Vibe Coding Glossary: 40 Terms Every AI-First Developer Should Know

Vibe coding glossary collects the 40 terms every AI-first developer hears in Discord, on X, and in tool changelogs — often used interchangeably when they mean different things. MCP is not the same as an API key. Vibe coding is not the same as vibe engineering. Smart Animate is not Framer Motion. This reference defines each term in plain language with enough context to use them correctly in prompts, job posts, and architecture docs. Updated November 2026.

Start with the vibe coding roadmap, what is vibe engineering, and vibe coding vs vibe engineering. Deep dives: what is MCP, context window prompts, Smart Animate.

Table of contents

1. Core concepts
2. AI tools and agents
3. Prompts and context
4. Design and handoff
5. Backend and security
6. Workflow and quality
7. FAQ

Core concepts: vibe coding and beyond

Vibe coding

Building software primarily through natural-language prompts to AI coding tools, iterating on generated output rather than typing most code manually. Coined in the indie builder community; implies speed and intuition over formal specs. See roadmap.

Vibe engineering

The disciplined upgrade of vibe coding: same AI tools, but with tests, reviews, architecture docs, security checks, and deployment pipelines. Production-minded. See definition.

AI-first developer

A builder who defaults to AI agents for scaffolding, refactors, and debugging — but owns validation, architecture, and shipping. Not “AI writes everything without review.”

Agentic coding

AI systems that plan multi-step tasks, edit multiple files, run terminal commands, and loop until tests pass — beyond single-line autocomplete. Compare Codex vs Claude Code.

Prompt-driven development

Software workflow where prompts (and context files) are first-class artifacts alongside code — versioned, reviewed, and reused like functions.

Ship velocity

Time from idea to deployed feature. Vibe coding optimizes this; vibe engineering optimizes it without sacrificing reliability.

Human-in-the-loop

Design pattern requiring human approval at critical steps — merges, deploys, schema changes — rather than fully autonomous AI deployment.

Export moment

When you move from a hosted AI app builder (Lovable, Bolt) to local git repo for advanced work. Often where layout and auth bugs surface.

AI tools and agents

Cursor

AI-native IDE (fork of VS Code) with inline completion, chat, Composer agent mode, and .cursor/rules. Common daily driver for vibe coders. Compare Claude Code vs Cursor.

Claude Code

Anthropic’s terminal-first coding agent using Claude models, CLAUDE.md, and MCP. Strong for multi-file TypeScript features.

OpenAI Codex

OpenAI’s coding agent/product line for generation and agentic tasks across ChatGPT and IDE integrations. Strong for scripts and polyglot glue.

Lovable

AI app builder generating full-stack React apps with Supabase — fast MVPs, export to git for extension. Review.

Replit Agent

Replit’s AI agent building and deploying apps in cloud workspaces — strong for prototypes and education. Review.

GitHub Copilot

Microsoft/GitHub inline AI assistant — completion, chat, PR summaries. Less autonomous than Claude Code; pairs well with agents.

Windsurf

AI IDE competitor to Cursor with Cascade agent flow. Compared in Copilot vs Cursor vs Windsurf.

v0

Vercel’s AI UI generator — React/Tailwind components from prompts. Starting point, not production app.

Bolt.new

StackBlitz AI app builder — in-browser full-stack prototypes with fast spin-up.

MCP (Model Context Protocol)

Open standard for AI agents to call external tools — databases, Figma, browsers, WordPress — with structured schemas. Full guide.

MCP server

A service implementing MCP that exposes tools (e.g., query, get_post) to agents. You configure servers in Cursor or Claude Code settings.

Prompts, context, and tokens

Context window

Maximum text (tokens) an model can process in one session — code files, chat history, rules. When full, early context drops. Manage with context window prompts.

Tokens

Units of text for LLM billing and limits — roughly 4 characters English. “Running out of tokens” = context or API budget exhausted. Not the same as design tokens or motion tokens.

CLAUDE.md

Repo-root markdown file loaded by Claude Code (and compatible agents) defining stack, conventions, forbidden patterns, and key file paths. External memory for agents.

Cursor rules

.cursor/rules/*.mdc files with optional glob scopes — project conventions injected into Cursor agent context automatically.

System prompt

Hidden instructions shaping agent behavior — tone, constraints, output format. Template library: system prompt templates.

Prompt chain

Sequence of dependent prompts where each step’s output feeds the next — safer than one mega-prompt for refactors. Guide.

Prompt pack

Curated copy-paste prompts for a domain — auth, API, accessibility, debugging. Reusable across projects.

Chunking

Breaking large tasks into vertical slices with explicit scope and handoff notes — essential for context limits.

Session primer

Opening prompt in a new chat: read CLAUDE.md, confirm stack, state slice scope before any edits.

Handoff note

Short markdown summary at end of agent session — decisions, files changed, next steps — for the next fresh context.

Design, motion, and handoff terms

Design tokens

Named variables for color, spacing, typography, radius — single source of truth from Figma to CSS. Variables handoff.

Motion tokens

Named duration, easing, delay, and spring values for animation — bridge Smart Animate to CSS/Framer Motion.

Smart Animate

Figma prototype transition interpolating changed layer properties between frames — position, size, opacity, radius.

Dev Mode

Figma inspector for developers — specs, variables, code snippets, asset export. Handoff anchor for AI builders.

Figma Make

Figma’s AI design-to-code feature — generates UI from prompts inside Figma; production gaps remain. Make to production.

Design-to-code

Pipeline from Figma (or sketch) to implementation — manual, plugin-assisted, or AI-generated. Always needs review.

Handoff

Transfer of design intent to engineering — specs, tokens, motion, assets, edge cases — not just a PNG export.

Component library

Reusable UI primitives (shadcn, Radix) that agents should compose instead of reinventing buttons each prompt.

Backend, data, and security

RLS (Row Level Security)

Postgres/Supabase feature restricting which rows each user can read/write — critical for multi-tenant SaaS. AI often forgets policies.

BFF (Backend-for-Frontend)

Your API routes proxy third-party APIs — hides secrets, shapes responses, avoids CORS. Pattern AI violates without rules.

Server actions

Next.js pattern for server-side mutations callable from client — preferred over client-side secret calls in vibe-coded apps.

Idempotency

Repeating an operation (e.g., webhook) does not duplicate side effects — store processed event IDs.

Environment variables

Config secrets and URLs outside code — .env.local, Vercel dashboard. NEXT_PUBLIC_ exposes to browser; use sparingly.

Zod

TypeScript schema validation library — validate API bodies, forms, env at boundaries. Stops AI any types from reaching production.

Supabase

Open-source Firebase alternative — Postgres, auth, storage, realtime. Default backend in many vibe coding stacks.

Workflow, quality, and deployment

Vertical slice

One feature end-to-end (UI + API + DB) shippable independently — preferred chunk unit for agent sessions.

Diff review

Human inspection of every line AI changes before merge — non-negotiable for production. Review guide.

Debugging prompts

Structured prompts for fixing errors — paste logs, expected vs actual, files in scope. Pack.

Starter stack

Default tool combo for new projects — e.g., Cursor + Next.js + Supabase + Vercel. 2026 stack.

CI/CD

Automated test and deploy on git push — catches agent regressions you miss in manual review.

Prefers-reduced-motion

OS/browser accessibility setting — animations must degrade gracefully. Required for inclusive vibe-coded UI.

WCAG

Web Content Accessibility Guidelines — contrast, focus, labels. AI UI often fails without explicit prompts. Accessible UI prompts.

Soft launch

Deploy to limited users before marketing push — validate AI-generated code under real traffic.

Technical debt

Shortcuts that speed today and cost tomorrow — unreviewed agent output is a debt accelerant.

Vercel

Hosting platform optimized for Next.js — common deploy target for vibe-coded apps. Deploy guide.

Quick reference table

Use precise terms in prompts — “add RLS policies for user_id” not “secure the database.” Agents mirror your vocabulary in generated code and comments.

Using the glossary in prompts

Reference terms by exact name: “Implement BFF pattern for OpenAI calls” not “proxy the API somehow.” “Add RLS policies” not “make database secure.” Agents trained on community content recognize MCP, CLAUDE.md, and vibe coding — vague prompts produce vague code.

Onboarding new collaborators: share glossary link before repo access. Junior devs and no-code founders align vocabulary before first PR comment war over “what is a token.”

Terms we did not include (on purpose)

Brand names change quarterly — we skipped granular model version strings (GPT-5.x, Claude 4.x point releases). Check vendor docs for current SKUs. We also skipped meme terms without technical meaning (“slop,” “magic”) — they do not belong in CLAUDE.md.

Extended reading by term cluster

Update personal copy quarterly — add terms from your stack (tRPC, Prisma, Drizzle) in docs/glossary-local.md linked from CLAUDE.md without bloating the canonical 40.

Flashcard-style recap: 10 must-know terms

1. MCP — agents call tools, not guess APIs
2. CLAUDE.md — repo memory for agents
3. RLS — database rows scoped per user
4. Vibe engineering — fast AI builds with discipline
5. Prompt chain — sequenced safe refactors
6. Design tokens — named visual variables
7. Motion tokens — named animation variables
8. BFF — server proxy hiding secrets
9. Vertical slice — one shippable feature end-to-end
10. Export moment — leave hosted builder for git

Master these ten before the long tail — they appear in 90% of indie builder Slack threads and job posts in 2026.

FAQ

Is vibe coding a real engineering practice?

It is a workflow, not a job title replacement. Vibe engineering is the mature form employers and customers trust.

What’s the most confused pair of terms?

MCP vs API, and tokens (LLM) vs design tokens. Specify which you mean in team docs.

Should I add this glossary to CLAUDE.md?

Link to it, do not paste all 40 entries — save context window for conventions. One line: “Terminology: see docs/glossary.md.”

Which terms change fastest?

Tool names and agent features — verify quarterly. Concepts (RLS, idempotency, chunking) are stable.

Vibe coding glossary — 40 terms defined for AI-first developers navigating tools, prompts, design handoff, and production guardrails. Bookmark it, link from CLAUDE.md, and use precise language in your next agent session. Continue learning on the vibe coding roadmap and best tools 2026.

Quick reference: token naming cheat sheet

Keep motion tokens boring and predictable. A small shared vocabulary beats clever one-offs that AI reinvents every session.

• --motion-duration-instant: 80–120ms for toggles and checkbox feedback
• --motion-duration-fast: 150–200ms for hover and micro UI
• --motion-duration-base: 250–300ms for most transitions
• --motion-duration-slow: 400–500ms for page-level or modal entrances
• --motion-ease-standard: a calm cubic-bezier for general UI
• --motion-ease-emphasized: a slightly punchier curve for primary CTAs
• --motion-ease-exit: a curve that feels slightly faster on the way out

When you prompt Cursor or Claude, paste this cheat sheet into your rules file and say: “Use only these tokens. Do not invent new duration values.” That single instruction prevents half of the random 0.37s ease-in-out noise that shows up after AI UI generation. Pair it with the handoff habits in our Figma motion handoff guide and you get motion that survives production, not just a pretty prototype.