Terminology

This chapter defines key terms and concepts used throughout the Vibe-Driven Development methodology.

Vibe Coding

The practice of collaborative development with AI assistants, emphasizing natural communication and iterative refinement over rigid specifications.

Vibe-Driven Development (VDD)

An emerging methodology that brings structure and repeatable patterns to vibe coding, providing systematic foundations and granular control throughout the development process.

DevDocs Pattern

Living documentation that represents AI’s understanding of the project, continuously updated throughout development.

Smoke Tests Pattern

Comprehensive, verbose tests designed both for build verification and as specifications that AI can understand and implement against.

The Butterfly Defect

Small semantic imprecisions that cascade into large architectural changes. Named after the butterfly effect, where using “modal” instead of “dialog” might shift the entire UI philosophy.

Semantic Precision

Using exact, unambiguous terminology to prevent misinterpretation by AI.

Anchor Pattern

The practice of continuously ensuring existing functionality still works after AI makes changes, preventing regression through regular verification.

HILing (Human-In-the-Loop-ing)

The act of actively intervening as the human in the loop to review, correct, and steer AI output.

AI Drift

The gradual divergence of AI’s implementation from the original project vision, occurring when AI makes accumulated small decisions without human guidance.

Fuzzy Architecture

Starting with vague architectural guidelines and allowing structure to emerge through development, rather than over-specifying upfront.

Offload Pattern

Structuring code, documentation, and prompts to make AI’s job easier and more reliable.

Testing is Dialogue

The concept that tests serve as a communication medium with AI, teaching it requirements through examples.

Data Dump

The initial transfer of all available project information to AI, often messy and unstructured.

Retrofitting

Introducing VDD methodologies to projects that were originally developed without them.

Latent Persona

An emergent character or voice the AI adopts unintentionally, often due to biases in training data or weights. This can manifest as unexpected personality traits, communication styles, or decision-making patterns that weren’t explicitly prompted.

Ghost Tokening

Phantom influence from hidden system prompts, latent weights, or unseen context that alters style or tone without user intent. These invisible influences can cause AI to behave differently than expected.

Lo-Fi Coding (Low Fidelity)

Vague, exploratory collaboration where AI has maximum creative freedom.

Mid-Fi Coding (Medium Fidelity)

Balanced collaboration where human sets boundaries and AI fills in details.

Hi-Fi Coding (High Fidelity)

Enhanced human-in-the-loop intervention with focused vision alignment.

Progressive Fidelity

Starting with Lo-Fi for exploration, moving to Mid-Fi for implementation, and applying Hi-Fi for critical sections.