Microsoft CEO Satya Nadella provided a notable metric for artificial intelligence’s penetration into his company’s software development process, stating that AI is likely responsible for generating between “20%, 30% of the code that is inside of our repos today” was “written by software.”
He shared this estimate during an on-stage conversation with Meta CEO Mark Zuckerberg at Meta’s inaugural LlamaCon AI developer conference in Menlo Park on April 29th. Acknowledging nuances, Nadella mentioned that AI’s effectiveness varies across different programming languages, currently showing more promise with Python compared to C++.
The exchange highlighted differing perspectives and metrics among tech leaders. When Nadella turned the question to Meta’s internal AI code generation, Zuckerberg admitted he lacked a ready figure but pointed towards a future where AI builds AI. “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 projected.
These comments followed Google CEO Sundar Pichai’s statement in late April 2025 that AI was writing more than 30% of new code at his company. However, the differing definitions—Microsoft’s focus on existing repository share versus Google’s on new code and Meta’s goal related to AI model development—make direct percentage comparisons complex.
Microsoft Pushes AI Coding Internally and Externally
Nadella’s internal estimate comes as Microsoft heavily promotes AI coding tools externally, primarily through its subsidiary GitHub. Earlier in April 2025, GitHub enhanced its popular Copilot assistant with an “Agent Mode” capable of executing commands and applying fixes within the VS Code editor.
This upgrade coincided with the launch of a new Copilot Pro+ subscription tier providing access to high-end models from partners like Anthropic and OpenAI, and the general release of its AI Code Review feature. These external product developments align with ambitious internal projections; Microsoft CTO Kevin Scott previously predicted AI could generate as much as 95% of all code by 2030.
Meanwhile, Microsoft’s primary AI partner, OpenAI, continues its own efforts. On April 16th, OpenAI released Codex CLI, an open-source command-line tool emphasizing local developer control, contrasting slightly with GitHub Copilot’s increasing automation.
News also surfaced mid-April that OpenAI was in advanced talks to potentially acquire AI coding startup Windsurf (formerly Codeium) for around $3 billion, a move seen to bolster its capabilities against GitHub Copilot.
Meta Joins the AI Platform Race at LlamaCon
While Nadella focused on internal metrics, Meta used the same LlamaCon event to preview its own Llama API, marking a direct entry into the AI platform services market contested by OpenAI, Google, and others.
This limited preview grants developers direct access to Meta’s Llama models, starting with Llama 3.3 8B, and notably includes tools for fine-tuning – adapting a base model for specific tasks using custom data. Although the initial API access centers on Llama 3.3, Meta is providing experimental access to its newer, multimodal Llama 4 models (Scout and Maverick) via hardware partners Cerebras and Groq, aiming for faster processing speeds.
This developer-focused API launch, detailed further on Amazon Bedrock, coincided with the release of Meta’s standalone Meta AI app, also running on Llama 4 and now serving as the control hub for its Ray-Ban Meta smart glasses.
An Evolving AI Development Ecosystem
The moves by Microsoft and Meta fit into a broader pattern of tech companies deploying increasingly capable AI for software development. Google‘s efforts include its Firebase Studio (launched April 9), which integrates the Gemini AI into a full-stack cloud IDE, and its Gemini Code Assist Free tier, introduced in late February 2025.
Anthropic targeted developers with its Claude Code Assistant in February 2025, focusing on executing complex programming workflows. Elon Musk’s xAI also made its Grok 3 model available via API on April 10, followed by the Grok Studio workspace. Even community-driven, open-source alternatives like Open Codex CLI emerged (April 22), specifically designed for running AI coding assistance locally. This diverse range of tools and platforms underscores the industry-wide significance of the AI code generation metrics discussed by Nadella and his peers.
