Google spent 1 year on a task, Claude Code did it in 1 hour, says Google engineer

Google

• Google engineer: Anthropic's Claude Code equaled a year's work in one hour
• Claude Code built a multi-agent AI system from a brief public prompt
• AI coding tools now rebuild entire codebases, far outpacing expectations

Did our AI summary help?

Imagine spending one year building something at work, with meetings, whiteboards, debates, snacks, and more meetings. Now imagine a rival shows up and does almost the same thing in one hour. That’s basically the story Google engineer Jaana Dogan accidentally dropped on the internet, and of course, it exploded.

Dogan, a principal engineer working on Google’s Gemini API, recently tested Anthropic’s Claude Code, an AI tool that helps write and organise code. She fed it a problem her team had been wrestling with for a year. The result? Claude delivered something shockingly close to Google’s work in 60 minutes. Dogan posted on X saying, “I’m not joking and this isn’t funny. We have been trying to build this since last year. I gave Claude a description, and it generated what we built last year in an hour.”

To make things clearer for the non-Google crowd: the task was about building systems that manage multiple AI agents working together. Think of it like traffic control, but for AI bots instead of cars. Google had tried many ideas, but the team still hadn’t agreed on a final design. Dogan wanted to test Claude without using Google’s private data, so she created a simplified version of the problem using public ideas. Her prompt was just three short paragraphs. No secret sauce. Just curiosity and three paragraphs.

She admitted Claude’s answer isn’t perfect and still needs polish, but she encouraged people who doubt AI coding tools to test them on problems they already understand deeply. Because that’s where the shock value really hits.

Someone asked if Google uses Claude Code internally. Dogan replied that it’s allowed only for open-source projects, not for Google’s internal work. Another person asked when Gemini will catch up. Dogan said, “We are working hard right now. The models and the harness.” Translation? They’re trying. Calmly panicking, but trying.

But Dogan didn’t make it a villain arc. She praised the competition too, saying AI progress isn’t a race where one side wins and the other loses. “Claude Code is impressive work. I’m excited and more motivated to push us all forward,” she wrote.

She also gave a timeline of how wild AI coding has become. In 2022, AI could write one line at a time. In 2023, it could finish full sections. In 2024, it could handle multiple files. In 2025? It can now rebuild entire codebases. Dogan admitted she never saw this pace coming. In 2023, she thought these tools were five years away. In 2022, she doubted they could ever scale into a global developer product. Surprise, surprise.

Her post crossed 4 million views and 4 million developer emotions. Users chimed in saying AI moves faster because it doesn’t have corporate red tape. One comment said, “It’s fascinating how AI sidesteps bureaucracy, sparking a renaissance in creativity.” Others called Claude Code genuinely impressive and said developers will soon move at a completely different speed.

The moral of the story? AI doesn’t replace snacks and whiteboards yet. But it definitely replaces one year of meetings.