reddit

It always has been a hobby. A lot of late 90s, early 2000s open source was just hobbyists doing stuff for the pure joy of it.

This shows a woeful underestimation of a lot of open-source projects of the late 90's / early 2000's.

[-]

_ak@reddit

I grew up in that environment and was socialised as a programmer, and I'd say it's the opposite of it. "Hobbyist" or "amateur" are not pejoratives or terms to downplay the quality of projects or the achievements as such. The etymology of "amateur" really shows what it's all about: it comes from Latin amare, "to love".

[-]

GuyWithLag@reddit

I've been writing code since the late 80s. There's 2 kinds of people in this profession: those that wouldn't code if this wasn't their profession, and those that would code even if it wasn't.

The first ones will get you products, the second ones will get you tools, philosophies, and science.

[-]

CrushgrooveSC@reddit

You mean things like Linux, Redis and Nginx? Funny how quickly hobby turns into actual real world impact.

[-]

_ak@reddit

Hobbies have real world impact, always have. The motivation of creators is just a different one, namely curiosity instead of commercial success.

[-]

Which-World-6533@reddit

That Linux thing will not go anywhere.

Plus we have Apache. Oh, wait...

[-]

Gooeyy@reddit

Their point is that it will exclusively be a hobby.

[-]

loganbrownStfx@reddit

Unfortunately, the answer is that you don’t get paid to code, you get paid to build and run systems. Coding is (was?) just a means to an end. There will always be value in being able to read and understand code I think. But manually writing code it’s already losing its value as a skill.

If you love to code and that brings you fulfillment, then do it as a hobby or try really hard to find a place that moves slow enough that allows you to write code manually.

However, most companies will likely want to move too quickly for that to be a reality.

[-]

IXISIXI@reddit

Should be the top comment. I feel like a lot of people in this profession are quite naive or childish. Do you think most professionals have the luxury of being able to complain about a “fun” part of their job going away? As a career changer i wonder how a lot of devs would survive in an actual bad job.

[-]

loganbrownStfx@reddit

I don’t get it sometimes. I feel like a lot of people don’t actually realize what the value they provide to the companies they work for is, it’s not code.

I absolutely empathize with them though, if they feel like a part of the job they enjoy is going away, then that would be very difficult. Unfortunately though, that doesn’t really mean anything to the companies they work for.

[-]

BusinessWatercrees58@reddit

It's entitlement. Many of these people haven't had to struggle and have only known being paid well to do what they love. They can't fathom anything else.

[-]

Fyren-1131@reddit

I struggle so much with this tbh.

I am a musician at heart, and programming has been a nice overlap with the creative / compositional aspect of musicianship.

Now I'm just stressed tbh, and not enjoying my day-to-day at all as an 8 yoe backend developer in enterprise.

[-]

daraeje7@reddit

Same here. Graphic designer was my passion as a teen. Took on CS for the overlap with that creative instinct.

The creative part is gone now, so I’ve suffered huge cognitive decline in programming over the past few years.

On top of that, I am not expected to pretty much all of the full stack tasks with AI, including infra.

[-]

nanotree@reddit

I get where both of y'all are coming from. I grew up interested in art and music. I drew for hours on end and then in my teens learned to play guitar and bass guitar. I was the kind of instrument player that learned by ear and couldn't read music. I don't do either activity but once in a blue moon now.

What's weird is I don't find the loss of creativity to be the case. I work backend, and have developed a lot of methods over the year to keep things organized and to use repeatable patterns. I don't simply let AI do what it wants. Sometimes I'll go prompt first, then refactor what it spits out to be more generic and structured, since it's usually not either one of those things.

I've found AI does the bare minimum to get things functional. Which isn't how I write code and certainly not anything I'd want to maintain. AI does better when you give it a working example of what you want with loads of context.

Personally, I still get creative satisfaction. Because I'm still doing the parts I enjoy the most creatively. When I do go "full AI", I use it to build scripts to automate tasks, and I find that process enjoyable because I still have to devise ways to get the AI to do what I need reliably.

Maybe your experience is different somehow, I don't know. It'd be interesting to compare how and what you are using it for.

[-]

Fyren-1131@reddit

I think we are looking down the barrel of enshittificatiom. My hope is that tokens will be expensive and these ai gurus with 10 agents in parallel will price them selves out of existence. I don't mind using ai, but that whole management style of the job is something that makes me need a vacation after 7h of work.

I think that as we enter usage based billing, it will cool down. For better for you and me, for worse for some others.

[-]

pydry@reddit

It's already getting harder and harder to keep a lid on the smoldering dumpster fires agentic coding has spawned.

Even if usage based billing never happened I think the retarded executive enthusiasm would dissipate. Usage based billing though - that is going to kick that enthusiasm in the nuts.

[-]

corny_horse@reddit

Then few people won't have the skills and there will be even fewer juniors to potentially be competition.

[-]

pvgt@reddit

perhaps an unpopular view, but I kind of like LLMs for the single reason that my brain is less fried at the end of the day, and I have a bit more energy for things like practicing music or reading. I do get exhausted using agents, but it's different than pouring over language syntax and perhaps abstract enough that I still have energy for non-work activities.

[-]

Legal-Software@reddit

I went the opposite direction. Close to 30 YOE programming, now recently took up modular synthesis as a hobby to scratch that creative / problem solving itch I no longer get from my job. The NerdSEQ especially has managed to fill a happy middle point.

[-]

Fyren-1131@reddit

Happy to hear that! It's something I'd love to get into. It looks so fun to fiddle with all that hardware. My main passion sits at the production and composing side of progressive metal, so it's a bit more hands on. Still very fun!

[-]

neonskimmer@reddit

also a music nerd and old man programmer here, and modular synthesis is a great way to liquidate all that disposable income we've accumulated!

joking aside i've managed to stay away from modulars, mostly because i like polysynths better. good luck and have fun!

[-]

experienceddevsb@reddit

This flair is only allowed on wednesday, saturday (UTC). Please repost on an allowed day. Intentionally trying to circumvent this rule will result in a suspension. See: https://www.reddit.com/r/ExperiencedDevs/comments/1rfhdrg/moderation_changes/

[-]

JohnWH@reddit

I still write code by hand and in general keep up. I am no longer the top code contributor, but I still get enough out each day. I use AI to review my code and provide feedback on how to write it better, but I write it out myself because that I how I better understand the true problem and all the nuances.

Something I noticed with a lot of my AI heavy coworkers is that they get to 90% really quickly, but the last 10% takes them weeks to months. We are now behind on a number of projects, even though they definitely review their own code and typically take an incremental approach to their AI use.

I still introduce a lot of bugs, but I feel like I catch them a lot quicker because I know the code and flow well. When asking a coworker how a feature worked they had a decent understanding of it, but missed a lot of the nuance that came with the problem, and were dependent on AI to point out the bug. On the other hand I have been able to quickly fix a lot of my issues because I know the flow.

Will this last for me? I highly doubt it, but I am going to try to enjoy my job for as long as I can before I have to do the same. AI isn’t going away, but I do wonder if its usage will change once companies need to make a profit.

[-]

DowntownLizard@reddit

Coding is basically dead. Software engineering is very alive. I would consider myself to be a solid coder but I havn't written code in 6+ months. With the proper skills and practice, because I use AI a lot, it is capable of writing better code than I would have in way less time. Things that I would have considered over optimized in the past it can just do in 2 minutes. All the things that we as engineers think are best practices but really never get to implement because they would take too long at my companies size we now have. Extensive automated tests, logging, resiliency, etc. All the maintainability perks that delay delivery time are completely feasible now.

I was in the camp of loving to code because it scratches a certain itch of diving into the nitty gritty details of how to make a task happen through logic. Thats kinda sad to have lost, but we gained a lot. 10x used to be a meme and now I'm pretty sure I am. That feels good to me. Im able to deliver awesome solutions in fractions of the time. That still brings me a lot of enjoyment. That was the point the whole time anyway. Solve a problem using software. The end users or business doesn't care about your code and how elegant it is. They just want it to work well and make their lives easier.

Most companies are not on the leading wave of this but its definitely coming for everyone at some point. Nothing in the future will prevent you from writing code yourself but you probably wont be allowed to in a business environment thats trying to be competitive. Engineers wont lose their jobs to AI they will lose them to people who are good at leveraging AI. Its a major disruption event and you'll have to adapt or get left behind.

I would argue as well that most of the skill of being a senior+ engineer has nothing to do with writing the code anyway. Taking a story with pre crafted requirements and implementing it on a lot of levels is an entry level task in my opinion. A lot of the thinking has been done for you already

[-]

Monowakari@reddit

I'm sure there people who love analogue typesetting, or scribing by hand with quill or fountain pen. And these can be art forms or hobbies, but I'll tell you what they aren't anymore, massively employable skills. Enjoy your hobbies, just because an automated process exists doesnt mean you can enjoy it

[-]

TopSwagCode@reddit

Keep on coding my friend. The market for good developers will not die. AI ain't perfect and when some companies AI slop them selfs into a corner, they need people with exp. to come help out.

I dont know for sure how the future will be, but I will keep trusting my gut. AI is for sure going to be part of it. Perhaps smaller models and selfhosted, that is more of a helper tool. But having these huge models that can do everything is a huge strain on the economy, infrastructure and environment.

I work in power / electricity field and the grid as I see it is struggling in most of the world now. Incidents are slowly getting bigger and bigger. Many places it hasnt been built to handle.the load and kinda load we are doing.

The world used to be run on "few" big generators that slowly scaled up Nd down production based on normal patterns. But now we have shit loads of solar and wind which is highly variable, where clouds can cripple production.

Sorry 😜 went on side rant. Go buy personal generators. Thin foil hat 😜

[-]

Blecki@reddit

AI can't invent. We are entering a world of vast applications being invented over and over that don't innovate on anything in anyway.

[-]

uber_neutrino@reddit

If everyone writes C or Cobol code are my punch card assembly skills still useful?

And what happens to people who genuinely enjoy the craft of programming itself?

In the same way nobody is stopping you from getting a keypunch and making some punch card code nobody is going to stop you from hobby programming if that's what you want to do.

[-]

druidgaymer@reddit

Yeahhhh this is what I'm struggling with. I'm 5 YOE and the coding is the fun part. Everything else is blah you know? I've been coding since I was 12 and I enjoy it.

[-]

rcls0053@reddit

I think all developers have thought of this. We get that dopamine rush from solving problems and now it's being taken away.

[-]

blood__drunk@reddit

No it's not, just the tools you use to solve the problems are

[-]

ghost_of_erdogan@reddit

It’s not the same rush anymore.

[-]

grimsleeper@reddit

Ya, ai heavy toolchains feel more like middle management.

[-]

loganbrownStfx@reddit

I’m not quite sure I agree here. How is solving problems being taken away?

Writing code isn’t solving problems, solving problems is solving problems. The way we solve those problems has just changed

[-]

Successful-Actuary74@reddit

You don't get it

[-]

loganbrownStfx@reddit

I’m not sure what you mean by “I don’t get it”

I totally empathize with people who enjoyed this aspect of the job, but the value engineers deliver isn’t writing code and never has been. It’s solving problems, which is absolutely still a part of the job

[-]

Fancy_Ad5097@reddit

On the architectural level, sure, we still make decisions. But what I loved about coding was the micro decisions in each line of code. And how all of that came together to make a working product. There was a huge sense of “wow, I did that.”

[-]

e430doug@reddit

I guess it all depends on what level of problem you get your joy from. I still get exactly the same dopamine hit using AI tools as I did when I manually wrote my code.

[-]

sweetno@reddit

You can code in your free-from-work time. A good retirement hobby too, keeps you sharp for longer.

[-]

GotAim@reddit

I think a good analogy is craft work, like sowing for example. First it was just thread and needle, but then we made machines to automate the work. People still sow or knit stuff themselves, but mostly as a hobby.

[-]

boring_pants@reddit

I mean, sure, we also made machines to automate the work of converting assembly mnemonics into machine code. We made machines to automate the work of converting C# or Go code into assembly. We've done a lot of automation within our field. It does not automatically follow that "in the future all software will be developed by arguing with the hallucination machine"

[-]

GotAim@reddit

You’re entitled to your opinion, but I think it’s very obvious that AI will be doing most/all coding in the future

[-]

boring_pants@reddit

And describing using AI the way you’re doing tells me you either have some weird bias against it or you use it incorrectly

Do you not know how an LLM works?

It is a hallucination machine. Sometimes its hallucinations are useful, and you can make an argument that they're a net benefit. But it's still a hallucination machine.

I'm a software engineer. When I discuss software I'm going to discuss it in terms of what it is and what it does and how it works. What we call hallucinations are a core element of LLMs. They are intrinsic. Like, we have research papers on this. I'm not making it up.

The question that is worth asking is whether these hallucinations are useful. They certainly can be, at times. At other times they're detrimental.

[-]

GotAim@reddit

calling LLMs hallucination machines is not «saying it how it is. It’s an unnecessarily negative interpretation of what LLMs are.
I never talked about LLMs, that might not be the best way to do AI in the future. You assuming that AI = LLM when talking about the future tells me a lot

[-]

Nofanta@reddit

It’s the same as people who love to knit or churn butter.

[-]

Purple-Cap4457@reddit

We are basically fucked 💀

[-]

roger_ducky@reddit

Everyone gets “promoted” a level, responsibilities wise, without changing job titles.

So ICs will absolutely still code. I mean, current leads of small teams do.

[-]

Battleaxe19@reddit

work is for money. everything else is what brings me joy.

[-]

justly_tuneful@reddit

Make a side project and start to love money instead. Show up for the salary not the work

[-]

e430doug@reddit

I’m at the 40- year mark in my career. I love to code. I use AI tools to generate most of my code these days. I still do all of the tasks that you mentioned. I still debug. I still design. I still refine implementations. I still use abstractions. I’m deriving as much joy as ever, because I also like solving problems. If someone’s joy comes from typing into a computer, then that person will have a problem. Typing code into a computer is going to be more of a hobby than a career. Just like writing assembly code became more of a hobby except in certain niches. That said, knowing how code gets compiled into assembly is a really important skill to have in this new world of AI tools. Computers put constraints on how problems can be solved. The more you know that in-depth, the better you can use AI tools. Having a long career in software engineering depends on adaptation and continual learning. This is just another phase.

[-]

Outside-Storage-1523@reddit

I have the "luxury" of boring job so I do some side projects to satisfy my "builder" side. Since I'm not interested in my job, it's fine to just generate code -- unfortunately AI is yet to generate good quality code because my work needs a lot of context that even I don't know all of them.

[-]

blissone@reddit

Well we are SOL. I haven't coded since february, I'm incredibly bored, the fun and interesting aspects are gone, honestly looking to get away from tech but not sure it will happen. Architecture and high level solutions are still interesting but it wasn't what kept me running day to day. The good thing is that I have much more mental energy nowadays so just need to channel it somewhere else and not look beyond the paycheck at work (tbh this probably was always smart).

[-]

RubOk1972@reddit

I watched an (AI slop) video on the topic recently. Usually dont like ai slop but it made interesting points in that LLMs are trained on “good” code but good code is often just microservice slop with 10 layers of abstraction and thus doesn’t factor in things like cost, performance, and tradeoffs.

Not to mention the hallucinations are very convincing and kinda go against the fail fast principle

[-]

WheresTheSauce@reddit

You aren't paid to do a job because you love to do it

[-]

PorkChop007@reddit

This. At the end of the day, it's a job. I'm being paid to deliver value. It was nice when that necessity aligned with me having a good time coding, but things are changing and I still would like to be able to pay bills and put food on the table.

[-]

amtcannon@reddit

I’ve been getting paid to write code for 17 years now (I just worked out how long exactly and it made my bones hurt).

But for most of the last 4 years AI has been doing a lot of the boring parts for me. The joy of coding has definitely been lost, the artistry of it. I’ve taken up carpentry so I can scratch my “make something good, use your skills, pay attention to the little details” itch.

I have felt a bit of a sense of loss over the last 12-18 months since Claude code completely changed they way software is created. There is no going back now though, this is what the job is now. Glad I’m close to retiring, feel sorry for the young folks who need to figure out the rest of their working lives

[-]

zilchers@reddit

Ai has brought back my love of coding - so much of my day used to be repetitive or just flavors of the same stuff, now all that just works, and I get to work on the hard problems

[-]

timwaaagh@reddit

There will probably be an assisted living facility or something

[-]

Comprehensive-Pin667@reddit

I used to love coding in assembly. Knowing how to do that still sometimes helps with some higher concepts.

AI code is a bit different because it's pretty garbage until you code review it, and it's not something that's improving. So knowing how to code won't become optional any time soon as code reviews will remain necessary. But actually writing code might become a hobby instead of the job.

[-]

SoUpInYa@reddit

I't spita out so much code, I'm worried that something is gonna break and I won't be able to fix it. Any performance concerns are 'solved' by throwing more horsepower at it

[-]

revolutionPanda@reddit

Your company doesn’t care about what you like. Line must go up.

[-]

Bubbly-Watch6214@reddit

I would wait for a few years before you decide whether or not this generative AI thing hurt your career or ended up boosting it. I work in a high level position within a company and I don’t quite see the productivity gains across large teams. Rather, I see true productivity gains being roughly distributed with skill. Excellent developers can build good projects quickly with generative AI. Mediocre developers can make bad decisions faster than ever. I’m not convinced that skill levels are increasing and highly skilled and experienced developers are the most likely to leave when they realize a project cannot be saved. There’s a possible future where cleaning up the last two years would be a calling for people like you.

[-]

Ok-Letterhead3405@reddit

TBH I've found my love of code challenged a bunch of times in my career, mainly by things like shared UI libraries done by platform teams I'm not on and building things on top of packages like Material. That said, I'm a frontend dev who loves UI and building composable systems of components, so what it really means is that I should try to get on a platform team. I've just found that very difficult to break into, and even then, there's a lot of stupid politics and such.

That doesn't stop me from doing it for fun on the side or still finding things about my tickets and even my glue work that I can enjoy. It's actually getting old and tired and burned out that stops me, most of the time. Sad.

And maybe coming from a UI perspective and being at a relatively small, quiet, technologically conservative-ish company, I don't feel AI really breathing down my neck at this moment. Management wanted us to pilot test using something they wanted to purchase for us, but I ended up turning it off and stating in the survey that I found the tool was more distracting than helpful and offering too much unasked for CSS slop.

I can see a possible future where AI tools become less faddy and start coming down closer to Earth, where they help speed up parts of our jobs in a way similar to frameworks and other tooling. Probably, the market will become even smaller and pickier, but I don't see it fully retracting. I think companies might start using AI to get more production out of the same amount of devs. Projects on the back burner for a few years might see the light of day more quickly. Dev overall becomes cheaper, so more of it is produced.

What is more worrying to me is that we're pulling the ladder up behind ourselves on junior devs, more and more, and at the same time, it sounds like many juniors are leaning on AI too much and not developing the skills they should. We'll probably come to a place where we end up with too few engineers again, just not in the same way as it was in 2022 or the 2010s.

One of the things that I loved about the earlier web was that if you liked it, buckled down and learned it and built some stuff, you could turn it into a career, one where employers often treated you better than people get treated in a lot of other jobs. All you had to do was work for it, and I felt there was more variety of people and backgrounds who shared this career with me. Being able to break in self-taught, though it was difficult to do, basically saved my life. I'd be sad to see that go.

[-]

Empanatacion@reddit

Did we stop limiting AI posts to only Wednesday and Saturday?

[-]

dom_optimus_maximus@reddit

This pressure existed before AI and will recalibrate again in a few years. I am classically educated and find all the inspiration and joy in coding through the elegance and poetry of the craft. I very often felt like I was pushed to get things out fast without craftsmanship before AI. If anything I feel on the front foot now for the first time.

The key is to find out how to use AI tools to get leverage back. This can be harder for earlier career devs because the AI can get away from you particularly if you let it hop into code generation. By the time its done, the code is "good enough" that you have difficulty finding problems with it and tend to amend here or there and ship.

Instead, use AI tools in thinking mode to make designs, refine the state machine, run TLA+ tests on entire systems and gain deep modular understanding of a PRD. Then take those designs (which are free of code just models, diagrams, state transitions and flows) to create plans for the user stories where you carefully comment on your boundaries, naming, folder structure, etc. I can genuinely say that with this kind of attention I can hop in and out of the code I write almost seamlessly with the AI, rewriting functions, shaping, refining, and writing custom test cases if I want.

The AI still does things I genuinely despised and I make sure that it fixes those so the final shipped code has my approval and my personality. I honestly have shipped features and apps at a quality that I would never have been able to in years past and genuinely understand what I wrote.

[-]

itix@reddit

I can finally finish the projects I had no time for. With kids and all that life, it is very difficult to work on hobby projects, but now it is finally possible.

[-]

mancunian101@reddit

If you want to write code then write code, even if your employer is full on AI development only there’s nothing stopping you from not using AI is personal projects.

[-]

The_Other_David@reddit

Some people love to knit. That doesn't make knitting by hand efficient enough at scale to work in real manufacturing.

Maybe you can have a stand at the local farmer's market selling hand-coded To-Do Lists.

[-]

mancunian101@reddit

That makes no sense, software engineering isn’t the same as manufacturing, you build one app, not thousands of copies of the same app every day.

Hand writing code in production is what people have done for decades and it has scaled just fine. Yes it’s not as fast as AI (AI is about 30% faster) but I’m not completely convinced that how many PRs you can slop out a week is a good metric for measuring whether AI is “better”.

[-]

_JaredVennett@reddit

I could just about deal with being forced to delegate the coding to an AI agent but I don't want to nanny it, correcting its mistakes all the time thinking I wish i wrote it instead. And sure while I never write bug-free code first time at least the bug is because of my code so it hits different fixing it.

[-]

Dimencia@reddit

You can always just get a job at one of the many places that haven't fallen for the AI hype. AI has its place, but it's not replacing anyone, and the costs are already prohibitively high, and likely to rise because LLM providers generally operate at a loss already. It just hasn't been around long enough for the major problems to become evident, such as bad practices that get ingrained in the codebase which take years before anyone realizes there's a problem at all, but by then it has spread to everything and would require major rewrites to fix.

And when nobody owns the code or understands the codebase, the only way to maintain it is with AI agents that eventually become even more expensive than they are now. The whole LLM marketing model relies on this - get companies hooked now with low prices, and then they have no choice but to keep using it in the future when the prices reflect the actual costs of running them

I'm still pretty confident that this bubble is right on the edge of bursting, as companies start to realize the costs they've incurred by focusing so heavily on AI tooling in the first place

[-]

another_dudeman@reddit

Now you gotta read a shitton of code

[-]

memoriten@reddit

I like coding as well, but I code mostly for hobby projects. For a few years it's been just meetings, architecture decisions, warrooms and delegating stuff to others. Now it's just a job to pay my bills.

[-]

SituationNew2420@reddit

I think people are really speaking from their own experience when answering this question. We all write code in dramatically different contexts. Not everything is a web app. We have big tech, gaming, finance, med device, aerospace, defense, on and on. We have different levels of the software stack. Businesses with different goals.

Every single one of us has to contend with the existence of LLMs, and figure out the best way to use them in our contexts. There are contexts where you need an LLM to rip through thousands of lines of code and ship fast. There are contexts where you're writing a safety-critical class where you do it by hand, because the cost of a defect or misunderstood requirements is too great.

I think both extremes, either coding as a hobby or LLMs as useless, are totally incorrect. The only honest answer imo is it depends.

[-]

fdeslandes@reddit

I get you. I love the actual coding part, it's what prevent burnout by putting me in a flow state. Context switches are burning me out and AI flows with all the waiting and switching tasks are replacing the one nice part of the job with more tiring shit.

I'm going to continue coding by hand until it changes or they fire me. Hopefully I'll have accumulated enough money by then to switch to a lower pay field or lower pay job in the field where they don't require AI. Good thing right now I am still more productive without AI than most colleagues using it, as 20% gain was not enough to fill the productivity gap.

[-]

Flashy-Whereas-3234@reddit

Dunno man, I'm trying to think of parallels and the best I can think of is 3D modelling.

Used to be that you would vertex poke, manually map UVs, cull faces, trim, teak, optimise. Now the workflows are around the tools and hugely fluid and automated in comparison, people work so much faster to produce so much more complex stuff. They can still go down to the details if they need to, but usually that means something has gone wrong.

I feel like coding is headed the same way, the tricky part is code isn't visual, it either does or it doesn't. I'm wondering if code and interfaces will evolve because now we can throw so much more metadata at the problem, even if it gets compiled away at runtime, the keystrokes don't matter just like the vertecies and tris.

You lament the old way like Phil Tippett laments stop motion fx. There will always be an art to it, there will always be experts who knows better and watch the industry punch itself in the nuts. Opening the doors to more people means there WILL be a generation of shit while they slowly learn what part of our jobs was actually art.

But it's an evolving art form, AI is just a tool. Becoming more powerful by delivering more isn't about speed to delivery, it's about building better, building systems, thinking deeper and further than your keystrokes could get you originally.

[-]

oVtcovOgwUP0j5sMQx2F@reddit

remember when folks resisted JavaScript frameworks?

[-]

yxhuvud@reddit

Yes, eventually they invented HTMX and Hotwire and brought the joy back into building simple webpages.

[-]

oVtcovOgwUP0j5sMQx2F@reddit

they were scriptaculous

[-]

Gunny2862@reddit

Heard a pretty good conversation that had me thinking about this. The people were talking about how photography changed art into being more emotional/subjective than realistic because that wasn't necessary anymore. Not sure what the answer is for coders, but there could be something here to think about.

[-]

jcradio@reddit

I've struggled with this too as a founder who still programs. While some of the tools have been helpful, there are still things the tools cannot do well and it takes someone with experience and passion for the craft to see something wrong and fix it.

I actually felt a twinge of depression when I thought "of we automate the party of the job that was fun and never the problem, and leave only the pain points of unclear requirements and code reviews, I won't want to do this anymore."

[-]

bartekus@reddit

I’ve only ever done this job for the love of my craft. You can be very far and very close to it at the same time. It’s a matter of perspective and personal drive. For me stagecraft-ing/open-agentic-platform became a therapy and a re-alignment at the same time.

[-]

_itshabib@reddit

Use that to learn new stuff but be very involved in the plan and design docs LLMs produce. It helps

[-]

Neat_Strawberry_2491@reddit

Get your bills paid however you need to and then you can do whatever you want with your free time

[-]

nomiinomii@reddit

"what about people who love to ride horses"

There will be elite clubs and meetups and events where you can hand-code

[-]

djnattyp@reddit

What about people who used to love manually astroturfing reddit?

Sorry bro, AI slopaganda is inevitable.

[-]

Mobile_Reward9541@reddit

Its ok that you love coding. Whats changing is, people needed your love for coding so they paid you for it. Its no longer happening. Because people who love coding are resisting the ai coding change. Now people who dont love coding but rather love solving business problems will thrive.

[-]

davedrave@reddit

I don't know what is to happen or if the AI bubble might burst due to costs vs manual coding. But I do miss coding both due to my role in management and due to what AI is doing to remove coding from software engineering roles. I fear that if I take a step overto a coding role, the role is simply not what it was

[-]

dark_mode_everything@reddit

I tried agentic coding after being literally forced to use it and while it does work well enough in most cases I feel like I'm just reviewing other people's work. There's no sense of accomplishment, sense of ownership, or any sort of learning. Generate the code, review it, push it and forget. It's just made the job joyless. Luckily I have my own projects where I can do what I want.

[-]

Stamboolie@reddit

I like making software - I haven't enjoyed corporate coding for a while, maybe never if I think about it. LLM's mean I can code and experiment a lot quicker so it's good. Can I make money from it still, probably, there's still skill involved - maybe even more than previous. Coding itself is pretty tedious

[-]

Unlucky_Data4569@reddit

They aren’t paying you to code because you think its fun. You just might not like your job as much moving forward

[-]

allo37@reddit

I feel like the industry in general *hated* programming, well before LLMs came along.

For a long time engineering has felt to me like being a professional shopper: Why write something yourself when there is a library to do it for you? Why use code when we can just drop in a no-code or low-code solution? These are the maxims that I constantly heard.

It's almost as if something about having to find people who actually know what they were doing gives this industry a cold sweat. Too expensive. Too risky.

Now LLMs have come along and they cranked this up to 11. But also maybe not? Will all the whacky abstractions on top of abstractions of old disappear now that code is easier to produce? Will there be a sort of renaissance for coders in at least reviewing and maintaining the massive amount of bespoke code that will now be pooped out by these LLMs instead of being provided by an external dependency that we couldn't touch?

[-]

ritchie70@reddit

I don’t usually code very often lately, but I’ve done a decent amount of it in the last three months. I do Windows applications, so I’ve spent some time with Copilot in visual studio.

I actually enjoy maintenance programming – and what I mean by that is figuring out someone else’s code and making changes. That’s what using Copilot feels like to me. It writes code, but it’s not quite the right code that it wrote.

[-]

almarcTheSun@reddit

Programming was a profession. Now system design will become a profession and programming - a hobby. We're still a ways off and it's no guarantee that this wave of automation will work out, but there is no going back already.

[-]

FatefulDonkey@reddit

Yeah, noone cares about what you love. Else they wouldn't be paying you.

[-]

ShodoDeka@reddit

The same thing that happened to people who genuinely enjoyed programming in Asm. Some are still around in niche fields, some still do it for fun, but mostly it is something you need for troubleshooting hard problems.

[-]

Western-Image7125@reddit

You’re kind of combining a few things together. You mentioned “ debugging, designing systems, refining implementations, learning better abstractions” you absolutely 100% will continue to need to do this even in a world where a lot if not all code is auto generated. But if you’re talking about the actual process of typing out lines of code to implement specific features then that is likely not something you’ll make a career from. You might think of woodworking or building novelty cars from scratch as analogies, but unfortunately it doesn’t hold up because physical objects and art you can clearly see and feel the difference vs a machine made item/car whereas a piece of code is going to look and perform exactly the same. No one goes and looks inside a feature to see the code anyway they just want to use it and make sure it works. So I feel like the only things where a human can separate from a machine is in stuff that exists in the real world.

So if you enjoy “solving problems and debugging in general” - that’s not going away not even close, ai generated code is going to create a just of new problems that need competent problem solvers. But if you enjoy the act of pressing your fingers against the keyboard and physically typing out code then you can keep doing just don’t expect people to pay for that specific effort and skill the way they used to.

[-]

ThlintoRatscar@reddit

The unfortunate truth is that we decline. Especially people who love coding in a specific language.

But, if you love to code for the sake of coding, then there will likely always be a place. It won't be like the 90's with what feels like infinite growth, and it won't be an unregulated path to a solid upper middle class professional life, but lots of code is still needed to do things for people.

In terms of your specific questions:

The compiler/AI does the actual coding. Back in the day, assembly programmers existed who took great joy out of writing assembly but they have largely disappeared from our profession. The compilers generate the vast amount of binary code and do it quite well, most of the time.

In terms of what happens, people move on to other work-like hobbies. Developers ( before the great LeetCode FAANG infestation ) are the kinds of people that enjoy a craft, be that craft beer, artisan coffee, fine woodworking, boat repair, off grid design, organic integrated farming, or whatever. Or, they just continue to code in obscure languages for fun and show off the work in multiple online communities. Not as a resume point or a side-gig but simply to show off to their like-minded friends.

[-]

itdoesntmatterrly@reddit

The same with people who loved to count using an abacus, they are not needed anymore.

[-]

Guess-Severe@reddit

The narrative that coding is going away is false. This shit is all going to balance out in the near future. The way marketing hype is elevating LLMs is absurd and not sustainable. These companies are bleeding money.

[-]

bigorangemachine@reddit

Agentic coding can only go so far.

It's a split between how much time are you willing to explain what needs to be done vs how long it takes to get done

So far a Database migration is by far the worst thing to do with agentic coding I've found...

UI you can get something started but anything beyond a page hit and load contents it sucks.

[-]

ZenX22@reddit

What about the people who love anything that used to be manual and is now largely (or entirely) automated?

[-]

Asdas26@reddit

You can still handcraft code. Probably not everything but I think it's good to keep writing code manually from time to time. Really understanding code might become a rare and thus valuable skill in the future.

[-]

Coxian42069@reddit

Someday you'll be an old person saying "back in my day we did it manually" while young people smile and nod, just like how we smiled and nodded at our elders.

[-]

HoratioWobble@reddit

It's a job, the fact you've enjoyed your job up until this point has been great but your job is to provide services in exchange for money, whether you enjoy that or not is irrelevant.

Hobbies are things we enjoy, sometimes you enjoy your job, sometimes you don't

[-]

DateMasamusubi@reddit

Coders will be like telegraph operators when telephones went mainstream. Still some niche uses but it will dwindle and become an antiquated hobby like crochet.

[-]

Chuu@reddit

A lot of people enjoy creating music but there is virtually no money creating music unless you're one of the best in the world at it. Tons of other examples out there, even in some engineering trades.

I would be sad to see coding go the same way, but the trend has always been for code to be representing higher and higher abstractions it is arguably inevitable that one day natural language is the coding language and digging deeper is the domain of a very small number of specialists.

Fwiw, I too love to actually code and selflessly hope that we don't get there in our lifetime.

[-]

08148694@reddit

The same thing that people did when the car and printing press were invented

Get a horse, ride it as a leisure activity

Get a nice pen and practice calligraphy as a hobby

The world changes, you need to adapt with it

[-]

03263@reddit

You can officially get fucked.

Sorry, I'm in the same boat and the outlook is not good.