What is vibe coding? Tools versus coding agents
By Flavio
Vibe coding is describing an app to a tool that scaffolds it while you stay out of the repo. How that differs from Claude Code, Codex, Cursor and Grok Build.
Vibe coding means you describe the app you want, and a tool scaffolds a running version for you, while you stay out of the repository. You talk about screens, data and the happy path. The tool picks the stack and the file layout.
IDE, CLI, or vibe coding tool is the post that owns this choice. It names Bolt, Lovable, v0 and Replit as the vibe coding tools, and it separates them from an IDE with an agent and from a CLI coding agent. This entry is the short definition, plus what those four coding agents document.
How vibe coding works
You write a description. The tool generates the app, hosts it, and lets you ask for changes in the same chat. You can ship something small without installing a bundler or wiring auth.
The cost is the decisions you did not make. When the app needs a behavior the tool did not assume, you are editing generated code you have not been living in, or you move the project into a repository and keep going with a coding agent.
None of Claude Code, Codex, Cursor or Grok Build documents a mode named vibe coding. A loose prompt in those tools is still agentic coding: the agent edits a real project.
Vibe coding in each tool
| Claude Code | Codex | Cursor | Grok Build | |
|---|---|---|---|---|
| A vibe coding mode | Not documented | Not documented | Not documented | Not documented |
| What the product is | A coding agent with tools that edit and run commands | An agent for a codebase, in the app, CLI, IDE or cloud | Agent in the editor, opened with Cmd+I | A coding agent in a terminal UI, a script, or another editor |
| Where the code lives | Your repository | Your repository, or a Codex cloud environment | Your project folder | The directory you started grok in |
Claude Code
Claude Code’s glossary defines agentic coding as the model reading files, running commands and making changes, as opposed to a chat that only returns text you apply yourself. That is the opposite shape from a vibe tool that keeps you out of the files.
A prompt such as “build me a status page” still runs inside the repository you opened. Claude looks for files, edits them, and can run the project’s commands. See the Claude Code glossary.
OpenAI Codex
Codex is for work with code, a codebase, or developer tools. A useful prompt names the behavior, points at the relevant code or the reproduction steps, keeps the constraints, and says how to check the change. The IDE extension includes your open files. In the CLI you mention paths yourself.
Local commands run in a sandbox. Crossing that boundary follows your approval policy. There is no documented product mode that scaffolds a hosted app from a description and leaves the repository out of it. See Prompting Codex.
Cursor
Cursor Agent completes coding tasks in the project, runs terminal commands and edits files. You open it from the side pane with Cmd+I. The agent page describes instructions, tools and a model. It does not describe a separate builder that hosts the app for you.
If you are not sure which files matter, the prompting page says to skip @ mentions and let Agent search. That is still work inside the project. See Cursor Agent.
Grok Build
Grok Build’s overview calls it a coding agent you run as an interactive terminal UI, headlessly with grok -p, or through the Agent Client Protocol in another app. The sample prompts are “Explain this repo.” and a file mention with @.
The overview does not describe a hosted app builder. See Grok Build.
Details checked against each vendor’s documentation in November 2026.
A real example
This prompt is the shape the vibe-coding post is about. It describes the product and leaves the implementation alone:
A public status page for flaviocopes.com. Three rows: the site, the newsletter, and the downloads host. Each row is operational, degraded, or down. One incident note with a time. No accounts and no monitoring probes.In a vibe coding tool, that description is the whole job. The v0 tutorial shows that style of build on Vercel’s v0: describe the app, then connect data and deploy.
The same words in Claude Code, Codex, Cursor or Grok Build need a repository and a check. Add the files that already exist, and say what must not change:
Add a public status page to this Astro site. Three rows: the site, the newsletter, and the downloads host. Each row is operational, degraded, or down. One incident note with a time. Do not change the RSS feed. Run npm run build when the page is in place.Plan mode is the step to take first when that second prompt touches a site you already maintain.
When to use vibe coding
The decision post sends you to a vibe tool when you are new and the setup itself is the thing blocking you, or when you already know how to build the app and you want a first version of the forms and the screens without spending the day on them. The concrete cases it names are landing pages, small CRUD apps, internal dashboards, and checking whether an idea is worth two weeks.
Skip it when you need to understand the architecture, set conventions, refactor, or write tests. That post sends that work to an IDE or a CLI agent, because you have to be in the codebase to do it.
15 apps you can vibe code instead of subscribing is a list of small replacements, each with a scope and a list of things the first version should not include.
Pitfalls
The decision post’s tradeoff is the one that shows up in practice. You outsource a lot of decisions, and a specific or unusual requirement is where the generated app stops being easy to change. That is the signal to move the project into a repository.
A coding agent does not remove that problem by accepting a vague product prompt. Claude Code’s prompt library says to describe the outcome and to give the agent a way to check its work. “Build me a status page” without a repository, a boundary, or a command to run is how an agent edits the wrong files.
Vibe coding tools and these four agents also bill and host differently. This entry does not compare their prices. Check the product you are about to use.
Related on flaviocopes.com
- IDE, CLI, or vibe coding tool is the choice between a vibe tool, an editor, and a CLI agent, based on the project in front of you.
- 15 apps you can vibe code instead of subscribing keeps each build small and names what the first version leaves out.
- v0 tutorial walks through building and deploying one app with Vercel’s v0.
- Draw an architecture diagram of an app your coding agent built is the step after the app is a real repository. Architecture Dissector draws the map from the code, which is the point where you have left the vibe tool.
Related glossary entries
- Agentic coding: Agentic coding means directing an agent that reads your repo, edits files and runs commands. How that loop works in Claude Code, Codex, Cursor and Grok Build.
- Coding agent: A coding agent uses a model and tools to read a repo, edit files and run commands in a loop. How Claude Code, Codex, Cursor and Grok Build each do that job.
- Plan mode: Plan mode makes a coding agent research and propose a plan before it edits code. How it works in Claude Code, Codex, Cursor and Grok Build, and when to use it.
Want me to talk about your product? You can sponsor this site.