Anthropic explains how it grades task priorities in Claude Duet

Anthropic told The New Stack how it tests Claude Duet, a background AI assistant on Claude Opus 5.5 that monitors connected apps and flags tasks needing attention. The company built internal graders for task prioritization, admitting urgency is subjective and the eval is imperfect.

• Duet debuted October 21 on Claude Opus 5.5, working in the background across connected apps
• A separate automated check built on the Claude Code classifier verifies actions match user intent
• Background checks run every few hours and count against Pro and Max plan limits
• Research preview is waitlist-only for U.S. Pro and Max subscribers