>>107943674
No wonder I keep finding bugs in the most unsuspecting places

>>107943674
I really think that at some point this year one of these FAANG companies is going to ban the use of LLMs on their codebase
The issues are stacking, and you simply can't trust devs to not use these tools to be lazy, it's human nature

the suits thought this would make them 10x more productive, but reality is it's just making people lazy and sloppy with what they push to prod

>>107943674
literally happened to me an hour ago with the snipping tool and had to fuck around for half an hour trying to make it work, great timing nutella I was right in the middle of discussing screenshots of plane tickets which were getting more expensive by the minute

AI is fine for the first 50 lines of code and setup but on every function you add onto your project the motherfucker removes/modifies already working code.

>>107943674
>nobody knows how anything actually works and Microsoft's code just turns into AI being asked to fix the fuckups of AI being asked to fix the fuckups of AI
It's perfect for business software to be absolute mystery meat with no guaranteed functionality or security.

>>107943817
skill issue

oh, it shows

File: 1768186166283.png (354 KB, 986x456)

354 KB PNG

>>107943674
>>107943834
>>107943878

>>107943896
i bet google paid him well for this ad

Line goes up, shareholders are happy.

>>107943896
linus is a known retard

>>107943761
Shit hasn’t gotten bad enough. I don’t think it’ll be this year, but I think it’ll be after some catastrophic, publicized failures where they will try to hide the cause was ai slop but we’ll know

>>107943817
>the motherfucker removes/modifies already working code.
Never run agents. It's like letting a 10 year old on LSD run wild with your code. Only ever use conversational modes where you click accept for every code change.

>>107943896
anon, did you even read the shit you posted?
>which was also generated with help from google, but of the normal kind
>I had to figure out what the problem with using the builtin rectangle select was

>>107944396
You can easily check the diff and approve or disapprove of any change for any agentic model. It's just like maintaining a git repo with contributors

>>107943896
torvalds never has been a good programmer. he had great ideas with linux and git which translated into functioning things which are well maintained but the code he wrote is far from phenomenal. someone like fabrice bellard or jeff dean are great programmers in comparison to him.

>>107943896
this was on a side project written in python
he doesn't know python so i imagine the machine could write better python code than him

>>107945064
What makes them great programmers? How can I learn that?

>>107943674
thats why it just werks.

>>107944497
Welcome to 4chan, where anons cherry pick everything. If they posted the whole thing it'd show where Linus says he doesn't know python language

>>107943674
I remember this every time a Microsoft update breaks any Microsoft product.
Which is practically daily at this point.

>>107943896
What was it merged to?
Not the Linux Kernel.

>>107943896
https://github.com/torvalds/AudioNoise

I work at a mid size shop (~500 engineers) . This quarter management let go 300 of them. Literally all intermediates and 90% of juniors were let go, seniors were largely spared with only about 15% let go. All openings were removed from the site and instead they put on more openings for seniors.

I had a conversation with the CTO and CEO over lunch (I joined when it was just a startup 9 years ago so I know them personally) they told me that they found juniors and intermediates had a lot more issues productively using AI agents. Juniors and intermediates still needed to write code by hand while seniors had enough experience to let agents do 100% of the code as they have enough experience to describe the entire stack + patterns + algorithms + architectural angle of attack to Claude Code so that it could write practically bug free highly performance code.

It actually makes sense when I think about it. I barely delegated work to juniors in my team and just spun up a couple of independent Claude Code sessions instead.

On the point of maintenance the CTO and CEO also had great points that I didn't even consider. The CTO pointed out that by the time maintenance becomes a true issue which on large codebases is 2 - 5 years after being written on average the models will already be good enough to maintain the code of earlier models.

CEO added that the clients don't care as long as the code works in the moment and the money keeps flowing in so it's a no-brainer. Insurance covers both companies in case of catastrophic failure so neither the client nor us actually care as long as it works.

Client actually upped the contract because we can deliver faster now.

As a principal I probably have a few more years left but honestly it's just a matter of time before software engineering in its entirety is gone. I mean the entire end-to-end pipeline. The faster we just come to terms with this the better.

File: Screenshot 2026-01-23 at (...).png (427 KB, 1164x1020)

427 KB PNG

>>107943674
Glad that I use Windows 10 IoT LTSC for gaming and macOS for working.
When I buy a AMD GPU in the next years I will switch to Linux for gaming.

>>107943674
The other 70% is written by jeets, so likely worse.
Microsoft is one of the companies where vode quality might go up with AI

>>107946519
>As a principal I probably have a few more years left but honestly it's just a matter of time before software engineering in its entirety is gone.
I don't code for a living but, as someone working on the chip design world which (at least until now) was ahead of the curve in terms of automation, I don't think it will go away, it will just change.
At the start, all transistors were laid out by hand, even for digital circuits, and that process of laying them out was a huge part of the chip design workflow.
Nowadays, however, this part of the workflow has been so automated that digital chip designers don't do it at all anymore and, instead, focus on designing the logic circuitry and clocks. The physical implementation of those circuits is then done by few specialized engineers (usually called a Physical Design or Physical Implementation engineer) who are experts at using the automated layout tools to carry out the layout and routing of millions of transistors without breaking anything. This has allowed digital circuits to reach levels of complexity and scale that were unachievable before those tools existed.
I think programming is going through what digital circuit design went through decades ago. In the future, there will be software system design engineers who will think at a higher level than code to design systems that seem impossible now, and then specialized software implementation engineers will use the AI tools to turn that design into code.

>>107943674
horrible day to be a winjeet, there's always the rope though

>>107946606
Hey don't worry I'm actually old enough to have had silicon design in my computer science curriculum where we had to make our own computer from discrete transistors making the ALU, registers, decoders by hand. I even had a floating point unit made from analog comparator which was pretty innovative back then. My graduation project and thesis was a verilog FPGA block that mined bitcoin back in 2010. Honestly if I had just focused on that instead of thinking about it as a toy I might have been a billionaire now.

That said my point is that yes I understand the layers of abstractions and automating one layer just exposes the next layer of abstraction. However I think AI technology is the first time where the next layers of abstraction can immediately be done by AI as well. Right now of course the layer of abstraction is managing AI agents and verifying output which requires you to be a senior with a lot of experience. However there are just some loose ends that can probably be glued together to make it go end to end. Meaning business requirements, scoping, architecture, patterns, algorithms, unit testing, QA and maintenance planning and execution will all be managed by an AI agent. What you will then see is the entire "job" can just be reduced to the CTO managing the entire technology department on his own. Or in extreme cases even by a non-technical CEO. The layers of abstraction being added so high it needs essentially 0 human input anymore. I can see that being just a couple of years away.

Interns were fucked 2 years ago. Juniors last year. Intermediates this year. Seniors next year and then it's me on the chopping block the year after that. Pattern will probably be the same for most white collar work.