Google DeepMind Forms Coding Task Force to Compete with Anthropic, Brin Personally Oversees Efforts

According to monitoring by Dongcha Beating, The Information cites three informed sources stating that Google DeepMind has formed a task force composed of researchers and engineers specifically aimed at enhancing its coding model capabilities. The task force is led by DeepMind research engineer Sebastian Borgeaud, who previously oversaw DeepMind’s pre-training efforts; co-founder Sergey Brin and DeepMind CTO Koray Kavukcuoglu are also directly involved. The immediate trigger for forming the team was Anthropic’s recent model release. DeepMind internal researchers believe that Anthropic’s coding tools have surpassed the code generation capabilities of Gemini. In a recent memo, Brin wrote that the team must ‘urgently close the gap in agent execution capabilities and turn the model into a primary developer for writing code’ to win this final sprint. ‘Agent’ refers to AI capable of handling multi-step tasks. The gap has specific figures: Boris Cherny, head of Anthropic’s Claude Code, stated in January that the company has ‘almost 100%’ of its code written by AI; Google CFO Anat Ashkenazi mentioned in February’s earnings call that coding agents at Google handle only about 50% of coding work internally. The task force focuses on long-cycle coding tasks, such as writing new software from scratch, which require the model to read multiple files and understand user intent, representing the most challenging aspect of current AI coding tools. The training corpus is also being adjusted: Google has begun using its private codebase to train models, as internal code differs significantly from public code, and general coding models do not perform well on internal projects. These internally trained models cannot be released externally but can help iterate publicly available versions. On the internal promotion front, Google has set up a leaderboard for the use of an internal coding tool called Jetski; some teams outside of DeepMind have begun organizing mandatory AI training. In the memo, Brin requires that every Gemini engineer must use the internal agent when performing complex multi-step tasks. The longer-term goal is what Brin refers to as ‘AI takeoff,’ meaning self-improving AI. He has repeatedly told employees that enhancing coding capabilities is key to reaching this stage; in conjunction with AI that can perform mathematics and run experiments, it theoretically allows for large-scale automation of AI researchers and engineers’ work. OpenAI already has similar internal tools to help researchers generate experimental code more quickly.