Glossary

Plain-language definitions of the terms behind AI-assisted app building and pre-build planning.

80% Wall

The 80% wall is the point in an AI-assisted build where the app looks done but breaks once real complexity like auth, payments, or multi-user state enters.

Acceptance Criteria

Acceptance criteria are the conditions a feature must meet to be done. For AI coding agents, they define done before the task starts.

Agentic Building

Agentic building is using an AI coding agent, like Cursor, Claude Code, or Codex to autonomously write, test, and iterate on code across multiple steps.

Architectural Decision Record

An ADR is a short document that captures a significant architectural decision: what was decided, why, and what tradeoffs were accepted.

Architectural Guardrails

Architectural guardrails are constraints embedded in your planning layer that prevent an AI coding agent from conflicting with your intended architecture.

Architecture Design Document (ADD)

An ADD is the pre-code blueprint for a software system: data schema, tech stack, routing logic, and the boundaries agents must not cross.

Architecture Drift

Architecture drift is when an AI agent invents patterns, APIs, and structure to fill gaps because it has no master plan to follow.

Coding Ticket

A coding ticket tells an AI coding agent what to build, which constraints to follow, what must exist first, and how to verify the result.

Context Rot

Context rot is when an AI coding agent's context fills with contradictory instructions, errors, and expired rules, so it hallucinates instead of executing.

Data Contract

A data contract defines how data moves between frontend and backend. Without one, AI agents rewrite API payloads to fix local bugs.

Database Schema

A database schema is how your data is organized. Let an AI guess it on the fly and your app will eventually collapse under bad migrations.

Jenga Code

Jenga Code is a fragile AI-built codebase: fix one bug and unrelated features collapse because nothing was architected upfront.

Minimum Viable Product (MVP)

A minimum viable product is the smallest version of an app that delivers its core value. Defining that scope keeps a vibe-coded build focused.

Phased Implementation Plan

A phased implementation plan breaks an app build into isolated, sequential prompts so your AI agent builds on a stable foundation, not a scrambled codebase.

Plan Mode

Plan Mode makes an AI outline steps before coding. It cuts local mistakes but does not replace global pre-build architecture.

Pre-Build Planning

Pre-build planning is architecture work before you open an AI editor. Skip it and you hit the 80% wall when complexity shows up.

Product Requirements Document (PRD)

A Product Requirements Document defines what a product should do. But if you feed it directly to an AI agent, your build will likely collapse.

Prompt Pack

A Prompt Pack is phase-based prompts from your architecture spec, each tailored to a build stage and platform like Lovable, Cursor, Bolt, or Replit.

Spec-Driven Development

Spec-driven development (SDD) is a planning method where you write a full architecture spec before generating any code with an AI coding agent.

User Stories

User stories describe features in plain English. They help humans plan, but fed alone to a coding agent, they invite guessed architecture.

Verification Gates

Verification gates are manual checkpoints in an AI build. Test that the app still works before you give the agent its next prompt.

Vibe Coding

Vibe coding is building software by describing what you want in natural language and letting an AI write the code: intent first, not syntax-first debugging.

Vibe Coding Doom Loop

The doom loop hits when your AI agent starts breaking features faster than you can fix them. Here is why it happens and how to stop it.