Does dictation work inside the VS Code or Cursor inline comment threads?

Yes, both editors render inline review comments as standard text inputs that accept OS-level keyboard input, which is what Amical's push-to-talk hotkey produces. Click into the comment textarea, wait for the cursor to blink, then hold your hotkey. The same applies to the GitHub Pull Requests extension panel, the Cursor review view, and any third-party plugin that uses a normal HTML textarea or NSTextField. If dictation feels laggy inside the editor specifically, check that no other voice tool — macOS Dictation, Wispr, Whisper Flow — is also listening on a global hotkey and intercepting your audio.

How do I dictate code symbols like double-equals or curly braces?

You generally should not. Dictation engines produce inconsistent results for operators, brackets, and language-specific syntax, and fighting the engine costs more time than typing the characters yourself. The reliable workflow is to dictate the prose explanation of a code change by voice and then type the actual code with your hands, or paste it from the diff. For comments that need to reference a symbol like getUserById or a flag like --no-verify, add those terms to Amical's custom vocabulary so the transcriber knows to keep them as one token instead of breaking them into 'get user by I D'.

Will voice review work if I am pair-reviewing on a call?

Push-to-talk dictation is well-suited to pair review because the hotkey only captures audio while you hold it. You can talk to your reviewer freely, and only the words you intend as comment text get transcribed. Mute your microphone in Zoom, Meet, or Slack Huddles while dictating into the comment box if you do not want the other person to hear you narrating your own commentary. Many teams adopt the inverse pattern: one person reads the diff aloud while the other dictates the actual review comments, splitting the cognitive load between speaker and scribe.

How to do voice-driven code review (Cursor, VS Code, GitHub)

A repeatable workflow for reviewing pull requests by voice — dictating inline comments, structuring nitpick versus blocker feedback, and writing the review summary.

Code review is the single biggest typing tax most senior engineers pay. You read a hundred lines of someone else's code, then type two hundred words of feedback per file, then type a summary on top. Voice flips this: you read at full speed, talk through what you see as you see it, and let the dictation engine keep up. This guide walks through reviewing a real pull request in Cursor, VS Code, or directly on GitHub using Amical's push-to-talk hotkey. It covers leaving inline comments without losing your place, separating nitpicks from blockers in voice, and dictating a structured PR summary that the author will actually read.

Step by step

1
Open the PR in split view with the diff on one side and the comment box ready
On GitHub, open the Files Changed tab and resize the window so the diff fills two-thirds of the screen. In VS Code or Cursor, use the GitHub Pull Requests extension and pin the description in a side panel. The goal is zero context-switching while you talk. When you spot something worth commenting on, click the plus icon next to the line, wait for the comment box to appear, then hold your push-to-talk hotkey. Do not start dictating before the textarea has focus, or your first sentence vanishes into the void.
2
Prefix every inline comment with nit, question, or blocker
Train yourself to start every dictated inline comment with one of three words: 'nit colon', 'question colon', or 'blocker colon'. So you say 'nit colon this could be a constant instead of an inline literal' or 'blocker colon this overwrites the user record without a where clause'. The prefix tells the author exactly what action to take and stops you from drowning real issues in style nitpicks. Authors learn to scan the nits last and the blockers first, which makes your reviews land harder with less total commentary.
3
For complex suggestions, dictate the explanation then the GitHub suggestion block
When you want to propose a concrete code change, dictate the why first as plain prose, then type or paste the GitHub suggestion block syntax with the actual code inside. Voice handles the explanation; your fingers handle the code, because no dictation engine reliably produces correct syntax for arrow functions, generics, or async iterators. A good pattern: 'This will throw if the array is empty, period, new paragraph, suggestion colon' then drop in the three-line fix. The author gets clear reasoning plus a one-click apply.
4
Group related comments by file before writing the summary
Once you finish leaving inline comments, scroll back through the Files Changed view and look for clusters: three nits on the same file, two blockers in the same module, a question about test coverage. Open the main PR conversation tab and hold push-to-talk. Dictate a summary that groups by theme, not by file: 'Two blockers around input validation in the auth handler, period, several style nits in the formatter helper, period, one open question about whether we need a migration here, period.' Theme-grouping makes the summary scannable.
5
Close with a clear request-changes, approve, or comment verdict
Before you submit the review, dictate one final sentence stating your verdict and what you need from the author next. Examples: 'Requesting changes on the auth blockers, period, happy to approve once those are addressed, period.' Or: 'Approving with comments, period, the nits are optional but the migration question needs an answer before merge, period.' This single sentence stops the back-and-forth where the author cannot tell whether your fifteen comments are blockers or suggestions. Pick the GitHub review action that matches the sentence you just dictated.

Experience the future of dictation

Local and Cloud Models

Choose from powerful cloud-based AI models or run everything locally for maximum privacy and control. Switch seamlessly between providers to find the perfect balance of speed, accuracy, and security.

Custom Vocabulary

Customize the AI to recognize your specific terminology, jargon, and proper nouns for industry-specific accuracy and personalized transcription results.

Project PhoenixMcConaughey

CI/CD pipelinek8s

SOC2PostgreSQL

JiraROI metrics

⌘Space

⌥⌘A

CtrlAltSpace