Defined Concepts

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.

Context Engineering is the practice of structuring information to optimize LLM comprehension and output quality.

The phenomenon where explicit context biases an LLM toward specific concepts or solutions, even when marked as deprecated or irrelevant to the immediate task.

A software development methodology emphasizing high-frequency feedback, testing, and continuous refactoring, which maps perfectly to the Agentic SDLC.

A lightweight document that captures a significant architectural decision, its context, and consequences at a specific point in time.

A collaborative proposal document for significant changes that require team consensus before becoming formal decisions.

The iterative cycle between exploratory implementation and spec refinement, balancing vibe coding velocity with captured learnings.

Methodology that defines specifications before implementation, treating specs as living authorities that code must fulfill.

A cognitive model codifying four essential competencies—Delegation, Description, Discernment, and Diligence—for effective generative AI use.

A collaborative specification methodology that defines system behavior in natural language scenarios, bridging business intent and machine-verifiable acceptance criteria.

A structured, domain-specific language using Given-When-Then syntax to define behavioral specifications that are both human-readable and machine-actionable.

A text-based diagramming language that renders flowcharts, sequences, and architectures from markdown, enabling version-controlled visual specifications.

The Observe-Orient-Decide-Act decision cycle—a strategic model from military combat adapted for autonomous agent behavior in software development.

A human-readable data serialization language that serves as the structured specification format for configuration, schemas, and file structures in agentic workflows.

SAE-inspired taxonomy for AI agent autonomy in software development, from L1 (assistive) to L5 (full), standardized at L3 conditional autonomy.

Framework for industrializing software development where agents serve as the logistic layer while humans design, govern, and optimize the flow.

The chain of custody and intent behind software artifacts, distinguishing high-value engineered systems from 'slop'.

How AI compresses the observe → validate → learn cycle, shifting the bottleneck from code production to code understanding.

The distance between a working generative AI demo and a secure, scalable production system.

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.

A structured methodology combining PRD, codebase context, and agent runbook—the minimum spec for production-ready AI code.

The practice of engineers thinking about user outcomes, business context, and the 'why' before the 'how'—the core human skill in the AI era.

Natural language code generation without formal specs—powerful for prototyping, problematic for production systems.

The universal connector for AI agents to access tools and data, acting as the supply chain infrastructure for agentic workflows.

Open standard for packaging procedural knowledge and workflows, serving as the 'how-to' layer for AI agents.

The principle that AI tools amplify existing engineering practices—making disciplined teams faster and chaotic teams fail sooner.

Coverage measures task completion reliability: the proportion of required work units an agent successfully completes in a long-horizon task.

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