Table of Contents
Can AI agents safely rewrite millions of lines of code without human oversight?
Recent internal communications from Microsoft indicate a strategic pivot toward aggressive code modernization. Galen Hunt, a Principal Software Engineer at Microsoft, posted a recruitment notice in late December 2025 outlining an ambitious objective: the elimination of C and C++ code from Microsoft repositories by 2030. The proposed method involves a hybrid approach utilizing Artificial Intelligence and algorithmic translation to convert these legacy systems into Rust.
This initiative aligns with CEO Satya Nadella’s broader directive. Nadella has integrated AI mandates across the organization, requiring employees to adopt these tools. Reports suggest that 30% of Microsoft’s code base is currently AI-generated. This top-down pressure has created friction, with some users and developers expressing concern over the quality and security of forced AI implementations.
The “One Million Lines” Metric
The controversy centers on the specific operational metric Hunt proposed: “1 engineer, 1 month, 1 million lines of code.” This ratio suggests a level of automation that minimizes human oversight to a mathematically negligible degree. To achieve this, Microsoft has developed a code processing infrastructure that maps source code into scalable graphs. AI agents then utilize these graphs to perform mass translations.
However, industry experts question the feasibility of this throughput. A human engineer cannot meaningfully review, comprehend, or validate one million lines of code in a single month. The sheer volume implies that the AI would operate with substantial autonomy, raising significant concerns regarding quality control and logic verification.
Clarification: Research vs. Deployment
Following community backlash regarding the safety of rewriting core systems like Windows 11 solely with AI, Hunt issued a clarification. He specified that the initiative is currently a research project rather than an immediate roadmap for consumer products. The goal is to build technology capable of language migration, not to blindly deploy AI-converted code into production environments immediately. This distinction is vital for stakeholders to understand: the infrastructure is being built, but the “rewrite Windows” narrative was a misinterpretation of the project’s current scope.
The Hidden Cost of AI Code Generation
While the promise of AI is speed, current data suggests a trade-off in quality. A December 2025 report from CodeRabbit, an AI-based code review platform, analyzed 470 open-source pull requests (PRs) to compare human and AI output. The findings contradict the efficiency narrative:
- Higher Defect Rate: AI-generated PRs contained an average of 10.83 issues, compared to 6.45 issues in human-generated code.
- Increased Review Load: AI involvement resulted in 1.7 times more errors per submission.
- Complexity of Errors: The bugs found in AI code often involved logic, security, and maintainability, which are harder to detect than syntax errors.
Advisory Summary
For technical leaders and investors, this situation highlights a critical “AI Paradox.” Tools designed to increase productivity may currently decrease it by inflating the time required for code review and remediation. While Microsoft’s ambition to migrate to memory-safe languages like Rust is sound from a security perspective, relying on AI to execute this migration at a rate of one million lines per month presents a high risk of introducing technical debt. The industry consensus remains that AI should assist, not replace, the architectural judgment of experienced engineers.