Vibe coding vs agentic coding: what is the difference?
Updated Oct 26, 2026 · Sources listed at the end
Vibe coding is about the human’s attitude: you accept AI-written code without really reviewing it. Agentic coding is about the tool’s behaviour: an AI agent reads your codebase, edits files and runs commands to complete a task. You can do agentic coding carefully, or vibe code with an agent.
Vibe coding
Andrej Karpathy coined “vibe coding” in February 2025 for coding where you “forget that the code even exists” and accept all changes.1 Simon Willison narrows it to “building software with an LLM without reviewing the code it writes”.2
Agentic coding
Anthropic describes Claude Code as “an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools”.3 The defining trait is autonomy over several steps, not whether the human reviews the result. See what is agentic coding.
How they relate
| Vibe coding | Agentic coding | |
|---|---|---|
| What it describes | How much the human reviews | What the AI is allowed to do |
| Typical tools | App builders, AI editors | CLI agents, IDE agents, cloud agents |
| Review | Little or none, by definition | Your choice: review diffs, run tests, approve PRs |
An agent makes vibe coding easier, but the same agent can be used with full review. Willison’s rule for production work is not to commit code you could not explain to someone else.2
Footnotes
Tools mentioned
- Claude Code: paid from $20/mo
- OpenAI Codex: free plan, paid from $8/mo
- Cursor: free plan, paid from $8/mo
- Lovable: free plan, paid from $25/mo