Is 30% of Microsoft's Code Really AI-Generated?

A few months back, news outlets were buzzing with reports that Satya Nadella claimed 30% of the code in Microsoft's repositories was AI-generated. This fueled the hype around tools like Copilot and Cursor. The implication seemed clear: if Microsoft's developers were now "vibe coding," everyone should embrace the method. I have to admit, for a moment I felt like I was being left behind. When it comes to adopting new technology, I typically choose the slow and careful approach. But suddenly, it seemed like the world was moving on without me.

Here's the thing though, I use Copilot. I use Cursor at work as well. But I can't honestly claim that 30% of my code is AI-generated. For every function an AI generates for me, I spend enough time tweaking and adapting it to our specific use case that I might as well claim authorship. Is that what Microsoft employees are doing? Or are they simply writing prompts or a set of instructions, then letting the LLM write the code, generate the tests, and make the commits entirely on its own?

So I went back to reread what Satya actually said:

I'd say maybe 20%, 30% of the code that is inside of our repos today and some of our projects are probably all written by software.

Fair enough. But then I watched the video where he actually said it. Interestingly, it was Zuckerberg who asked the question. What you hear in the interview is a whole lot of "maybe," "probably," "something like". Not the confidence portrayed in the written headlines.

But here's what I really want to know: how are they tracking this? Are developers labeling all AI-generated code as such? Is there some distinct signature that marks it? How can you even tell when code is AI-generated? Unlike a written article where we can identify clear patterns, telltale phrasing, word choices that deviate from an author's typical style, code doesn't come with obvious fingerprints.

For example, there's no way to tell when a senior developer on my team uses AI. Why? Because they don't commit code they haven't thoroughly reviewed and understood. They treat AI suggestions like rough drafts, useful starting points that require human judgment and refinement. With junior developers, you might occasionally see a utility function defined for absolutely no reason, or overly generic variable names, or unnecessarily verbose implementations that scream "AI-generated." But these issues rarely make it past the code review process, where more experienced eyes catch and correct them before they reach production.

Before LLMs entered the picture, what we worried about was developers copying and pasting code from Stack Overflow without understanding or modifying it. These snippets weren't easy to identify either, unless they broke the logic or introduced bugs that revealed their origin. You couldn't reliably identify copy-pasted code back then, so what makes it any easier to identify AI-generated code now?

Both scenarios involve code that works (at least initially) and follows conventional patterns, making attribution nearly impossible without explicit tracking mechanisms.

The line between "AI-generated" and "human-written" code has become blurrier than the headlines suggest. And maybe that's the point. When AI becomes just another tool in the development workflow, like syntax highlighting or auto-complete, measuring its contribution as a simple percentage might not be meaningful at all.

---