Currently reading

How do you review code at the volume agents produce it? On the Claude Code team, once we had high coverage and deep reviews, we started optimizing for the minimum amount of human review time that doesn't raise reverts or incidents.

How can customer-service agents uphold refund and cancellation policies under manipulation? This paper introduces CRAFT, a multi-agent red-teaming system, and τ-break, a benchmark built on τ-bench, to evaluate policy adherence under adversarial conversations and explore the limits of defensive strategies.