Skip to content

Concepts — Overview of Design Philosophy

A bird's-eye view of the "AI Agent Architecture Design Philosophy" across seven chapters.

Document Chain

Chapter Overview

Ch.LabelCentral QuestionLink
01WHYWhy do AI agents need guiding principles?01-vision
02WHATWhat should be used as reference sources?02-reference-sources
03HOWHow should the system be structured?03-architecture
04WHICHWhich pattern should be chosen and when?04-ai-design-patterns
05REALITYHow do we address real-world constraints?05-solving-ai-limitations
06EXTENSIONDoes the three-layer model hold in the physical world?06-physical-ai
07DOCTRINEOn what basis should AI judge and act?07-doctrine-and-intent

Layer × Concern Cross-Reference Matrix

Shows which chapters cover which concerns for each layer.

ConcernAgent LayerSkills LayerMCP LayerDoctrine Layer
Structural Definition03030307
Design Patterns040404
Constraints & Countermeasures05050505
Edge Extension06060606
Judgment Criteria070707
Reference Source Taxonomy0202
Design Philosophy (WHY)01010101

Mermaid Diagram Color Legend

The following color codes represent layers consistently across all chapters.

LayerColorMermaid fill
Agent LayerLight Blue#87CEEB
Skills LayerLight Green#90EE90
MCP LayerPink#FFB6C1
Doctrine LayerLight Orange#FFE4B5

Normative Strength Ladder (shall / should / may)

This site's documentation uses normative keywords conforming to RFC 2119.

KeywordStrengthMeaning
MUST / SHALLRequiredAn absolute requirement. Violation constitutes a design defect
MUST NOT / SHALL NOTProhibitedAn absolute prohibition
SHOULDRecommendedDeviation only with justified reason
SHOULD NOTNot RecommendedAdoption only with justified reason
MAYOptionalEntirely discretionary

Constraints within doctrine (07-doctrine-and-intent) and normative requirements extracted from spec MCPs are interpreted according to this strength ladder.

Concepts → Implementation Exit Checklist

A checklist to confirm that your understanding of the Concepts section is sufficient to proceed to the implementation phase.

Minimum Readiness Conditions

  • [ ] Reference Sources Minimum Catalog — Have you identified the authoritative sources your project will reference, and prioritized them for MCP integration? (See 02)
  • [ ] Three-Layer Separation Understanding — Can you explain the responsibility boundaries of Agent / Skills / MCP, and recognize anti-patterns (layer confusion)? (See 03)
  • [ ] Pattern Selection Rationale — Can you justify whether to adopt RAG, MCP, or Fine-tuning, and explain the reasoning? (See 04)
  • [ ] Constraint Boundary Awareness — Can you distinguish between constraints solvable by technology (knowledge constraints) and those not solvable by technology alone (institutional constraints)? (See 05)
  • [ ] Human Intervention Point Agreement — Has your team agreed on the agent's autonomy level and the conditions for escalation to humans? (See 07)
  • [ ] Evidence Trail Minimum Requirements — Does your design include mechanisms for post-hoc verification of AI decisions (verification status, source records)? (See 05)

Once These Are Met

→ Proceed to Development Phases and implement MCP integration at each phase → Refer to the Skills Design Guide and formalize domain knowledge as Skills

Correspondence with AI Research

The conceptual framework of this site corresponds to standard structures in AI agent research as follows.

Standard AI Research StructureCorresponding Concept in This SiteChapter
GoalIntent07
PolicyDoctrine07
ReasoningAgent Layer (inference & judgment)03
Tools / SkillsSkills Layer + MCP Layer03
ExecutionTool execution via MCP03, 04
Physical ActionPhysical AI06

Released under the MIT License.