Jaana Dogan, Google's tech chief surprised with a rival AI recreating her team's year-long work within an hour

Jaana Dogan, a Principal Engineer at Google working on the Gemini API, has drawn attention in the tech community after sharing an experience involving Anthropic’s Claude Code, an AI tool developed by a competitor.

She gave Anthropic’s Claude Code a description of the problem. Not internal documents. Not proprietary code. Just a few paragraphs describing the challenge at a high level. What came back, she said, looked strikingly familiar. “I’m not joking and this isn’t funny,” Dogan wrote in a post on X. “I gave Claude Code a description of the problem, it generated what we built last year in an hour.”

What made Jaana Dogan unsettling?

Dogan was careful about how she framed the experience. She did not say the AI produced a finished system. She did not claim it replaced her team. The AI output mirrored architectural ideas that had taken months of discussion, disagreement and refinement among experienced engineers. In later clarifications, Dogan explained that the AI produced a prototype — something closer to a draft than a final answer. Still, the speed mattered. It compressed time in a way that felt impossible just a year ago.


Dogan explicitly noted that she was experimenting with a competitor’s product. As a senior engineer involved with Google’s own AI offerings, she acknowledged the significance of testing an external system and openly shared the outcome. In follow-up remarks, Dogan clarified that what Claude Code generated resembled a prototype or architectural direction rather than a complete solution. However, she said the similarity to her team’s earlier work was close enough to be striking.

Check Jaana Dogan's X (formerly twitter) post here:

Reaction within the engineering community

Dogan’s post resonated with engineers largely because it came from someone deeply familiar with the problem space. She encouraged others to test AI tools against problems they already understand well, suggesting that this is the best way to assess current capabilities. Some commentators highlighted that the outcome doesn’t diminish the value of long-term development work; instead, it reflects how AI tools are becoming capable of quickly generating drafts or prototypes when given clear, domain-specific inputs.

She did not suggest that human engineering effort was unnecessary or obsolete. Instead, her comments focused on the speed with which an AI system could arrive at a familiar design after receiving only a problem description.