A Claude workflow that runs my morning.

I’ll describe the workflow exactly, because the platform’s position is that AI workflows should be published, named, and reproducible — not vaguely gestured at.

The morning workflow is one Claude Project. The Project has three reference files: my standing operating context (role, priorities, recurring obligations), the format I want briefings in, and a running file of patterns I want it to flag (when a deal is going stale, when a follow up is overdue, when a meeting could be an email).

I run the prompt at 7am. It pulls in (manually pasted) the day’s calendar, the unread starred Slacks, and yesterday’s open Linear tickets. It produces: a one paragraph briefing on the day’s shape, a list of three priorities with reasoning, a list of items I’ve probably been avoiding, and a question or two it thinks I should be thinking about.

Where it fails: it sometimes weights novelty over importance. It occasionally invents a detail that wasn’t in the input (rare in long context, more common when I’m sloppy with the paste). It can’t see anything I don’t give it, which means the workflow is only as good as the inputs.

Refusals worth naming: no AI generated audio of me reading these briefings back. No AI avatar standing in for me in meetings. No outbound emails written without my review and edit. These are not abstract principles; they’re lines in the Project’s reference file.

— Alvin

— Alvin