This is what I mean, its not a low number far from it. But I know people that only value that number, so previously they churn out a big LOC, higher ups praise. Now its just big number good, bigger number even better. And LLMs do exactly that.

Now LOC comes down to how much tokens you are willing to spend for a given day.

Depends on the kind. 400 locs JS or Java or something is barely worth mentioning. Can do more, a lot of it is boilerplate bs.

400 locs of critical C code, sure.

Leave your card to fix the mess later.

"But I saw on Instagram, with AI it can be done cheaper and faster".

You’re a doctor. What’s your reaction when a patient rejects your diagnosis due to something they saw on Instagram?

The people that would pay me don't care for good docs, eyesore, or if the programmer actually knows how the app works.

Do they care for liability?

with AI it can be done cheaper and faster

And they are probably right. But they are probably also missing that this will translate to worse quality. You know, the classic "you can only pick two among good, cheap and fast"

well even if the results are crap, it still has big economic influence:

"since I can use AI to create this app this much, I'll pay you less"

"although current AI made this crap, it'll outpace you eventually, so I'll pay you less"

The same feeling, do AI knows the hardware? Does it have the context of the changes? Did it read Linus decisions why something should be the way he thinks is correct? I would prefer such approach where AI knows so much context as Linus and the most important persons in the kernel team

That's a good question. I don't have anything to critique regarding consistency or quality. You are right to say that the newest frontier models are quite good on that front. Especially when you are doing a task you have little experience in.

I like programming, it fun! Using AI feels like I am optimizing the fun out of it. I learn so much about different programming languages and topics I would simply not learn when using AI. I also like learning things!

In my case, it would be to make sure I have a good mental model of the problem at hand instead of offloading it to a LLM. I'd also rather not put myself in a position where I'm reliable on a LLM to do my job.

Oh and there are also the huge moral and ethical issues with the companies that offer LLMs.

Yes, people who know ASM are in demand in a tiny very small portion of dev work. And even in most of their main areas like embedded a lot is moving to HLLs these days so even that is leaving but you don't hire asm devs to build your product anymore maybe a feature or something because they are just not productive enough.

When C and C++ came out we had a whole new class of problems too with people not using them correctly and over time best practices developed to make it safer and better quality. The same thing is happening today with LLMs.

I code because the software i need is not there yet and these tools still fail at doing those.

Not sure why the next xy (already done) app has any economic value outside being a cheaper clone.

Still there is nothing wrong with having an AI point out problems which i assume is the case here and someone still is responsible and knowledgeable about the patched (!) code.

At the very least, AI can review all the code you have written. You can still reject every suggestion, but in the worst it will just slow you down. No citation needed for that. If an AI were to truly confuse you sufficiently to accept a wrong suggestion based on phony logic, then it just shows you didn't grok the logic deeply in the first place. (Which is a helpful finding for the second iteration.)

You didn't say if you did, but just in case if you truly never had a serious look at LLM tools, you really need to step up. Never even trying something like Claude Code borders on negligence nowadays.

AI is still a bubble, but pretty much every software developer having a hard-dependency on these companies nowadays, is was sustains it. It sucks, but currently there is no realistic way around it.

Exceptions apply for data protection reasons, obviously don't upload all your personal data and business-critical intellectual property into some 3rd-party AI tool.

It very well may. Quite a few among those that have been fired because of AI (or at least, with the companies claiming AI as the cause).

But what is that cobbler supposed to do during that time?

Optimally? Move to the best available trade that appears not to be deprecated for a good 50 years in the future, and in which they'd hire someone of his background.

Please don’t do this if your livelihood depends on it. You’re just going to make your life more painful.

As a professional programmer, you’re selling a service. The people buying your service have fully bought into AIs and accepted their shortcomings. You can’t fight this.

You can still scratch your itch with hobby apps, or contributing to human-first projects like Zig.

But what is the cost?

The cost is similar to doing any other task in a rushed, careless manner, with little regard for the consequences.

sometimes I feel like I am being left behind.

And why does that bother you? “Forward” does not equal “good” or “desirable”. If a van is driving towards a cliff at high speed, would you not rather be left behind? “Being left behind” is a talking point to cause you FOMO, don’t fall for it.

In my case very much yes to both questions.

Hey we should see this increase of so called productivity in the global GDP growth numbers. If AI really is this productive the slope of GDP growth should be steeper the last 2~ 3 years. If there is no change in the GDP growth slope, there's no general productivity improvement.

The code itself indeed has become cheap but the question still is out whether it solves the right problem. I've shifted my work significantly toward designing more high-level portions where multiple things would interact in somewhat unpredictable manner (read as failure modes) that's where I found the models being quite poor, they do suggest many things to solve these or that problems but in a unity these suggestions miss second order effects. I was quite surprised that models can't hold all of the connected pieces simultaneously to reason about. But once I have a complete design after a week of deliberations and various attempts, the models can produce lots of correct code to cover that. And then another pass starts because they still deviate in implementation of various pieces and once you start questioning those the models start self-correcting. Judging the output is still way faster than writing the code yourself and indeed models still show some cases that slipped once mind. It's a definitive increase in terms of quality and value of the end code no doubt. Whether it can one-shot a complex problem, not really.

Code is merely the means to an end. It does not matter who writes it as long as it does what you want. Correctly.

Stands to reason though that someone who knows programming AND has domain knowledge can get LLMs to produce much better output compared to someone who does not.

I never used to have time to make all the stuff I needed or was interested in. With LLMs, I can.

This "shipped a whole app while being in the gym" does not work for me though. It takes me a couple of days to a week to produce solid, functional software (~10KLOC). Simple tools (3-400LOC)? Yeah, those you can produce in 30-60 minutes.

As is with any engineering… a key skill is knowing which tool to use when and for what.

There will always be people who push the extreme limits of either way. “I will not use AI for anything!” Vs “I will vibe code this while I’m at the gym”.

The winner is the one that knows when and how to use AI as a _tool_ and focus on the better outcome.

Building a business on vibe coded software feels like building your house on sand.

Sooner or later you'll end up troubleshooting something you just don't understand.

And the tricky part about AI is that it only has to make one bad decision confidently, to then base all future decisions on that one bad one.

AI-assisted coding helps me a lot to churn through boilerplate, straightforward implementations, test writing, exploration of unfamiliar codebases, and other ancillary tasks in coding. It does not replace tacit knowledge/domain knowledge required to design a good solution, no matter how much of the architecture is described as documentation it cannot predict the future to have judgement about a proposed design.

In general AI is an assistance, it doesn't replace the intellectual work that is needed for a good implementation.

In my opinion, approach it like you'd approach any other tool: there are good use-cases for it while being definitely not the right tool for other cases.

I think if you review the code and keep a good mental model of the codebase, the costs are pretty minimal but the value is huge. I'm not going back...

You’re missing out on people getting better at working with ai to improve better long term goals, like testing suites and guardrails, velocity is inconsistent in my experience.

And vibe coding apps in the gym, is like playing a game by typing a cheat code, honestly it’s not satisfying

Or maybe.. people like the tech and use them? Instead of all these theories?

No issues, glad that we agree. I personally do use it for generative and I fully see the value for my CRUD engineer type of use. I think kernel development requires more scrutiny and we are not there yet where we can blindly use LLMs for Linux.

But my main point is that it is a bit strange to see these patches that are self disclosed as AI use and then come to conclusion that they were pressured to use it rather then use it because it is useful. I have never come across this kind of pressure, which is why I call it a conspiracy theory.

How big is your circle? What industries does it span? Maybe the people writing 0% code manually are outliers and your circle is relatively small?

Small changes can have a huge impact.. far beyond your context window ;-)

My follow up would be - how many are unemployed vs employed. From what I do know, most linux devs also work someplace, so either directly or indirectly have the means for a 20-100 dollar subscription a month.

More than that, I've not seen people do agentic coding in one part of their work, and manual coding in another. Is there any first hand account of someone doing this? It kind of rewires your brain, and makes it hard to do both at the same time.

I would assume that linux devs are relatively older. Furthermore they are probably very knowledgeable in that area, and one thing I've noted on here is people who are knowledgeable not liking the quality of code output from LLMs, if they have stringent quality requirements it might be easier for them to write the code themselves.

Embedded code and systems code tends to have a lot of extra constraints the model doesn't know about and aren't necessarily written down where it can see. This produces a higher rate of "superficially plausible but doesn't actually work" code.

(There are probably wider benchmarks, but this was a definite finding in our company wide evaluation of effectiveness. LLMs aren't great at RTL either)

The Linux kernel is essentially the nightmare scenario for a LLM.

No tests, not much documentation to work with, a very high acceptability bar, multiple ways to do the smallest change and every single word on your commit messages is very important.

You can start with a LLM but there going to be some manual modifications you will have to make, even with the best model available

Just thinking loudly here, I guess Co-Author in commits is a hint. PRs where LLMs have been involved is probably another measurement. Github Copoilot has its own account too.

They require disclosure of what tool was used if submitting AI-helped code.

(Which probably means this is slightly underreported from people not disclosing.)

Yeah, I wonder which part of this figure is people doing "let me ask the LLM to do X... aw shucks, that's not how you do X you dumb clanker, fine, I'll do it myself" and not deleting the metadata left by the LLM in the initial draft (a very common scenario in my personal experience). Or people using LLMs for "dumb" large scale changes instead of conjuring an ad-hoc grep+sed script.

Midwits can only downvote when they can't engage in a convo/debate.

Yes. Codeberg running on top of linux and git just shows that they are huge hypocrites.

I still don't get how are they dealing with copyright here. Looks like the ostrich algorithm (hide your head in the sand and pretend nothing happens).

Those patches are not copyrightable unless human made. There's a reason why Oracle doesn't accept AI contributions to Java.