Google uses AI for half of its code, and it’s pushing for even more

The strike team is reportedly led by Sebastien Borgeaud, a Google DeepMind research engineer who has also led pretraining for DeepMind in the past. Other notable contributors include DeepMind Chief Technology Officer Koray Kavcuoglu and Google co-founder Sergey Brin. Information Google DeepMind engineers say they believe Anthropic’s latest cloud code models outperform similar Gemini models, prompting the company’s latest push.

Google co-founder Brin stressed the urgency of the strike team’s mission in a leaked internal memo Information. “To win the final sprint, we urgently need to close the gap in agentic execution and turn our models into primary developers,” Brin wrote. To do this, Google plans to build AI coding models specifically for internal use and double down on agentive functionality.

While the Gemini, DeepMind, and Google Cloud teams have typically built AI coding models for third-party users and customers, there is a new focus on coding tools for in-house deployment. Internal Google AI coding models are said to be trained on the proprietary Google codebase. Using the private Google codebase may improve internal performance, although models trained on Google code reportedly cannot be released publicly.

The move comes as major technology companies, including the aforementioned Spotify but also AI competitors like Anthropic, rely on coding agents for growth. Anthropic’s Boris Cherny, who leads cloud code, said earlier this year that almost all of the company’s code is built using AI. Meanwhile, as the publication notes, Google uses AI for about half of its development, according to Google Chief Financial Officer Anat Ashkenazi, speaking during a February earnings call.

The change includes internal AI coding tool usage tracking, mandatory training sessions, and usage requirements for certain tasks. Google’s internal AI coding tool is called Jetski, and it wants to find out which employees are using it most aggressively.

A Google spokesperson said Information It said in a statement that it is seeing “tremendous adoption” of these devices. “Their use is turbocharging our model and AI tooling development – ​​that’s what we’re really focused on here.”