Skip to main content

How I actually dictate code (and what I still type)

· 2 min read
Driss Lahbil
Software engineer, editor of Wispr Flow Guide

Every few months someone posts that they now "code with their voice", and every developer who has tried it winces. Dictating literal syntax is miserable. But there is a version of voice-driven coding that genuinely works, and I use it daily. Here is the honest split.

What I dictate​

  • AI prompts. This is the big one. Talking a full, contextual request to an assistant in my editor beats typing a terse one, and the code I get back is better for it.
  • Comments and docstrings. Explaining intent out loud is faster than typing it.
  • Commit messages and PR descriptions. Pure prose.
  • Slack, issues, and docs. All the writing around the code.

What I still type​

  • Syntax. Brackets, operators, indentation. Dictating these is slower and error-prone.
  • Exact names. Unless they are in my dictionary, function and variable names get mangled.
  • Small precise edits. The keyboard wins for one-character fixes.

The workflow​

I keep the AI chat pane focused, dictate the change I want in detail, let the assistant write the code, then type the small corrections. The commit message gets dictated too. My hands do far less, and the parts they do are the parts that need precision.

The one setup step that matters​

Add your stack's names to a custom dictionary: libraries, APIs, your product. It stops the constant re-correcting that otherwise kills the whole thing. I wrote up the details in the dictate code guide and the custom dictionary guide.

If you want the tool I use for this, it is Wispr Flow; the honest review covers where it does and does not help.