Builder Methods library
Training AI with taste
The second pattern lesson makes the course-wide voice thread explicit: two mechanisms, a self-check, where the training lives, and the kit for doing it on your own work.
Voice training takes two mechanisms, and you need both: written training material that holds the rules, and live calibration from your most recent finished work. This lesson names what you saw in the content example and turns it into a repeatable move for any business that publishes with its name on it. We cover how a voice doc gets built from proof instead of description, why the slop list has to be yours, what makes a self-check work, and why the training folder lives in one central place. It closes with the voice-training kit in your course assets and how to run it on your own writing.
What's covered:
- The two mechanisms: documented rules and live calibration from recent work
- Building voice.md from evidence: every rule with a quoted line
- slop.md as a running list, seeded and maintained by you
- The self-check: a specific checklist the agent runs before you see the draft
- Training at the repo root so every skill reads the same files
- The voice-training kit: builder prompt, voice template, slop starter
- Where this applies: newsletters, articles, support replies, sales emails