# Vibe Coding

AI Glossary · Prompting & Context
Canonical URL: https://aisat.uk/glossary/vibe-coding
Author: Lucy Gilmore, Assistant General Counsel, Product, Privacy & AI at Sprinklr | Practical Privacy, Product & AI Governance & Defensible Risk Decisions (https://www.linkedin.com/in/lucy-gilmore/)
Breadcrumb: [Glossary](https://aisat.uk/glossary) > [Prompting](https://aisat.uk/glossary?category=prompting) > Vibe Coding

## TL;DR

> **Vibe coding**, coined by Andrej Karpathy in early 2025, is a style of software development where a builder describes features in natural language to an AI coding agent and iteratively accepts generated code based on runtime behaviour rather than manually inspecting every line.

## How Vibe Coding works in practice

In a vibe-coding workflow, the developer shifts from typing syntax to directing an agentic IDE or CLI tool (such as Claude Code, Cursor, Windsurf, or GitHub Copilot Agent) using conversational prompts, screenshots, and pasted error logs. The coding agent reads the repository, plans multi-file edits, installs packages, runs terminal commands, and launches a local dev server. When a layout breaks or a test fails, the builder pastes the stack trace or asks the agent to fix the behaviour until the application works.

To keep coding agents effective as a repository grows beyond a weekend prototype, experienced teams moving from pure 'vibe coding' to disciplined agentic engineering add structural guardrails to the repo itself: a concise `AGENTS.md` or `CLAUDE.md` file documenting architectural rules and commands, strict TypeScript compiler flags, automated linter hooks, and fast unit/Playwright test suites that give the coding agent deterministic feedback loops.

For example, a London product manager with no React background vibe-coded a working internal pricing-calculator web app in an afternoon to validate workflow requirements with sales teams before handing the validated spec to platform engineers.

## Why it matters and trade-offs

Vibe coding has collapsed the cost of disposable software, internal admin tools, UI prototypes, and one-off data scripts to near zero. Product managers, designers, and domain experts can now validate ideas with working software rather than static slide decks, while senior engineers use agentic coding to scaffold boilerplate and migrations in minutes.

However, shipping unreviewed, purely vibe-coded software straight into production customer systems accumulates severe security and maintainability debt. Without human code review and architectural stewardship, agents routinely introduce duplicated logic, N+1 database queries, permissive CORS policies, missing authentication checks on API routes, or hallucinated package dependencies (slopsquatting). The golden rule for engineering teams is simple: vibe-code prototypes freely, but enforce strict human review, automated security scanning, and deterministic test suites on every commit that touches production.

## Related AI concepts

- [AI Agent](https://aisat.uk/glossary/ai-agent): An AI Agent is a software system powered by a large language model that perceives environment context, maintains state, formulates multi-step plans and executes actions across external tools, APIs and databases to complete a delegated goal.
- [Context Engineering](https://aisat.uk/glossary/context-engineering): Context Engineering is the engineering discipline of dynamically selecting, structuring, compressing and ordering the system instructions, retrieved knowledge, tool definitions, memory and examples placed inside a language model's context window at inference time.
- [Shadow AI](https://aisat.uk/glossary/shadow-ai): Shadow AI is the unsanctioned use of consumer chatbots, unvetted SaaS AI features, browser extensions or personal API keys by employees to process proprietary company data or customer records outside IT and information-security governance.
- [Evals (Evaluations)](https://aisat.uk/glossary/evals): Evals (Evaluations) are automated, repeatable testing pipelines that score an AI application's outputs against a golden dataset using deterministic assertions, reference metrics and LLM-as-a-judge rubrics before any prompt or model change ships.