Overview
- Jaana Dogan said Anthropic’s Claude Code produced a working design for a distributed agent orchestrator in about an hour that was comparable to what her Google team built over the past year.
- She noted the prompt was only three paragraphs and based on a simplified version of the problem because she could not use internal details.
- Dogan called the output a non‑production prototype that needs refinement and urged developers to test coding agents in domains where they can judge quality themselves.
- She said Google permits Claude Code only for open‑source projects, and she added that Google is working to advance its own models and tooling.
- Claude Code’s creator, Boris Cherny, recommends verification loops and plan‑first iteration, citing use of Opus 4.5 and integrations such as Slack, BigQuery, and Sentry, as Dogan also pointed to rapid progress from line‑level help in 2022 to full codebase generation by 2025.