Defined Concepts

Concepts are the theoretical foundations — immutable ideas, taxonomies, and definitions that underpin the Agentic SDLC. They describe the "physics" of agent-driven software development.

Start Here

Three landmark concepts that define the domain.

Live Concepts

• Architecture Decision Record A lightweight document that captures a significant architectural decision, its context, and consequences at a specific point in time.
• Behavior-Driven Development A collaborative specification methodology that defines system behavior in natural language scenarios, bridging business intent and machine-verifiable acceptance criteria.
• Context Anchoring The phenomenon where explicit context biases an LLM toward specific concepts or solutions, even when marked as deprecated or irrelevant to the immediate task.
• Context Engineering Context Engineering is the practice of structuring information to optimize LLM comprehension and output quality.
• Extreme Programming A software development methodology emphasizing high-frequency feedback, testing, and continuous refactoring, which maps perfectly to the Agentic SDLC.
• Gherkin A structured, domain-specific language using Given-When-Then syntax to define behavioral specifications that are both human-readable and machine-actionable.
• Mermaid A text-based diagramming language that renders flowcharts, sequences, and architectures from markdown, enabling version-controlled visual specifications.
• Model Context Protocol (MCP) Open standard for connecting AI agents to external tools and data. Covers MCP's role in agentic development, its limits, and why implementation quality matters.
• Model-Driven Development An early 2000s software engineering paradigm that attempted 100% code generation from models, serving as a cautionary tale for modern spec-as-source AI hype.
• OODA Loop The OODA Loop (Observe-Orient-Decide-Act) is a decision-making framework by John Boyd. Learn how it applies to software development and autonomous AI agents.
• Provenance The chain of custody and intent behind software artifacts, distinguishing high-value engineered systems from 'slop'.
• Request for Comments A collaborative proposal document for significant changes that require team consensus before becoming formal decisions.
• The 4D Framework (Anthropic) A cognitive model codifying four essential competencies—Delegation, Description, Discernment, and Diligence—for effective generative AI use.
• The Learning Loop The iterative cycle between exploratory implementation and spec refinement, balancing vibe coding velocity with captured learnings.
• Triple Debt Model A diagnostic framework for software system health built around three interacting debt types that accumulate as AI accelerates development.
• Vibe Coding Natural language code generation without formal specs—powerful for prototyping, problematic for production systems.
• YAML A human-readable data serialization language that serves as the structured specification format for configuration, schemas, and file structures in agentic workflows.

Experimental & Draft

• Agent Skills Open standard for packaging procedural knowledge and workflows, serving as the 'how-to' layer for AI agents.
• AI Amplification The principle that AI tools amplify existing engineering practices—making disciplined teams faster and chaotic teams fail sooner.
• AI Software Factory An industrial-scale approach to software engineering. Explores the dichotomy between Safe ASDLC Factories (L3) and high-risk Dark Factories (L4).
• Coverage Metric Coverage measures task completion reliability: the proportion of required work units an agent successfully completes in a long-horizon task.
• Digital Twins Virtual replicas of complex systems. In software engineering, these are behavioral clones of third-party services used for high-volume scenario testing.
• Event Modeling A system blueprinting method that centers on events as the primary source of truth, serving as a rigorous bridge between visual design and technical implementation.
• Feedback Loop Compression How AI compresses the observe → validate → learn cycle, shifting the bottleneck from code production to code understanding.
• PR Slop PR slop is the flood of AI-generated code that passes automated checks but overwhelms human review capacity. It looks correct, compiles clean, and hides architectural drift.
• Product Requirement Prompt (PRP) A structured methodology combining PRD, codebase context, and agent runbook—the minimum spec for production-ready AI code.
• Product Thinking The practice of engineers thinking about user outcomes, business context, and the 'why' before the 'how'—the core human skill in the AI era.
• Production Readiness Gap The distance between a working generative AI demo and a secure, scalable production system.
• Software Civil Engineering The engineering discipline required for agentic software production, defined by formal specification layers, declarative lifecycles, and adversarial verification.

Deprecated

• Guardrails Why we deprecated the term 'Guardrails' in favor of strict separation between deterministic Context Gates and probabilistic Agent Constitutions.

Next Steps

• Patterns — See how concepts become architectural designs.
• Getting Started — Deploy the methodology from scratch.
• Resources — Field manual, design system, and governance.