Engineer would like to rewrite Windows — in Rust — using AI

1_rick said:

Microsoft denies rewriting Windows 11 using AI after an employee’s “one engineer, one month, one million code” post on LinkedIn causes outrage​

https://www.windowslatest.com/2025/...million-code-post-on-linkedin-causes-outrage/
Click to expand...
As of now it is a Lone LinkedIn Lunatic trying to catch Satya Nadella's eyes in the next quarterly performance review

But if this idea takes wings within Microsoft then you never know

G84gYm4XgAAvmL0.jpeg
 
Nytegard said:
At the very least, you need someone to verify the code and look over it. That alone would take well over a month unless you want an operating system that makes the current version of Windows vastly more secure in comparison.
Click to expand...
Window users are the new alpha and beta users.......
😜
 
Advil said:
  • Individuals and interactions over processes and tools - Maybe they consider this internally, but externally the user experience as an individual and their interaction aren't considered much over processes and tools which seem to be sacrosanct to Microsoft.
Click to expand...
I'm not trying to be a microsoft fanboy. Far from it. But consider the Windows Insider program.

On the other hand, MS forcing features like CoPilot is the triumph of corporate strategy over (individual) user needs.

Advil said:
Click to expand...

Nonetheless, Agile isn't meant to be a straitjacket. It's best practices. And it was developed based on mostly smaller projects. Windows taken as a whole is one of the most complex projects around. And it's not internal to the organization. It's vcery public.
Advil said:
  • Customer collaboration over contract negotiation - Yeah. No. And hell will freeze over before MS starts listening and collaborating directly with customers rather than being all about the contract.
Click to expand...
End users or corporate customers?
Advil said:
  • Responding to change over following a plan - If they believed in this one, a whole lot of features and UI in all their software would have been scrapped (or left alone) over the last 20 years. Following their plans into a pit they have to dig out of later is what they do best.
Click to expand...
Mostly true.
 
Aseem Rastogi, Principal Researcher, and Pantazis Deligiannis, Principal Research Engineer from Microsoft Research FoSSE (Future of Scalable Software Engineering) discuss the technical results from ICSE'2025 on using Large Language Models (LLMs) for safe low-level programming. The results demonstrate LLMs inferring machine-checkable memory safety invariants in legacy C code, and how LLMs assist in fixing compilation errors in Rust codebases.


LLM assistance for memory safety (in C?):
https://www.microsoft.com/en-us/research/publication/llm-assistance-for-memory-safety/

RustAssistant: Using LLMs to fix compilation errors in Rust
https://www.microsoft.com/en-us/res...-llms-to-fix-compilation-errors-in-rust-code/


This session aired on February 25, 2025, at Microsoft Research Forum, Episode 5.


View: https://youtube.com/watch?v=kYGLH-WYPdI&t=125s&pp=2AF9kAIB

https://x.com/_h0x0d_/status/2004634265007579230?s=20
 
"Design Note 1 was intended to be a straw man.

One of the things I do as a researcher is I put out a stake that is very aggressive and causes people to question it and question their understanding and what the space looks like and it provokes discussions."


https://x.com/_h0x0d_/status/2004621307162599427?s=20


Mr. Sinofsky once sent an email to my VP suggesting that I should be fired ......

https://x.com/TimSweeneyEpic/status/2004288064408134109?s=20


https://x.com/stevesi/status/2003553104982344000?s=20
 
Last edited:
MicroSpeak term of the day — North Star

I raised my hand. “What do you mean by north star? Because if you follow the north star, you end up at the north pole, and not where you actually want to go.”​
​
The speaker seemed a bit frustrated by this question. “Who is this idiot who doesn’t know what a north star is? Certainly this person hasn’t been in all the meetings I’ve been in, where people are saying ‘north star’ all over the place.”​
​
The speaker noted that I might want to look it up in the dictionary, because it would have told me that the north star is the goal you have beyond your immediate goal. It’s a guiding principle that keeps you on the right path for your journey. (Curiously, this definition doesn’t appear anywhere in any online dictionary I could find. It also doesn’t match the citation at the top of this article.)​


https://devblogs.microsoft.com/oldnewthing/20151103-00/?p=91861
 
Would "guiding light" have been better? Seems a bit pedantic. It was a metaphor.
 
What a bunch of garbage.
If it were so simple to refactor a million lines of code a month, why are there so many bugs and performance issues in pretty much every MS product. Shouldn't they have already cleaned up their codebase?
 
zerogg said:
What a bunch of garbage.
If it were so simple to refactor a million lines of code a month, why are there so many bugs and performance issues in pretty much every MS product. Shouldn't they have already cleaned up their codebase?
Click to expand...
The main problem here is that in order to achieve this type of metric would require complete faith in the AI, because it's impossible to go that speed. And if the AI was perfect, having it output Rust code is pointless. If the AI can write perfect code, who cares if it writes unsafe code, because it can't make mistakes. Just have it output the Operating System components in pure machine code.
 
Nobu said:
Would "guiding light" have been better? Seems a bit pedantic. It was a metaphor.
Click to expand...
Raymond (who wrote that quote) occasionally complains about people at Microsoft who tend to make up their own meanings for words or phrases that aren't quite the same as the normal usage, but also, that nobody ever defines them. Some manager will just say something and the next thing you know everyone is using the term.

Also, he is more than a bit pedantic.
 
well the initial message can be a bit optimistic, there is really a lot of strange reaction to a c/c++ RUST new automated tools, you can still do giant volume while obviously needing human and not expected it to be perfect.

by 2030 (years of work) and needing to create a whole new tool to do it, do not imply they think it is simple (specially if you mostly avoid c2rust like unsafe rust output).

of those 45,000 line a day, a lot can be boiler plate, but it took years of works making that translator. Has for who will know the complete windows codebase if so much was automatically translated... after a certain size and many decades the answer will always be kind of no-one, all by hands the work would have been split among many.
 
Last edited:
Just one engineer and a team of agents ported the GitHub Copilot agent runtime to Rust, shipping 800,000 lines of production code while retaining code quality.

https://x.com/github/status/2102103572867358977?s=20

The GitHub Copilot CLI, GitHub Copilot app, and GitHub Copilot SDK are all backed by the Copilot agent runtime, an agentic harness that can be embedded into applications and services. It was originally written in TypeScript on Node.js and the V8 JavaScript engine for what is now the GitHub Copilot cloud agent (CCA), and the runtime stayed on that stack as the runtime and its capabilities grew rapidly.​
​
That has now changed. Using the GitHub Copilot app and the Copilot CLI, we completely rewrote the runtime into more than 800,000 lines of production Rust. AI agents wrote most of the code, spanning 128 pull requests that landed in main and shipped incrementally rather than waiting for a single cutover at the end. The few inevitable regressions were discovered and fixed quickly along the way, while the performance of the runtime improved by orders of magnitude. ​

https://github.blog/ai-and-ml/gener...um=social&utm_campaign=github-app-cli-Q1-push
 
Marees said:
Just one engineer and a team of agents ported the GitHub Copilot agent runtime to Rust, shipping 800,000 lines of production code while retaining code quality.

https://x.com/github/status/2102103572867358977?s=20

The GitHub Copilot CLI, GitHub Copilot app, and GitHub Copilot SDK are all backed by the Copilot agent runtime, an agentic harness that can be embedded into applications and services. It was originally written in TypeScript on Node.js and the V8 JavaScript engine for what is now the GitHub Copilot cloud agent (CCA), and the runtime stayed on that stack as the runtime and its capabilities grew rapidly.​
​
That has now changed. Using the GitHub Copilot app and the Copilot CLI, we completely rewrote the runtime into more than 800,000 lines of production Rust. AI agents wrote most of the code, spanning 128 pull requests that landed in main and shipped incrementally rather than waiting for a single cutover at the end. The few inevitable regressions were discovered and fixed quickly along the way, while the performance of the runtime improved by orders of magnitude. ​

https://github.blog/ai-and-ml/gener...um=social&utm_campaign=github-app-cli-Q1-push
Click to expand...

There is also Bun.

https://bun.com/blog/bun-in-rust

You could actually follow the branch on Github while it happened and see the progress.

Their Github in general is pretty cool. At one point before the rewrite shipped, I submitted a bug I found. An agent replied, reproduced the issue, opened a PR with the fix and tests, and it was merged within a few days.

Shockingly, it appears to have been a very successful rewrite. Claude Code runs on it (and ran on the pre-release version for a while).

I don't think this is the go-ahead to just rewrite an entire kernel or OS this way, but it's clear the friction for these kinds of migrations is going waaaay down. Granted, it doesn't neccsarily always produce good code (actually, the migration seemed to produce a lot of dog shit) since going language to language doesn't really always map cleanly, but it's fixable.

At that point, even though you have some patches of hideous garbage, your tests are passing and you've condensed a year timeline into 2 weeks, so... I can see why it's an interesting proposition.

And - agents are getting really good at optimizing now. They're very good at assembly and SIMD, and have no problem literally reviewing the profiler in a loop to experiment. They can even do graphics debugging. Like they'll actually snapshot the window as they work so they can debug something like a game engine. They even understand things like shadow acne and shadow mapping looking slightly weird and all kinds of subtle shit. When you give them a focused task, they're kind of insane. If you set a goal to look at the profiler and try to optimize the hottest function, you will probably wake up to test cases and benchmarks built, and a spreadsheet of shit they tried. It's kind of wild.
 
You must log in or register to reply here.
Back
Top