One year of Google’s work done in an hour by Claude Code: Google engineer

Jaana Dogan, a Principal Engineer for Google’s Gemini API, recently disclosed on X that Anthropic’s new AI tool, Claude Code, successfully reproduced a complex system architecture in one hour that her team had been developing for a year. The task involved creating distributed agent orchestrators—systems designed to coordinate multiple AI agents—a problem Google teams had wrestled with throughout 2024 without reaching an architectural consensus.

To test the tool without compromising proprietary data, Dogan provided Claude Code with a three-paragraph description of a simplified version of the problem using public concepts. Despite the minimal prompt and lack of secret internal data, the AI generated a working prototype that closely matched the design patterns Google had spent months validating. Dogan noted that while the output was a "toy version" and required further polish, its ability to navigate complex design choices independently was "shocking."

The admission highlights the rapid evolution of AI-assisted programming. Dogan outlined a timeline showing that in 2022, AI could only complete single lines of code; by 2024, it handled multiple files; and as of 2026, it can restructure entire codebases. She admitted that in 2023, she believed these current capabilities were still five years away. Currently, Google policy only permits the use of Claude Code for open-source projects, keeping it isolated from internal core systems.

Regarding the competitive landscape, Dogan described the industry as "not a zero-sum game" and praised the work of Anthropic. She stated that seeing such progress from a competitor motivates her team to push forward with Google's own models and harnesses. The viral post, which surpassed 5 million views, has sparked discussions among developers about how AI is beginning to sidestep corporate bureaucracy and legacy infrastructure constraints to accelerate technical execution.