The Standard For Building With AI
AI Build Playbook
A tier-scaled, vendor-neutral framework, plus three interview-driven master prompts, that takes you from idea to a reliable, correctly-scoped AI system in a single sitting.
Look Inside
Five pages from the book.
Every page designed like this. Doctrine, diagrams, and a teaching layer that assumes nothing.
“Push everything deterministically specifiable into code, reserve AI for genuine ambiguity, keep humans on irreversible decisions, and make every failure loud.”
The one-line thesis
The Problem
Over-engineered
Simple projects get dragged into three-week ordeals: multi-agent orchestration, vector databases, and frameworks for something that needed one good prompt.
Under-engineered
Client-facing systems ship with a weekend script's discipline, then break silently the moment the model gets something wrong, and nobody notices until it's expensive.
What You Get
-
PDF
The AI Build Playbook · Reader Edition65 pages · the doctrine, the diagrams, a worked example, and a full glossary
-
MD
Prompt 1 · Fresh InstallInterviews you, then designs and builds a new system from zero
-
MD
Prompt 2 · EnterpriseDesigns within company security, approval, and data constraints
-
MD
Prompt 3 · Audit & UpgradeInspects an existing AI system, then improves it with your approval at every gate
The Core Ideas
-
Tier by consequence, not code size
A four-level system (T0–T3) that scopes rigor to what actually happens when it's wrong, so you never over- or under-build again.
-
The code / agent / human allocation rule
What each should own, and the decisions AI must never make on its own.
-
The simplicity ladder
Start with the simplest thing that works. Only climb toward complexity when a lower rung provably fails.
-
The silent-failure doctrine
Make every failure loud, so nothing quietly breaks in the dark.
“For any component, ask: what would it look like if this were silently broken? If the answer is ‘exactly like it looks right now,’ you have found the next thing to build.”
The meta-test · Section 10.3
Questions
All of them. The playbook is deliberately vendor-neutral: no lock-in to any model, language, database, or platform. The prompts run in any capable AI chat or coding agent, and the doctrine applies whether you build with Claude, GPT, Gemini, or something local.
No. The Reader Edition explains every technical term in plain language the first time it appears, includes a full glossary, and the prompts interview you in ordinary language. Saying "I don't know" is an expected, supported answer. Experienced builders get the same doctrine at full depth.
One 65-page PDF (the reasoning layer: framework, diagrams, worked example, glossary) and three Markdown prompt files (the execution layer: paste one into an AI session and it runs the process for you). They're designed to be used together.
When the playbook is revised, you get the new edition through your Gumroad library at no extra charge. The field moves fast; the playbook moves with it.
Who It's For
Consultants, founders, product managers, engineers, and serious hobbyists. Anyone who wants their AI builds to be reliable on purpose, not by luck.