Claude Duet: proactive features and safety design
An overview of Claude Duet's proactive assistant features—morning briefs, app monitoring, task completion, reports—built on Claude memory and Opus 5.5 with safety limits.
Anthropic shipped another agent while we were asleep. Claude Duet is the proactive kind. It surfaces what actually needs you, keeps working while you're away, and only asks when it matters. It's a direct shot at Dots, Grok Bot, Hark and Muse. What it runs on: → your Claude memory and connected apps → Opus 5.5 for the judgement calls What it does: → morning brief → app monitoring → task completion → reports → building things Safety is baked in. No password access, a classifier on sensitive stuff, and the final say stays with you. It's an early research preview, but the product shape is clear. Anthropic is treating agents as a full product surface now. 𝗡𝗼𝘁 𝗷𝘂𝘀𝘁 𝗮 𝗺𝗼𝗱𝗲𝗹 𝗳𝗲𝗮𝘁𝘂𝗿𝗲.
This post quotes another post: x.com/claudeai/status/2112984664256241837