At the LlamaCon conference, Microsoft CEO Satya Nadella revealed that 20% to 30% of Microsoft’s code is now written by software, marking a significant shift in how humans and machines collaborate in software development. This evolution goes beyond just improving speed; it’s changing who builds the code, what’s being built, and how much control developers retain.
Nadella noted that Python has been more adaptable to machine-generated code than stricter languages like C++, showing that the integration of AI in coding varies across technologies.
When Nadella asked Meta’s CEO Mark Zuckerberg about machine-generated code at Meta, Zuckerberg admitted he didn’t know the exact figures but projected that machines would eventually handle half of their coding tasks. In contrast, Google CEO Sundar Pichai recently stated that over 30% of Google’s code is now AI-generated. However, the lack of clear definitions around what qualifies as “machine-generated” code—ranging from autocomplete suggestions to entire functional modules—makes these comparisons murky and potentially misleading.
Microsoft’s CTO Kevin Scott pushed the conversation further, predicting that 95% of code could be machine-generated by 2030. While this highlights growing confidence in AI tools, it also raises concerns among developers about overreliance on automation. As AI becomes embedded in software development processes—from writing and debugging to quality control—the line between assistance and dependency becomes increasingly blurred.
Despite the bold projections, the adoption of AI coding tools is uneven. Some programming languages and teams have embraced the technology, while others resist or experience less benefit. As tech leaders share optimistic stats, there remains uncertainty about what “AI-generated” code really entails. The industry is entering a new era where machines are no longer just helpers but co-authors—and not everyone is comfortable with that level of integration.