Field test3 min read
Trying to stop typing everything
Testing MacWhisper, Handy, Monologue, Willow, and Voxtype — why I ended up sticking with Hex, and where talking to a computer actually beats a keyboard.
If you looked through my Applications folder right now, you’d probably think I had a minor dictation problem: MacWhisper, Handy, Monologue, Willow Voice, Voxtype, and Hex have all spent time sitting in my menu bar.
A couple of years ago I’d have laughed at the idea of talking to my MacBook. Typing is fast, quiet, and predictable. Then AI coding agents happened, and suddenly half my day stopped being "type careful syntax into an editor" and became "explain three paragraphs of context, edge cases, and intent to an agent".
Typing out long prompts and Slack updates all day gets old fast. Talking at roughly 150 words a minute while looking at a diff or walking around the room feels way more natural — if the tool stays out of your way.
Why Hex has been the best so far
Most voice apps fall into one of two traps: either they send every syllable off to a cloud API with a monthly subscription and a half-second pause that makes you lose your train of thought, or they’re local but clunky enough that you forget to use them after day three.
I’m using Hex now, and it’s easily been the best of the bunch for me:
- Completely local on Apple Silicon. It runs speech recognition locally on Metal (Parakeet / Whisper models), so nothing leaves the machine and it feels practically instant when you let go of the key.
- Zero-ceremony push-to-talk. Hold a key, say what you mean, release, and the text lands straight where your cursor already is. No floating chat window, no copy-pasting out of a separate app.
- Custom vocabulary and cleanup pipelines. Developers say weird proper nouns all day —
Convex,APFS,CoreWLAN,Mise,Tillo. Being able to teach a local tool your own terms and run lightweight output rules without paying a cloud tax makes a massive difference. - Open source. If a tool is going to listen to my microphone all day, I’d much rather it be MIT-licensed and inspectable.
Having a proper mic quietly helps a lot
One slightly unexpected thing I noticed: desk audio gear matters way more for dictation than people admit.
Because I already had decent audio kit on my desk (Focusrite / Vocaster interfaces and a RØDE mic), feeding a dictation model clean, dry speech instead of a reverberant laptop mic sitting next to a cooling fan cuts down weird mishearings dramatically. You stop having to speak in that awkward "robot talking to Siri" cadence and just talk normally.
Where voice actually wins (and where it doesn’t)
I haven’t thrown my keyboard in the sea. Voice is brilliant for a few specific things and genuinely terrible at others:
Great for voice:
- Giving context-heavy instructions to coding agents, where explaining why matters more than exact punctuation.
- Rough first drafts of notes, Slack replies, and emails — it’s much easier to edit a messy paragraph you spoke in ten seconds than stare at a blinking cursor.
- Capturing quick ideas in Obsidian or Todoist before you forget them.
Still way better on a keyboard:
- Editing existing code, renaming symbols, or navigating files.
- Precision edits in the middle of a sentence.
- Anywhere other people are trying to concentrate in the same room.
It’s probably not a coincidence that this exact habit spilled over into hardware too — half the reason I spent a year building OpenCharm was wanting that same effortless hold-to-talk loop sitting on my desk without even reaching for the keyboard.
For now, Hex sits quietly in the background, handles 90% of the long-winded typing I used to dread, and lets me save my fingers for the code.