Amid the intense competition in AI programming tools, a "cracked" evaluation from within Google has caused a stir in the industry. Jaana Dogan, chief engineer of Google Gemini API, recently publicly praised Anthropic's Claude Code on the social platform X, stating that it generated a complex system framework that had troubled the Google team for a year - a distributed agent orchestration system - in just one hour, using only three short paragraphs of prompts.

 1 Hour vs 1 Year: A Qualitative Leap in AI Programming Capabilities

Dogan revealed that the Google team had tried multiple times to build this system but failed due to architectural disagreements. After she submitted the problem description to Claude Code, the AI quickly produced a system prototype with clear structure, complete logic, and directly runnable code. Although the code still needed optimization, its completeness was "comparable to the team's one-year iteration results."

"In 2022, AI could only complete a single line of code; in 2025, it can build an entire codebase from scratch," Dogan said. This rate of evolution far exceeded industry expectations, even making experts who had previously claimed "automated programming would take another five years" change their views.

 Envy Under Security Restrictions: Google Only Allows Open Source Projects to Use Claude Code Internally

Although giving high praise to Claude Code, Dogan also admitted that due to security and compliance requirements, Google currently only allows employees to use this tool in open source projects, while internal core systems still rely on self-developed models like Gemini. However, she emphasized that this external competition "is not a threat, but an incentive," driving the Gemini team to accelerate the optimization of code generation, tool calling, and engineering understanding capabilities.

 AIbase Observation: The Revolution of Programming Paradigm Is Here, the Key Lies in "System-Level Construction Ability"

Dogan's sharing reveals a key turning point: the competition in AI programming has evolved from "single-file completion" to "complex system design." Only models that understand advanced engineering concepts such as distributed architecture, cross-service communication, and state consistency truly have the potential to replace human junior to mid-level engineers.

The performance of Claude Code in such tasks may confirm Anthropic's advantages in code logic rigor, long context reasoning, and adherence to engineering standards. For Google, this is both a warning and a catalyst - in the "war of gods" of AI programming, only by continuously delivering reliable, maintainable, and scalable code can one win the minds of developers.