Build Log 01 — A daily AI briefing that writes itself overnight

Overview

An automated daily briefing system assembles overnight and arrives by midnight. The core insight: quality control comes from a written specification that the system reads on every run, not from clever prompting.

What it actually does

The pipeline runs on weekdays, pulling together:

• AI brief — named companies with measurable outcomes, excluding vendor PR
• Product and growth brief — actionable moves for client work
• Today's schedule — calendar with transit routing
• MMA check-in — training recommendations based on accumulated load
• Coffee recommendation — rotating single-roaster suggestions
• Content idea — one publishable writing prompt

The system integrates Google Calendar, Gmail, Google Drive, web search, and Zapier to deliver three minutes of reading with two or three actionable items.

Why specification beats prompting

Early versions were too broad and vague. The fix was a quality gate: if an item can't name a real company and a measurable outcome, cut it. This prevents outputs like generic trend statements from cluttering the briefing.

The approach mirrors product development — writing requirements before building, then iterating on the specification when results disappoint, rather than re-prompting and hoping for better output.

The refinement loop

The system evolved to include:

• Plain-English explanations of technical concepts
• Explicit rest-day overrides when training load indicates recovery needed
• Chained skills that call other documented specifications

Pattern for replication

1. Write specifications defining quality before the first run
2. Ship anyway — specs improve when they fail
3. Revise specifications, not prompts, when disappointed
4. Chain skills instead of duplicating rules

The guiding principle: no company, no metric, no item.