The RavenDB team shares their real-world experience rolling out AI coding agents across development teams. Two self-contained features were completed dramatically faster than estimated — one month of work done in two days, three months of work done in one week. The author emphasizes this is not 'vibe coding': it requires deep engineering discipline, with the bottleneck shifting from writing code to reviewing it. Reading 30,000 lines of AI-generated code, enforcing standards, and evaluating architectural approaches became the primary human responsibility. The key insight is that AI agents amplify existing engineering culture, and the review cycle becomes the critical quality gate going forward.
Table of contents
The numbers that got my attentionThis isn't vibe codingWhat this means going forward5 Impressions