The Agentic Email Engine
The document presents Mario Castelli’s “Agentic Email Engine,” a marketing‑automation framework that combines a writer‑AI, a judge‑AI, and a personal knowledge base to generate email copy that retains the author’s unique voice at scale. The system is split into three logical buckets: ENGINE (universal craft shared across users), PERSON (the author’s personal moat – life stories, proof, market observations), and JUDGE (a rulebook of hard‑fails and style constraints). Core files (core.md, feed.md, proof.md, references.md, skill.md) store the author’s worldview, real‑time life feed, verified proof points, pop‑culture references, and specialized skills. A continuous learning loop feeds every human edit back into the judge as a new rule, making the engine increasingly like the author. Six email campaigns are detailed, showing the raw voice note, the engine’s draft, Mario’s edited final version, and the specific rule(s) derived from each edit (e.g., one‑sentence‑per‑line, no “you’re not X, you’re Y” trope, avoid vague pronouns, name nouns, avoid over‑claims, keep emails stand‑alone). The document also outlines the philosophical premise that the moat is the author’s lived experience and voice, not the AI itself, and provides a roadmap (Phase 1 live engine, Phase 2 automated feeds, Phase 3 self‑driving marketing) plus a step‑by‑step guide for others to replicate the system using Claude Code and custom judge files. The overall claim is that this dual‑brain architecture yields fast, scalable, and uniquely personal email copy that outperforms generic AI output.
Topics
Engine‑Person‑Judge Architecture: The Three‑Bucket Design
Explains the separation of universal craft (ENGINE), personal moat (PERSON), and rule enforcement (JUDGE) and how they interact in the copy‑generation loop.
Voice Moat: Personal Material vs. Universal Craft
Details why the author’s life stories, proof points, and market observations form an uncopyable moat that differentiates the output from generic AI copy.
File Structure & Data Feeds (core.md, feed.md, proof.md, references.md, skill.md)
Lists each markdown file, its purpose, and example contents that feed the writer‑AI with style, real‑world events, verified proof, pop‑culture hooks, and specialized skills.
Judge Rulebook and Hard‑Fail Rules
Enumerates the universal anti‑AI rules (no "you’re not X, you’re Y", no dashes as pauses, one sentence per line, no vague pronouns, no over‑claims, stand‑alone emails, etc.) that the judge AI enforces.
Learning Loop: From Human Edits to Automated Rules
Describes how each edit Mario makes is reverse‑engineered into a new judge rule, making the engine progressively more like his voice.
Six Email Case Studies: Draft → Edit → Rule Extraction
Provides concrete examples of six promotional emails, showing the original voice note, the engine’s draft, Mario’s final edit, and the specific rule(s) derived from the edit.
Philosophical Premise: The Moat Is Voice, Not AI
Articulates the core belief that AI is a tool; the sustainable competitive advantage lies in the author’s unique voice and lived experience.
Roadmap & Vision: Phases 1‑3 of the Self‑Driving Marketing Machine
Outlines current Phase 1 (live engine), upcoming Phase 2 (automated market and business feeds), and Phase 3 (full multi‑channel self‑driving marketing).