Teaching AI the "Manners" of Your Code ── Designing Team Development That Balances Quality and Speed

In our team, after introducing Claude Code, the number of PRs increased, but we faced a new challenge: code that didn't align with the team's coding style led to repeated feedback and revisions, causing review congestion. Letting AI handle the work can boost speed, but without conveying the team's unique style, quality and speed remain a trade-off.

In this session, I'll share our practical case study of systematizing our lead designer's tacit knowledge into a style guide, teaching AI our team's coding rules, and achieving both quality and speed. I'll cover our approach of analyzing 300 past PRs — not just the code, but the PRs themselves; our three-layer framework that leverages AI for implementation, self-review, and review assistance; and a continuous improvement cycle where the style guide itself is refined through PR reviews on GitHub.
We propose this as a versatile framework that any team can adopt, not limited to Claude Code.