Morning Brief Agent
A two-stage Gmail and Calendar agent that tells me who I owe a reply, what's coming up, and what went stale.
refining
Summary
An agent that reads my job-search email and calendar and writes a morning brief: replies owed today with a ready draft, upcoming calls and interviews with context pulled from the threads, stale conversations, and who I'm waiting on. Stage one runs the agent loop against a saved snapshot of my inbox; stage two runs the same loop against live Gmail and Google Calendar. It caught an interview that was confirmed by email but missing from my calendar.
Problem
Job searching means dozens of threads with recruiters, referrals, and interviewers. Things fall through the cracks, and the cost of a dropped thread is high.
AI Usage
One model with tools (search Gmail, read a thread, read the calendar) decides what to read and writes the report in the same conversation. The Google API tools are deterministic code. A plain-text corrections file I edit overrides anything the model would infer from email. Every model call goes through my Model Router, which logs what each run costs.
- Cross-references email threads against calendar events and flags gaps.
- Drafts the reply for anything owed today, ready to send after I review it.
- Developed against a saved inbox snapshot, so real mail is never touched by accident.
- A standing-corrections file lets me fix the agent's understanding in one line instead of editing prompts.
- Every run is logged by the Model Router; cost per brief is being measured now.
Stack
Python, Anthropic SDK, Gmail API, Google Calendar API