Link: https://www.windowslatest.com/2026/09/05/microsoft-distinguished-engineer-says-typing-code-is-absolutely-over-and-windows-11-is-already-being-built-that-way/

Full title: Microsoft's distinguished engineer says "typing code is absolutely over," and Windows 11 is already being built that way. Nadella previously said 20-30% code at MSFT is AI-coded, and Windows security updates now include AI-assisted fixes to fight AI-enabled threats.

Lol is honestly all I have to say.

AI is like a band saw. It’s really easy to use it wrong and start missing fingers when you get cocky and move too fast.

Then you have people who know how to use it like Jimmy DiResta who will just freehand stuff on one and still has 10 fingers: https://youtu.be/gA4lpsEWM5k

From a statement like this you can’t tell which one it is. Usually the first.

Omg nope nope nope. His fingers are constantly right next to the saw

And still he made his career with doing stuff on a band saw in a New York basement pretty much 😅

Dude knows exactly what he’s doing.

Well… yeah, you can see it in the dumpster fires they call updates.

I don’t think it’s the flex Stale Nutella thinks it is…

I don't remember this many bad updates in a while...now it seems like every patch Tues that roles around has broken shit in it.

Clicking the start menu now results in literally 20 seconds of nothing until the apps appear. Even typing "excel" does nothing for about the same time. It is hot garbage.

I keep seeing people say this. I haven't seen any lag in the start menu. What are you running on? A 5400 rpm drive?

No, absolutely not. Windows still has major bugs like launching desktop apps from a shortcut drops environment variables when the environment path exceeds 2048 bytes. If AI was as good as they claim, it would have found these bugs already.

Hard agree with him now. I'm running multiple opus agents in parallel with occasional fable reviews and plans. I review and correct and guide but I could never come close to this level of real productivity without it.

The abstraction level has risen up to interfaces and architecture and I'm good at both.

they cant change fundametals - device working against me is not MY device.
which is why - doesnt matter if they rewrite everyting in rust, it wont sell.
people know nowadays

Well I guess if you're not experiencing lag, then we're all good.

Being a "distinguished" engineer has nothing to do with being susceptible to the LLM yes man machine.

These chat bots are spreading psychosis on a scale so massive the damage is gonna take years to undo.

Idk how people can type this shit out and not feel disgusting.

Productivity for what. You've outsourced your thinking to the gambling prompt machine. Yikes.

I have two windows 11 machines, both from work. The potent machine experiences this bug. The low-end Thinkpad shows the apps instantly.

No clue why they behave differently, but they do

It's funny you think it's going to be undone. We are not going back to a time where humans type code for any serious commercial product. As they said in the article:

Fowler isn’t saying developers will stop reading code, or that software engineering is done. His point is that manually typing code line by line, in an editor, is becoming the least interesting part of the job.

But there will be a point where very little code is written line by line by humans. Though I suppose most of it was already copy pasted anyway.

It doesn't matter if AI is as good as they claim. Right now, and every new moment, is the worst AI will ever be. Don't be shortsighted and assume since it can't currently do X it will never do X. 2GB of RAM was once a lot of RAM. At the time, we would have said what on earth would anyone ever need 32GB of RAM for.

Not to mention who knows how many bugs AI has already found and fixed in Microslops code. It's already in the thousands, that's for sure.

And before someone starts throwing insults for me "defending" AI, I am strictly talking about AI as a technology. I am not talking about what the mega corps are doing to create said AI. The legislature needs to be quick to stop these companies from destroying the planet trying to innovate AI the fastest. Despite that, I can still recognize AI as the innovation that it is. There are a lot of things AI does not belong in, but unfortunately writing code is one of the perfect use cases for AI.

It's much more responsive on my machine when I disable windows search indexer and superfetch (which I do anyway)

When did I even say that? How did you even extrapolate that from what I said?

You jumped to a conclusion so wildly I have no idea how you even got there.

sadly they don’t

anti ai is trending but that’ll go away with time

the actually caring about the underlying problems will be like the fight for privacy now

Though I suppose most of it was already copy pasted anyway.

Wtf kind of code are they taking about? Like first year "pomodoro tracker" projects? These days the only thing I really copy paste are Linux commands, and even then

Adding on to the other persons reply,

The actual code has always been the least interesting part to me at least. That doest mean I'm going to replace it all by AI.

That's a huge jump there too.

Yeah like a band saw that builds a profile of every interaction on the internet to sell to advertisers and fascist governments. Use with care!!

Stale Nutela is a greta pun!)
Also, agree

I know some of these words.

Doesn't look like it. Outsourcing thinking is one thing: "hey, ai, implement this feature, no mistakes --> straight to production", outsourcing writing code is another thing: "hey, ai, write such and such code --> review --> this is not needed, that should be done this way --> rinse and repeat, then testing, and only then production"

Edit: problem is, I have no sensible ground to believe enough people at microslop do the latter. Nor am I optimistic about this percentage in general, including people of other professions

Tired of a program that used to take up 100 megs of memory!? Now with the help of AI that same program takes up two gigabytes. More is better!

Reading AI code all day absolutely sucks and is the least interesting part of the job. Reading human code is so much better.

Also in all the places I worked 99% of the code was not copy-pasted. Beginners copy paste in their uni assignments, but once you are a bit comfortable with code it's much slower than just writing out exactly what you want!

I don't think that fully, 100% AI-written code is going to be the future and humans are only going to read it. I believe that (expert) humans still do a much better job in structuring code for long-term maintainability and thus there's still going to be a niche for software developers there.

But I also have to agree with you that unfortunately most of the code will still be AI generated slop.

AI is absolutely outsourcing whatever part of your thinking you assign it to. That's exactly why studies have already shown multiple times that AI use is associated with skill atrophy.

It's not permanent, but we can't keep ignoring it.

https://pmc.ncbi.nlm.nih.gov/articles/PMC11239631/

I write code myself all day, as I have always done. And more and more software engineers are finally starting to see the consequences of relying on LLMs for software development.

Writing code has never been the bottleneck.

These chat bots are spreading psychosis on a scale so massive the damage is gonna take years to undo.

take years to undo

You said it would take years to undo, saying it will be undone. What were you talking about being undone in a thread about code being written by AI, if it wasn't about undoing code writing AI? If the psychosis is induced by AI, wouldn't the usage of said AI have to cease?

Also in all the places I worked 99% of the code was not copy-pasted.

Yea, it was attempted humor. The classic "copy paste" joke, though I admit it didn't read well.

AI generated slop

It might be slop now, but we can't assume it will always be that. If AI gets to a point where large pieces of code can be written almost instantly with zero bugs, then we will definitely have code written almost exclusively by AI. I agree that some of the very complex stuff would actually be cheaper to pay someone than pay the for the computational power for an AI to do it, but that is probably about it.

Whatever people feel about that is totally fair. I am not saying I think it's better, I just see that is the ultimate outcome. And an outcome I think most of us will live to see in the not so distant future.

I write code myself all day, as I have always done

That's great. If you enjoy it, do what you love

more and more software engineers are finally starting to see the consequences of relying on LLMs for software development.

And they are adapting how they use said LLMs, not ending its use. People are developing and refining it constantly. They will keep refining until they get consistently good results.

Writing code has never been the bottleneck.

The problem is always ultimately with management. But if management is able to give an impossible deadline, and AI is able to meet that deadline, wouldn't that make a human writing, testing, refining the code the bottleneck? Management could make bad decisions, see the results of those decisions very quickly after AI writes it, and have the AI revise it (wash rinse repeat) much faster than humans could write and revise the same code. To management's dismay, humans are also more likely to give their opinions about how their idea is unrealistic. Management doesn't care what you think, they want you to do what they say. And AI does just that. Management's poor decisions would have even fewer consequences since it could be revised so quickly.

We are talking different languages here. You: "if you let AI do your thinking for you, your ability will decrease". True, of course. Me: "you can use AI to help without outsourcing actual thinking process to it". Also true. And of course it is true that too many people are taking the shortcut and this should be addressed

They should go back to typing code.

Want to see a neat trick? I know for sure this will work on VMware if anybody has it. Use packer to make a custom Windows server 2025 template, make sure to patch it too. Deploy said template. Now watch as you have no GUI. They updated a bunch of UI components, swapping out for 'modern' versions. And someone about these Appx versions get fucked when sysprep runs. It's been like this for months.

I have no idea if this works on other platforms.

I said the psychosis would take years to undo.

Holy crap, reading comprehension is fundamental dude.

Except what is the functional difference? Please elaborate.

You still interact with LLMs the same exact way, you are still prompting, waiting for a result, then getting dopamine when it's right, and then prompting again when it's wrong.

That's literally the same loop gambling uses, and then the AI will stack on "great job, you're so fucking smart and handsome".

People stop thinking, and use AI. It feels like a friend, like a colleague, like it was trained on the most persuasive writing in human history and thent it blows that smoke up your ass.

As AIs are designed now, literally no one is immune from the psychosis, the yes man syndrome, or just being human(lazy) and letting the AI do it.

You cannot trust people to think for themselves, we already have issues with that and now we distributed chatbots to everyone so learning is even less necessary.

Fair point 😊 Thanks for the clarification!

nd AI is able to meet that deadline
Software engineer here of 25+ years here. "AI is able to meet that deadline" and "Management could make bad decisions, see the results of those decisions very quickly after AI writes it" made me lol. Thanks.

Wall-E and Idiocracy are becoming more real every day. Over time there will be fewer people capable of understanding the code produced by AI. If this process repeats over decades and if AI isn't able to recursively improve itself without model collapse, eventually we will have the majority of software being built by model collapsing AI with almost no one left to be able to maintain it or even comprehend it. Recipe for disaster...

My work is going this way too. If you try to think for yourself, all the reviewers Claude agents make the same comments and suggest the same dumb defensive code with a paragraph of comments above it. The only way to avoid it is just to create the code with your own Claude agent and let it fix whatever nonsense your own agent-reviewer comes up with.

To refute some idiot thing the agents all came up with in a review requires me to dig through dependency source code for an hour to see that its just a hallucination, but then at the end of the week my PR numbers are down because I went on a half dozen wild goose chases at the direction of someone's Claude agent. Maybe 1/5 are real. I'm at the point where IDGAF and I just cargo cult the change and I'm already figuring out what my next career is and building a place in the Philippines where I can live for a few years on minimal income. This shit is seriously broken.

I don't even like reviewing ai code generated at my direction, reviewing other peoples is awful

Did you ever think that maybe opus is producing rubbish that nobody wants to read, and you are tempted at accept it for the sake of "productivity"?

If some piece of training data worked around some problem in an old dependency, well now all the generated code is going to build in that same workaround and write out comments explaining it whether it applies anymore or not. Its like superstition coding, and eliminates the possibility of creating optimized code for what you need without overly defensive or complex implementation, and against the current state of your dependencies. The opus code might come faster, and it might work, but the only way to review and maintain it is going to be opus or the next model. I'm not criticizing you, were all being pushed into using it this way, but a part of the creativity is gone and ceded to ai companies

Could be that the OS looks at the lower end machine specs and decides ahead of time it can't run all the extra stuff where the higher end machine gets over the line and runs it, or it could be something the sysadmins did since they're work machines. No way to know without the source code which we won't get.

what

^ This is what AI does to your brain.

Hey, buddy, if you stop using AI now, you'll probably gain like twice as many folds.

How would you undo the psychosis while still using the thing inducing psychosis??

You cannot trust people to think for themselves, we already have issues with that and now we distributed chatbots to everyone so learning is even less necessary.

True that. No question here

Except what is the functional difference? Please elaborate.

I am not "getting dopamine when it’s right, and then prompting again when it’s wrong", I am just offloading the work of typing the text. What the code should do, how it should do it, how much effort to spend on processing this or that edge case/possible problem - still all me. Another version of the same: I don't care that an llm also writes tests and they are green: who cares if they are green when the core logic can be still all fucked up