# VibeCode with your voice.

Vibe coding with voice is not about dictating syntax. It is about explaining the intent behind a change while the context is still in your head. Vara can turn that explanation into a prompt that Codex, Claude Code, Cursor or Gemini CLI can act on more reliably.

Developers write more text with AI, not less: tasks, constraints, edge cases, test requirements and context.

VibeCode mode lets you speak like a developer and get a clearer agent instruction on the other side.

## Speak intent, not syntax

The weak version of developer dictation is speaking brackets and semicolons. The strong version is: “this component should not own state anymore; lift it up but keep the API stable.”

That is exactly the kind of explanation an AI agent needs.

## Good tasks for voice

**Bugfix:** what happened, what you expected and where you suspect the issue.

**Refactor:** what should change, what must not change and how to test it.

**PR review:** risk, reviewer focus and what was intentionally left out.

**Docs:** explain the flow before fighting the perfect first sentence.

## Example output

**Spoken:** “This guide page has a CTA button inside prose and the text color is overridden by link styles. Fix it without breaking normal links.”

**Prompt:** “Find CTA links inside prose where `.prose a` overrides button color. Add scoped CSS so primary CTA text remains readable while normal prose links keep their styling. Verify build and pseo.”

## Where typing is still better

Use the keyboard for filenames, commands, snippets and exact errors. Use voice for context, goals and decisions.

That turns VibeCode from a gimmick into a faster path from loose thought to useful agent task.

## Sources

- [Wispr Flow: dictation for developers](https://wisprflow.ai/developers)
- [Superwhisper: Claude Code dictation workflow](https://superwhisper.com/claude-code)
- [Vara: Maskinrummet](/maskinrummet)