The Instinct Transfer
The document authored by Mario Castelli presents a detailed, step‑by‑step protocol for transferring a copywriter’s 15‑year‑old instincts into an AI system. It argues that instincts are embedded in the edits a writer makes, and that by capturing those corrections as data the AI can learn to apply the writer’s taste automatically. The system consists of three text files—Taste.md (exemplary work), Rules.md (hard rules and tendencies), and Pairs.md (before/after edit pairs)—plus a “Judge” instruction that forces the AI to self‑grade drafts against the Rules before the writer sees them. The guide walks the user through seven implementation steps (workspace setup, seeding the taste file, mining before/after pairs, tiering tendencies vs. hard rules, installing the judge, real‑time capture triggers, and drift guardrails), and then shares six higher‑order lessons learned from applying the protocol to daily emails, short‑form video scripts, and ad hooks. It outlines criteria for suitable deliverables, a 30‑day rollout plan, and provides a full appendix of ready‑to‑paste prompts. The final section promotes the author’s Genesis mastermind program, noting that the document itself was generated by the described AI system.
Topics
Instinct Transfer Premise: Teaching AI Your Taste via Edits
Explains the core idea that a copywriter’s instincts are revealed in the corrections they make, and that capturing these edits lets an AI emulate the writer’s taste.
System Architecture: Three Files and a Self‑Checking Judge
Details the three core text files—Taste.md, Rules.md, Pairs.md—and the Judge instruction that forces the AI to grade drafts against Rules before presenting them to the human.
Step 1 – Workspace Setup in Claude Code
Guides the user to create a fresh Claude Code folder, run the setup prompt, and generate the initial file structure.
Step 2 – Seeding the Taste File with Exemplars
Instructs the user to collect 5‑10 best pieces of the target deliverable, interview themselves via the AI, and annotate why each is great and what a mediocre version would look like.
Step 3 – Mining Before/After Pairs to Extract Principles
Describes how to feed the AI before/after edit pairs, have it name the failure, name the underlying principle, classify context, and propose a one‑line tendency for approval.
Step 4 – Tendencies vs. Hard Rules Tiering
Explains the two‑tier rule system: a single edit creates a tendency (with context), while a hard rule requires recurrence (3+ instances) or explicit author approval, and introduces blind‑testing for rule validation.
Step 5 – Installing the Judge for Pre‑Draft Filtering
Provides the prompt that makes the AI automatically grade drafts against Rules, fix hard‑rule violations, and present only cleaned drafts to the writer.
Step 6 – Real‑Time Capture Triggers (Save That / That Worked)
Sets up two short trigger phrases that let the writer instantly log a correction as a tendency or a win without breaking workflow.