Google I/O opens today, walking into the venue with an awkward fact: in coding AI, Google has clearly fallen behind Anthropic's Claude Code and OpenAI's Codex.
According to The Information, DeepMind inside Google has set up a dedicated coding-AI attack team, with Nobel Chemistry laureate John Jumper also being pulled in. Even rarer, Business Insider reports that Google has allowed some engineers to directly use Claude Code for their work — an unprecedented compromise for a tech giant known for full-stack in-house development.
Where did it go wrong? MIT Technology Review points out a huge gap between Google's public-facing coding tools and the internal version. Even the internal version is still not enough to compete head-to-head with Claude Code. This explains why a company with top-tier research talent needs to rely on a competitor's product to do daily work.
But that doesn't mean Google is in full retreat. In science and medical AI, Google remains the undisputed leader — it is the only company to have produced a Nobel laureate through AI research. Last year's AI co-scientist was described by Stanford scientists as an "oracle-grade" tool, and the AlphaEvolve system also showed Google's technical depth in AI for Science.
The real suspense: can Google put forward a sufficiently strong counterattack in coding AI at I/O? The answer may determine whether this search giant can prove itself in the industry shift where AI value moves from Q&A to task execution.