The Agentic Email Engine
The document is a comprehensive, hands‑on exposition of the Agentic Email Engine that powers Genesis’s email marketing. It explains why a simple voice‑clone bot is insufficient and introduces a three‑bucket system – ENGINE (shared universal craft), PERSON (the author’s unique voice, life, proof, and market feed) and JUDGE (a rulebook that rejects AI‑tells). The engine writes drafts using MarioBot, the judge AI grades them against hard‑fail rules, and the human edits are fed back as new rules, creating a two‑brain loop that continuously sharpens the copy. The text details the markdown files that store core voice fingerprints (core.md), weekly life/events (feed.md), verified proof points (proof.md), pop‑culture references (references.md) and skill definitions (skill.md). It walks through six real email examples (Announcement, Jason Statham, Losers Win Big, etc.), showing raw voice notes, engine drafts, edited versions, and the specific rule changes each edit taught (e.g., one‑sentence‑per‑line, ban “you’re not X, you’re Y”, avoid vague pronouns, name nouns, keep emails stand‑alone). From these edits a universal anti‑AI rulebook emerges (no balanced antithesis, no canned signposts, no dead adverbs, name the noun, no dashes as pauses, one sentence per line, no self‑contradiction, vary repeats, no overclaim, no stacked triplets). The system compounds: every edit becomes a rule, making future drafts more like the author while staying fresh. The feed layer pulls from three live sources – the author’s life voice notes, market conversation scraping, and business material – ensuring relevance, credibility, and humanity. The roadmap outlines three phases: current working email engine, upcoming automated feeds and multi‑format expansion, and a vision of self‑driving marketing across channels. Finally, a step‑by‑step DIY guide shows how to set up Claude Code, connect MarioBot, create the four personal files, build a custom judge, wire the loop, and feed edits back into rules, enabling anyone to replicate the system for their own voice.
Topics
Why Build the Agentic Email Engine – Beyond a Simple Voice Clone
Explains the four shortcomings of a standalone voice‑clone bot (no material, no quality gate, no learning, drift to average) and why a full system is needed to protect the unique voice and scale email output.
Three‑Bucket Architecture and Markdown File Blueprint
Details the separation into ENGINE (shared craft), PERSON (author’s moat), and JUDGE (rulebook), and enumerates the core markdown files (core.md, feed.md, proof.md, references.md, skill.md) that store voice fingerprint, life feed, verified proof, pop‑culture hooks, and skill definitions.
Two‑Brain Writing Loop – Writer AI, Judge AI, Human Feedback
Describes the cyclical process where MarioBot drafts using ENGINE+PERSON, the Judge AI bounces drafts that violate hard‑fails, the human edits the remaining copy, and those edits are codified as new rules for future drafts.
Style & Quality Rules – Hard‑Fails and Anti‑AI Tell Checklist
Lists the concrete rules that the Judge enforces (no balanced antithesis, no canned signposts, no dead adverbs, name nouns, no dashes as pauses, one sentence per line, no self‑contradiction, vary repeats, no overclaim, no stacked triplets) and explains their purpose.
Email Case Studies – From Voice Note to Final Ship
Walks through six real emails (Announcement, Jason Statham, Losers Win Big, Decoder, Open Book, etc.), showing the original voice note, engine draft, edited version, and the specific rule(s) learned from each edit.
Knowledge Feeds – Life, Market, and Business Streams
Explains how feed.md captures daily voice‑note life events, proof.md stores verified client wins and numbers, and references.md aggregates market conversation and pop‑culture hooks, providing fresh, uncopyable material for the engine.
Compounding Flywheel – Edits Turn into Rules
Shows how each human edit is reverse‑engineered into a rule, making the engine progressively more like the author while staying fresh, preventing stagnation and reducing future editing effort.
Universal Rulebook – Subtraction Layer for Any Voice
Presents the set of universal anti‑slop rules that ship with every forked engine, ensuring that regardless of the personal voice, the copy never sounds generic or AI‑like.