Skip to content

Internal · Living document

AI Playbook.
Use AI like it matters.

A shared, practical guide to understanding and using AI well, via AI-assisted tools or via API for in-app integration.

Start with the three limitations → Explore the guide

Ai (藍) is Japanese for indigo — a dye that deepens with every pass. So does this playbook.

New here? Read these two pages first — everything else builds on them.

  1. AI limitations — the mental model that explains why every technique in this guide exists.
  2. The two ways to use AI — AI in your products vs. AI as a dev tool. The rest of the guide is organized around it.

The map

Explore the guide

  • Introduction

    The mental model the whole guide runs on: AI and Human limitations, and the two ways to use it.

    Start here

  • Foundations

    Prompting, context management, planning, memory, verification, choosing models, and staying safe.

    Build the basics

  • Workflows

    How the work actually flows: parallel sessions, feature → plan → handoff, reusable skills, knowing when to reset.

    See the workflows

  • AI-assisted Tooling

    Getting started with the CLI, skills & plugins, superpowers, and wiring up MCP and config.

    Open the guide

  • Integrating AI via API

    Patterns for real LLM products: retrieval, tool loops, persistent memory, structured output, evals.

    Ship on the API

  • Team

    Onboarding new members, working with AI as a team, and adopting AI at the company level.

    Bring the team

  • Reference

    The quick-answer shelf: checklists, a glossary, and curated further reading.

    Look it up

Find your lane

How to read this guide

This guide is layered — you don't need to read it cover to cover.

If you are… Start with…
New to AI / non-technical The Introduction, then skim Foundations.
An engineer adopting AI dev tools Foundations → Workflows → AI-assisted Tooling.
Building a product on model APIs Integrating AI via API.
A team lead rolling out AI The Team section: onboarding, working with AI as a team, adoption at the company level.
Looking for a quick answer The Reference: checklists, glossary, further reading.

Deeper-technical sections are signposted in the text, and several pages (especially the API deep-dives) open with a plain-English "For non-engineers" callout that gives you the gist without the implementation detail.

The whole thing in a breath

The one-paragraph version

AI is powerful but bounded: it has limited memory, no real thinking, and outdated knowledge. Every good technique — chunking a codebase, writing plans to files, fetching current docs, managing context, choosing the right model — is a deliberate workaround for one of those limits. Used well, with a disciplined process around it and a skilled human guiding it, AI is a genuine force multiplier. Used carelessly, it just produces more mediocre work faster. This guide is about the difference.

This is a living document. If you find a technique that works, add it. The best version of this guide is the one the whole team keeps improving.