CEO Satya Nadella says up to 30% of Microsoft's code is now written by AI

Skye Jacobs

Posts: 663   +15
Staff
The big picture: The rise of AI in software engineering has sparked debate about the future of programming jobs, especially for those entering the field. While the percentage of AI-written code is likely to continue evolving, the need for skilled human judgment is not expected to disappear anytime soon.

At the recent LlamaCon conference, Microsoft CEO Satya Nadella offered a glimpse into how deeply AI has become woven into the fabric of modern software development. "I'd say maybe 20 percent, 30 percent of the code that is inside of our repos today and some of our projects are probably all written by software," he said.

Nadella joined Meta CEO Mark Zuckerberg on stage to discuss AI's evolving role, particularly in the context of open-source innovation. Their conversation underscored a broader trend: AI is no longer a futuristic concept but a practical tool already reshaping how software is built at the world's largest technology companies.

Nadella asked Zuckerberg what percentage of Meta's code is generated by AI. Zuckerberg said he did not have an exact number, but explained that Meta is working on developing an AI system capable of creating future versions of its Llama models.

"Our bet is sort of that in the next year probably … maybe half the development is going to be done by AI, as opposed to people, and then that will just kind of increase from there," Zuckerberg said.

Microsoft and Meta are not alone in using AI to write code. Last year, Google CEO Sundar Pichai revealed that AI generates a quarter of new code at the tech giant. Earlier this month, Shopify CEO Tobi Lütke informed staff that teams must demonstrate why AI cannot accomplish a task before seeking approval to add more employees. In a similar move, Duolingo CEO Luis von Ahn announced in a company-wide memo that the language-learning platform will gradually reduce its reliance on contractors for work that can be automated, as part of a shift to an "AI-first" approach.

It is not surprising that these tech firms are turning to AI, the primary value of which is handling repetitive, data-heavy, and predictable programming tasks. By automating these areas, companies can boost efficiency and reduce the need for entry-level coding positions – to the dismay of new graduates and aspiring software engineers.

Nadella noted that Microsoft has seen the most success with AI-generated code in Python, a language known for its straightforward syntax and flexible design. In contrast, languages like C and C++ present greater challenges for automation due to their complexity and their need for precise memory management.

Despite the growing reliance on AI, Nadella emphasized that human oversight remains crucial. While AI-generated code has advanced significantly, experienced engineers must review and refine what the algorithms produce.

As Nadella put it, certain critical projects, such as major updates to the Windows operating system, are unlikely to be entrusted to AI alone.

Permalink to story:

 
"CEO Satya Nadella says up to 30% of Microsoft's code is now written by AI."
Yes, We can tell.
I use Linux at home but my work computer is Windows 11. I've been getting this feeling that my work computer has been becoming more annoying over the last 3-4 months, I wonder if this is why.
 
Yeah, right, written by AI, then rewritten by humans, because what AI writes is garbage, it can only handle the most basic tasks, it cannot write complex algorithms or handle complex data flows, much less design them.

There is no software out there that's written by AI. The furtherst AI can get you - scaffold some generic stuff for you, and that's it. AI cannot produce code that implements business model of your company / product.
 
"AI" - which it isn't - is only as good as the data it has access to. This means any errors, incorrect data or incomplete data is what it has to works with. I wouldn't trust "AI" to writing any kind of program, let alone one to handle code for a large corporation.
 
More than 30% of the code is boilerplate / templates / etc., and none of that requires AI.
So the percentage is probably higher and will keep getting higher, but that's kinda normal.

Programming has always been that way, less and less code in lower level languages, more in higher level languages, requiring less and less narrow expertise for the vast majority of the tasks.
The gist of programming is not cranking out source code.

 
Good idea, we'll let AI do and run everything while we sit and let our brains atrophied into masses of liquid goo, becoming mindless drones subject to AI's whims. Why are we handing over our lives and work to AI? Insanity!!!
 
Written by AI trained on what kind of source code? If Windows 10 or 11 code was used to train the AI, we can look forward to still more gaping security holes, carrying forward a unique Windows feature.
 
Ah that must be why my start menu stopped working.
I dare them to let AI write the kernel code, see what happens.

So far my experience with AI code generation is that replaces search and replace. Is great for quickly creating templates/boiler plate and can be useful as a sort of pair-programmer approach.
Sort of like an advanced search and replace, Google and advanced autocomplete all in one.

However letting it straight up generate functional code usually results in having to debug and rewrite parts and the instructions have to be explicit to the point where writing it yourself isn't much slower. Alternatively writing a crude not quite working approach with some strategic comments of "// this should do that" and then letting the AI complete it can be fairly effective in some cases.
 
Yeah, right, written by AI, then rewritten by humans, because what AI writes is garbage, it can only handle the most basic tasks, it cannot write complex algorithms or handle complex data flows, much less design them.

There is no software out there that's written by AI. The furtherst AI can get you - scaffold some generic stuff for you, and that's it. AI cannot produce code that implements business model of your company / product.

Kind of nonsense what you write here. The public AI we see or test / feel is nothing compared to what company's like Google is cooking up.

You can pretty much with one prompt now, have AI generate a whole thing for you. I.e "Make me a copy of uber eats app including website but name it VitalyEats" and the thing spits out an app, an site and all the inner workings you could wish for exactly as the uber eats one.

 
If what AI sees is a problem and a solution, how far could it be to learn its own DNA? Solve it ?

 
Back