If your hands hurt or you cannot rely on them, you do not need to solve voice coding in one go. Start with the smallest change that cuts the most typing: dictate prose and AI prompts. Move to full hands-free control only if you need to replace the keyboard and mouse entirely. This post lays out that ladder, what the vendors' own pages confirm as of 2026-10-09, and a staged setup plan for Talon, the most capable option I found.
I did not install or run any of these tools for this article. Everything about products comes from their pages, read on 2026-10-09, and where a page did not load or did not say, I say so. This is not medical advice. Repetitive strain injury needs proper assessment; treat this post as a tooling guide only.
What are the options for hands-free coding?
Think in four rungs. Each costs more setup and removes more typing.
| Rung | What it does | What the vendor page says (2026-10-09) | Main limit |
|---|---|---|---|
| 1. Dictation app | Speech to text at the cursor | Wispr Flow lists dictation on Mac, Windows, iOS and Android, free up to 2,000 words a week on desktop; its privacy page says transcription is in the cloud | Does not control the computer; poor for symbols |
| 2. Built-in OS voice control | Speech commands and text | Windows voice access: Windows 11 22H2 or later and offline once set up, per third-party guides (see below). macOS Voice Control: replaces standard Dictation while on | Generic commands, not tuned for code |
| 3. Dictation plus AI coding agent | You speak the intent, the agent types the code | See below | You still review code, with eyes and some clicks |
| 4. Talon | Hands-free keyboard and mouse replacement | "Commands and dictation," mouse by looking, pop to click, hiss to scroll | Learning curve; setup time |
Rung 3 deserves emphasis. If your goal is less typing, an agent can do the typing while you speak the instructions. The workflow in voice coding with Wispr Flow, Claude Code and Cursor is built around that. The catch for accessibility is that you still have to read diffs, run commands and approve actions, and those need either a mouse, a keyboard or a hands-free system on top.
What do the built-in tools actually offer?
Both major desktop operating systems ship voice control, and it costs nothing extra.
Windows voice access. Third-party guides say it needs Windows 11 22H2 or later and works without an internet connection once the speech model is downloaded; I could not re-confirm the 22H2 minimum or the offline detail on Microsoft's own page on 2026-10-09, so check it there. Microsoft's update-history page does say fluid dictation is for Copilot+ PCs and English locales. You can switch between commands mode and dictation mode by voice, and say "Add to Vocabulary" to teach it terms. Fluid dictation requires a Copilot+ PC and is listed for English locales. I found nothing on coding commands.
macOS Voice Control. Apple's page says that when Voice Control is on, you use it to dictate text and standard macOS Dictation is not available, and that you can say "Show commands" to see what is available in your current app. It links to guides on custom vocabulary and customising commands, which I did not read in full. The page does not mention offline use.
Neither page promises good support for symbols like braces and underscores. Test that yourself with a ten-line function before you commit.
What is Talon and what does setup involve?
Talon's wiki describes it as "a hands-free input replacement for the keyboard and mouse" for people who have limited hand use or want to stop using their hands. Its homepage lists downloads for macOS, Windows and Linux (X11), and shows a short custom command file where saying "select all" presses the select-all shortcut. It does not state the speech engine, noise tolerance or system requirements on that page, so I cannot tell you those from the homepage.
The wiki's speech engine page adds more, read on 2026-10-09:
- Conformer is described as the recommended engine for new users. It is free and installed from the Talon tray icon.
- Dragon works on Windows only and costs roughly 300 to 500 dollars; the page recommends the professional version because the home version lacks command mode.
- Webspeech and Vosk are for dictation in other languages and need Talon Beta; they cannot run alone because they handle dictation only.
- The page says that as of April 2024 the main Conformer model supports only English. Check the wiki for anything newer.
The wiki also says beginners should install the Talon community user file set, and links separately to hardware advice on microphones and eye trackers. I could not load the install guide or the microphone page, so I cannot list steps or specific hardware here. Read those pages directly.
Talon is free to use and funded by Patreon donations, per the wiki, and a Talon+ tier is promoted on the homepage without a visible price.
A staged plan for adopting Talon
Do not try to switch overnight, especially if you need to keep earning. This plan is my suggestion, not a tested curriculum.
- Week 0: keep your current setup. Reduce strain with a physiotherapist's advice and a dictation app for emails and prompts. Treat Talon as a side project.
- Week 1: install and learn the alphabet. Install Talon, the community file set and a Conformer engine. Learn how to speak letters and dictate a sentence. Do nothing else.
- Week 2: editing and navigation. Learn the commands for moving the cursor, selecting and deleting. Editing is where voice coding gets slow, so practise it on plain text first.
- Week 3: your editor and terminal. Learn the commands for your own editor and shell. Keep a list of the ten you use most.
- Week 4 and on: custom commands for your daily work. Add your own
.talonfiles for repeated actions. The homepage example is the pattern: a spoken phrase, a colon, then an action. I have not verified more syntax than that. - Ongoing: keep a fallback. Keep a keyboard and an alternative input available for the days voice fails, such as a cold or a noisy room.
Expect the first weeks to feel slower than typing. Whether you recover your old speed is something I cannot promise and have not measured.
How do you combine voice with AI coding tools?
For many developers the biggest saving is not syntax by voice, it is letting an agent write the syntax. A practical split:
- Speak: goals, context, constraints, review feedback ("keep the function signature, change only the retry logic").
- Let the agent type: the code itself.
- Use hands-free control or short, rare keyboard use for: approvals, running tests, git operations.
Our guide to speaking a good AI prompt gives a four-part structure (goal, context, constraints, output) that works well when speaking is the main input. Dictating code reviews and commit messages covers two other prose-heavy tasks.
One caution: letting an agent run commands without keyboard review is risky, because a misheard instruction can become a destructive action. If your hands make approvals hard, narrow what the agent can do rather than approving everything blindly. See AI coding agent safety, sandbox and permissions.
What could go wrong?
- Your voice. Heavy dictation can strain your voice. Pace yourself, drink water, and stop if it hurts; ask a professional if it persists.
- Noise and shared spaces. Open offices and family homes are hard. Hardware advice for microphones is on Talon's wiki, which I could not load.
- Accents and language. Talon's main model was English only on the wiki's page. Windows voice access lists several English variants per secondary sources that disagree; check the languages panel on your own PC.
- Privacy. Cloud dictation sends audio to a server. If you work under NDA, use the on-device options or ask first.
- Fatigue and morale. Learning a new input system while in pain is hard. Move at the pace your work allows, and ask a colleague or community for help.
If you are comparing the dictation apps first, the dictation tool comparison covers Wispr Flow, Superwhisper, Talon and OpenWhispr side by side.
