A senior Google engineer revealed that Anthropic's Claude Code generated a functional system in just one hour—a project her team had been developing since last year.
Jaana Dogan, Google’s lead engineer for the Gemini API, wrote on X that she described the problem to Claude Code and received results consistent with what her team had built over the past year. The task involved a distributed agent orchestrator: a system that coordinates multiple AI agents. According to Dogan, Google had previously attempted various approaches to solve this challenge but failed to reach a consensus.
When asked about the prompt details, Dogan clarified it wasn't highly specific—only three paragraphs long. She tested Claude Code using a simplified version based on existing concepts, as she couldn’t share internal company information.
Dogan acknowledged the output wasn’t perfect and required further refinement. She encouraged skeptics of coding agents to try them out in domains where they possess deep expertise.
When questioned whether Google uses Claude Code internally, Dogan stated it is only permitted for open-source projects, not for internal work. In response to a user asking when Gemini would reach similar capabilities, she replied: “We’re working hard right now. Models and seatbelts.”
She added that the industry has never been a zero-sum game, so crediting competitors when due is both fair and motivating. “Claude Code’s work is impressive—I’m excited and even more driven to push all of us forward.”
AI coding tools are advancing faster than anyone anticipated
Dogan also outlined the rapid evolution of AI-assisted programming: in 2022, systems could complete single lines of code; by 2023, they handled entire code blocks. In 2024, they began working across multiple files to build simple applications. By 2025, they were capable of creating and refactoring entire codebases.
As recently as 2022, she believed the 2024 milestone would be difficult to scale practically for global developers. In 2023, today’s level of performance seemed like it would take another five years to achieve. “The quality and efficiency improvements in this field have exceeded everyone’s imagination,” she wrote.
Claude Code creator shares workflow optimization tips
Around the same time, Boris Cherny, the creator of Claude Code, shared his best practices for using the tool. His top recommendation: give Claude a way to verify its own work. This feedback loop can double or even triple the quality of the final output.
Cherny suggests most sessions begin in planning mode, iterating repeatedly with Claude until the plan is solidified. Once stable, Claude typically executes the task successfully in a single pass. For repetitive workflows, he leverages slash commands and sub-agents to automate specific tasks such as code simplification or application testing.
For longer tasks, Cherny runs background agents to audit work after completion. He also runs multiple Claude instances simultaneously to handle different tasks in parallel. His default model is Opus 4.5.
During code reviews, Cherny’s team directly tags Claude in colleagues’ pull requests to generate documentation. He noted that Claude Code integrates with external tools including Slack, BigQuery for data analysis, and Sentry for error logging.
