AI is eating your code for lunch

In an eye-opening fireside chat at Llamicon 2025, Meta's Mark Zuckerberg and Microsoft's Satya Nadella shared a startling prediction: within the next year, AI could be writing half of all code, with that proportion only growing from there. The discussion revealed how these tech giants are already heavily leveraging AI for software development and provided a glimpse into a radically different future of programming.

Key Points

• Microsoft reports that 20-30% of code in some of their repositories is already written by AI, with acceptance rates for AI-generated code steadily increasing across programming languages (though Python sees the highest success rates).

• Meta is focusing on building AI systems that can advance their own LLM development, potentially creating an "intelligence explosion" where AI can self-improve at an accelerating rate.

• Both CEOs envision a complete rethinking of software infrastructure, with developers evolving into "tech leads" orchestrating teams of AI agents rather than writing most code themselves.

• Neither company is dogmatic about closed vs. open-source models – Microsoft is deliberately diversifying beyond their OpenAI partnership to reduce platform risk.

The Code Revolution Happening Under Our Noses

The most striking takeaway from this conversation isn't just the impressive percentage of AI-written code today, but how dramatically the coding paradigm will shift. We're not simply getting better autocomplete – we're witnessing the early stages of a fundamental transformation where humans will primarily design and direct systems rather than implement them.

This matters tremendously because software development has been a bottleneck for innovation across virtually every industry. When Nadella mentions that "the world needs a new factor of production" to achieve 10% growth rates in developed economies, he's acknowledging that AI-powered development could be that catalyst. The productivity implications are staggering when you consider that tools like Cursor are already generating nearly a billion lines of accepted code daily.

Beyond the Transcript: What This Really Means

For working developers, this transition will be both disruptive and opportunity-rich. Consider GitHub's journey: they started with simple code completion, added contextual chat to keep developers in flow, and are now moving toward full agentic workflows where entire PRs can be assigned to AI agents. This progression shows