Google may be turning Play Store apps into the next big AI dataset

While Anthropic’s Claude Code and GitHub Copilot are in a league of their own with their AI coding tools, Google seems to be running low on free data to train its artificial intelligence — and the search giant is apparently turning to Android developers to fix the problem.

According to a new report from 404 Media, Google has quietly started offering money to select Play Store developers for access to their app source code. The move is reportedly part of a confidential pilot program designed to improve the company’s AI-powered coding tools.

Arguably, the absolute gold standard for real-time code autocomplete today is Microsoft’s GitHub Copilot, and for navigating complex, entire codebases autonomously, Anthropic’s Claude Code reigns supreme. Google’s own Gemini models have struggled to keep up using only the code the company can freely scrape from the open internet. To fill that gap, Google may be honing in on quality, real-world Android apps — even old side projects or archived prototypes gathering dust on developers’ hard drives.

According to one email reviewed by 404 Media, Google is inviting developers to share codebases from Android apps published on the Play Store. The company says the code would be used to “help improve Google’s developer tools and products,” a description that strongly suggests the data could be used to train or fine-tune AI models that are focused on coding.

If you’re an Android developer receiving this email, you get to sell access to your codebase and keep 100% of your intellectual property, states the report. It’s a non-exclusive license, so your app is yours, and you can monetize your data elsewhere too.

Google’s email pitch routes to a page presenting this effort as a “mission-driven opportunity” to help society solve global problems, but the immediate reality is clearly about making Gemini 3.5 and its new Antigravity 2.0 coding agent far more competitive.