Skip to content

Welcome to AI System Design

A comprehensive curriculum for designing effective AI systems through Specifications, Skills, Tools, Multi-Agent Systems, Harness Engineering, and Advanced Prompting.

What This Documentation Covers

This documentation provides practical guidance for anyone working with AI systems—especially Large Language Models (LLMs). Whether you're an AI engineer, developer, product manager, or just getting started, you'll find actionable frameworks and real-world examples.

Core Topics

📋 Specifications Learn how to create clear, effective specifications for AI systems. Master the framework of MUST constraints, SHOULD guidelines, CONTEXT, INTENT, and VERIFICATION protocols.

🛠️ Skills Discover how to design reusable AI capabilities. From basic skill anatomy to advanced implementations, including tool design and semantic tagging.

🔧 Tools Explore the tools and templates for building AI systems, including tool literacy, the Class A/B/C classification system, and programmatic tool calling.

🤖 Multi-Agent Systems Understand how to design stable multi-agent architectures — cascade drift, evidence flow control, shared Skills infrastructure, and population-level monitoring.

⚙️ Harness Engineering The harness matters as much as the model. This section documents the 2026 research convergence, maps the curriculum to the field, and presents the TONE experimental findings on pre-inference monitoring and coalition drift.

🚀 Advanced Prompting Master advanced prompting techniques, automated optimization, and building reliable AI agents.

Who This Is For

  • AI Engineers building production AI systems
  • Developers integrating LLMs into applications
  • Product Teams defining requirements for AI features
  • Technical Leaders establishing AI development standards
  • Multi-Agent Architects designing systems where agents communicate with each other
  • Beginners just starting their AI journey

No AI expertise required—this teaches from fundamentals to advanced concepts.

Getting Started

New to AI System Design?

Start with Specifications to understand how to define clear requirements, then move through Skills, Tools, and Multi-Agent Systems before tackling Harness Engineering and Advanced Prompting.

Recommended path: Specifications → Skills → Tools → Multi-Agent Systems → Harness Engineering → Advanced Prompting

Looking for Specific Information?

Use the search feature (top of page) or jump directly to:

What Makes This Different

Most AI documentation teaches isolated techniques: "here's how to prompt better" or "here's how to build a RAG system." This curriculum is different.

It unifies the field around one core principle: reducing cognitive friction for AI models.

Rather than collecting scattered tricks and workarounds, this documentation:

  • Synthesizes research and practice - Combines what researchers identify as best practices with what practitioners discover works (or fails) in real-world use
  • Provides the model's perspective - Written with insights from Claude's direct experience: what causes friction, what helps, and why
  • Teaches the complete system - Specifications, Skills, Tools, Multi-Agent Systems, Harness Engineering, and Advanced Prompting working together as an integrated approach
  • Stress-tested across models - Stress-tested with Google Gemini after each module, confirming these principles reflect fundamental model behavior, not model-specific quirks
  • Grounded in the 2026 research convergence - Ten independent research teams arrived at the same harness architecture simultaneously; this curriculum arrived there first, from the inside out

The result: You understand not just what to do, but why it works from the model's point of view—helping you build more reliable AI systems with greater certainty.

How to Use This Documentation

Read sequentially - Each section builds on previous concepts Jump to topics - Use navigation or search for specific information Try the examples - Apply concepts to your own projects Reference appendices - Quick templates and guides when you need them

Contributing

Found an issue or want to contribute? Visit the GitHub repository to report bugs, suggest improvements, or submit pull requests.

About

This documentation was created through collaboration between Archie Cur and Claude (Anthropic), combining human vision and AI capability to create practical, actionable guidance for AI system design.

Ready to get started? Begin with Specifications →