AI already composes half of Google's code, and their numbers are only increasing

Google is developing its own AI tools for coding to compete with Anthropic Claude Code

AI-based programming services are gaining momentum, and Google sees Anthropic Claude Code as a serious threat. According to *The Information*, the company plans to strengthen its internal AI‑coding capabilities.

What Google intends to do
- Create an operational group of researchers and engineers whose task is to develop more advanced models for code generation.
- The group will be led by Sebastian Borgeaud, a research engineer from DeepMind who previously headed the pre‑training project.
- The team will also include Google DeepMind CTO Koray Kavukcuoglu and Google co‑founder Sergey Brin.
- The goal is to “upgrade” internal models that currently lag behind Anthropic Claude Code in quality.

How the new system will work
1. Training on proprietary data
Models will be trained on Google’s codebase, which will increase engineers’ productivity. However, open access to them is not planned – the company wants to preserve its competitive advantage.
2. Increasing AI support in projects
Currently AI writes up to half of the code in the company’s projects. The projected growth will allow even faster development of models and tools.
3. Monitoring Jetski usage
Google already actively uses its own Jetski tool for code generation. The team will track which employees use it most frequently, conduct training sessions, and formulate rules for its application to specific tasks.

> “We see these tools becoming widespread right now,” said a Google spokesperson. “Their use significantly speeds up the development of our models and AI tools; we are truly focused on this direction.”

Thus, Google is preparing its own AI agent to keep pace with competitors and boost internal software development productivity.