A senior Google engineer says Anthropic’s Claude Code built a complex system in one hour that matched work her team spent a year developing. Jaana Dogan, Principal Engineer at Google responsible for the Gemini API, tested the AI coding tool on a distributed agent orchestration problem.
According to The Decoder, Dogan gave Claude Code three paragraphs describing the task. The tool generated a working system that compared to what Google had built after exploring multiple approaches over the past year. The task involved coordinating multiple AI agents across distributed systems.
Dogan later clarified her comments after they drew significant attention. She explained that Google built several versions of the system over the past year. When prompted with the best surviving ideas, coding agents can generate a decent prototype in about an hour. She stressed the output is a toy version, not production grade, but provides a useful starting point.
Quality Surprises Despite Simple Prompts
The engineer noted she was surprised by the quality of what Claude Code generated. She did not provide detailed design prompts, yet the tool offered good recommendations. Dogan emphasized that years of learning and grounding ideas in products precede such results.
She pointed out that having insight and knowledge makes building much easier today. Taking existing knowledge and rebuilding from scratch was not possible in the past. Building from scratch means final products are free from baggage, she wrote.
Rapid Evolution in AI Coding
Dogan outlined how AI-assisted programming has advanced quickly. In 2022, systems completed individual lines. In 2023, they handled entire sections. By 2024, they worked across multiple files and built simple apps. In 2025, they can create and restructure entire codebases.
She did not believe the 2024 milestone could scale as a global developer product back in 2022. In 2023, today’s level seemed five years away. Quality and efficiency gains in this domain exceeded anyone’s imagination, she stated.
Creator Shares Workflow Tips
Boris Cherny, who created Claude Code, shared his workflow recommendations. His top tip is to give Claude a way to verify its own work. This feedback loop doubles or triples output quality, he said.
Cherny suggests starting sessions in plan mode and iterating until the plan is solid. He uses slash commands and subagents for recurring tasks. For longer projects, he runs background agents to review Claude’s work. His team tags Claude in pull requests to add documentation automatically.
