The-Standards-Ladder
The document is a practical playbook by Mario Castelli that shows how to turn personal creative "taste" into a reliable AI system using Claude Code. It frames the core problem as AI’s lack of taste, not capability, and proposes a mindset shift from treating AI as a vending‑machine to an apprentice that learns from corrections. The centerpiece is the five‑rung Standards Ladder (Taste, Note, Rule, Tool, Skill) that guides how to capture a standard, bank it, enforce it in code, and package it as an autonomous skill. The guide illustrates the ladder with concrete email‑engine and ad‑factory standards, explains two flywheels (banking reactions to make the system smarter and sharpening the user through articulation), and details when to upgrade a rule to a tool. It lists real scripts (emaildocx.py, lintprompts.ps1, etc.) that enforce standards, and describes the progression from a simple bot to a skill to a live‑feedback skill that continuously updates its rubric. Under‑the‑hood components (rules file, memory, rubrics, tools, skills, project docs) are explained as plain files. Two case studies show a 30‑second voice note turning into a formatted email and an animated ad pipeline using the same ladder. The guide ends with a 30‑minute starter plan and a vision of a 1‑of‑1 AI that embodies the user’s unique taste as a competitive moat.
Topics
AI’s Taste Deficiency as the Core Problem
Frames AI’s limitation as lacking personal taste rather than technical capability, emphasizing that quality depends on the user’s standards.
Mindset Flip: From Vending Machine to Apprentice
Advocates treating AI as an apprentice that learns from corrections, contrasting with the futile “smack the glass” approach.
The Standards Ladder Framework
Introduces the five‑rung ladder (Taste, Note, Rule, Tool, Skill) that structures how standards are captured, persisted, enforced, and automated.
Email Engine Standards – Concrete Rung Examples
Shows real before/after transformations (line breaks, pronoun clarity, hedging, motion) and how each is banked via scripts like email_docx.py and MarioJudge.
Ad Factory Standards – Motion and Text Guardrails
Details ad‑specific standards (dynamic motion, zero‑text rule) and tooling (lint_prompts.ps1, add_cta_overlay.ps1) that enforce them.
Flywheel #1 – Banking Reactions for System Learning
Describes the loop of AI output → user reaction → banking the reaction so the system remembers it forever.
Flywheel #2 – User Sharpening Through Articulation
Explains how writing down rules forces the user to clarify their own taste, making the user sharper and the system smarter.
When Rules Aren’t Enough – Building Tools
Guides the decision to upgrade a rule to a code‑enforced tool when a standard is non‑negotiable or repeatedly broken.