Claude Duet vs Claude Code Routines: permission gap
The author notes Duet asks for approval before sending anything while Claude Code Routines does not, and shares rules for running a morning brief on Routines.
nobody is f*cking talking about this part. Claude Duet is built to ask for your yes before it sends anything. A Claude Code routine has no yes button. Duet is Anthropic's new personal agent that reads your email and calendar and briefs you each morning. Research preview, Pro and Max plans, US waitlist only. You can get a morning brief on a Pro plan today with Claude Code Routines (cloud jobs that run on a schedule, laptop closed). Anthropic's docs say a run does not stop to ask permission and can use every tool of every connector you included, writes too. So mine runs on these rules: 1. Connectors: mail and calendar, nothing else 2. Prompt rule: read everything, send nothing, save replies as drafts 3. Brief: 5 items, the one that needs you first 4. Schedule: every 3 hours from 8 to 20, 150 runs a month 5. Day one: open the first run and read what it actually did Every run eats your plan limits like a normal session. Hourly is 720 runs a month, every 5 minutes is 8640, and Claude Code rejects that one anyway. Set it up tonight and the first brief runs at 8:07.
This post quotes another post: x.com/0xmfox/status/2113303189764100108