The news: Google DeepMind has formed a team of researchers and engineers to improve its AI coding models, amid pressure from rival Anthropic’s recent AI model releases, sources told The Information.
The context: DeepMind researchers reportedly view Anthropic’s coding tools as out-performing the abilities of Google’s Gemini models, and both Google co-founder Sergey Brin and CTO Koray Kavukcuoglu have been directly involved with the strike team, a source told the masthead.
Google is putting more emphasis on models that write code the company can use internally, which requires training the models on Google’s code. While models that are trained on internal Google code cannot be released publicly, they can help to improve models which could then be released publicly.
The strike team is being led by Sebastian Borgeaud, a research engineer at Google DeepMind who previously led pretraining for Google DeepMind, three sources told the publication.
The strike team is focusing on improving model performance for long-term coding tasks which are typically complex and the models to read through files and to understand a user’s intent.
Anthropic has previously said that it uses coding tools for the majority of its engineering needs. Head of Claude Code, Boris Cherny, said in January that “pretty much 100%” of Anthropic’s code is written using AI, while Google CFO Anat Ashkenazi said in February that around 50% of its code is being written by coding agents.