Artificial intelligence writes 75% of new code for Google under the supervision of engineers
Most of Google’s software code is now generated using artificial intelligence, although human engineers are still responsible for its final review.
Most of Google’s software code is now generated using artificial intelligence, although human engineers are still responsible for its final review.
This was announced by Google, Business Insider reports.
According to the latest data, the share of code generated by Gemini algorithms is stable at 75%. This indicates the full implementation of the strategy started back in 2024, when this indicator was only a quarter of the total volume.
Google CEO Sundar Pichai noted that the company has fully adapted to «autonomous workflows.» According to him, the use of AI agents allows to close technical debts and carry out migrations of systems that were previously considered too resource-intensive.
«Complex code migration processes that used to take years are now completed six times faster thanks to the collaborative work of agents and engineers,» Pichai emphasized.
Google is not the only company that has transformed its approach to programming. Microsoft, as its CTO Kevin Scott predicted a few years ago, is moving closer to near-complete automation of code writing in many departments. Meta, according to its plans for 2026, has already engaged more than 65% of its engineers to write 75% of its code exclusively with the help of AI.
The introduction of these tools was accompanied by internal changes in the development culture. The use of AI tools such as Gemini Code Assist or Claude Code became a mandatory KPI, which directly affects the performance evaluation and career growth of Google employees.
It was previously reported that a group of Apple engineers working on Siri would be sent on a «multi-week bootcamp to learn how to program with AI.»
Read the country’s main IT news in our Telegram
Read the country’s main IT news in our Telegram
Anthropic releases Claude Opus 4.7: the model writes code and processes images better
Anthropic releases Claude Opus 4.7: the model writes code and processes images better
Apple sends Siri developers to boot camp to learn how to program with AI
Apple sends Siri developers to boot camp to learn how to program with AI
Former OpenAI employees said that Sam Altman has almost no programming skills and a poor understanding of machine learning.
Former OpenAI employees said that Sam Altman has almost no programming skills and a poor understanding of machine learning.
