Speaking Beats Typing

I’m three sentences into a thought, I fat-finger a word, and I stop to fix it. By the time the cursor is back where it was, the end of the thought is gone. I’d been doing that for twenty years and assumed it was what writing costs. Speaking doesn’t do that to me.

Typos derail the thought

Everyone who recommends voice-to-text leads with speed, and they’re right that it’s faster. I talk several times faster than I type. Speed still isn’t why I switched.

Speaking lets me stay in one mode. When I type, I’m composing and correcting at the same time, and the correcting keeps winning. A red squiggle pulls my eye. A missing comma nags. I go back to fix it, and the fix costs me the thing I was about to say. Multiply that across a long prompt or a rough draft and what comes out is shorter and shallower than what I had in my head. Slower too, but the shallowness is the expensive part.

Speaking splits those apart. I say the whole thing, mess and all, and then the AI and I clean it up in a second pass. That separation is the trick, and it’s why this works better now than it did back when dictation software stood on its own. Cleanup used to be my problem. Now it’s a paragraph of instruction.

You do have to check the output. While I was talking through this article, Handy heard “voice-to-text” and gave me “voice attacks.” Nothing catches everything, and the errors it makes are real words, so a spellchecker won’t save you. Read what comes out. Reviewing it is still faster than typing it, and reviewing is a different job from thinking.

I don’t use it for everything. I still type code, because identifiers and punctuation are miserable to speak and faster to type right the first time. And there’s the problem of talking out loud around other people, which you solve by lowering your voice or walking somewhere else. Neither of those is a reason to skip the other 80% of what you write.

One hotkey, every app

Claude and ChatGPT both have voice input built into their apps, and both work fine. If chatting with an AI is all you want, stop here, you already have it.

The trouble is that most of what I write isn’t in a chat window. It’s Slack replies, email, PR descriptions, meeting notes, a comment on a Google Doc, some text field on a web form I’ll never visit again. A voice mode that lives inside one app can’t touch any of that.

You want a hotkey that works everywhere. Hold it, talk, release, and the text lands wherever your cursor already was. Once it works in every text field on the machine, you stop thinking of it as a feature and start using it as your keyboard.

The options

All four work system-wide. On the privacy column: Handy processes your audio on your machine and never uploads it. superwhisper does the same if you pick a local model, though it also offers cloud models. Wispr Flow doesn’t document local processing on its site, so assume your audio leaves the machine. If you dictate anything about work, start the decision there.

Twenty dollars a month for a keyboard

superwhisper and Wispr Flow are both good. I used them. I stopped because I don’t want a monthly bill for the ability to talk to my computer.

The money isn’t the problem. Voice input is plumbing. You don’t subscribe to your trackpad or pay a monthly fee for copy and paste, and dictation belongs in that same category, handed to you by the operating system and working in every app. Charging rent on it is the wrong shape.

And it’s never one subscription. It’s the eleventh, each easy to justify on its own, adding up to a number you’d refuse if anyone showed it to you as a single line.

What I use

Handy. Local, private, free, open source, and good enough.

I mean good enough. The paid tools beat it in places. They don’t beat it by enough to bill me every month for the rest of my life.

Install it and hold a key while you talk. That’s the whole product.

What’s coming, and why it doesn’t change my answer yet

Apple is shipping much better system-wide dictation in macOS 27 Golden Gate and iOS 27, running a new model on-device. It handles punctuation, capitalization and formatting on its own. In Apple’s own testing, evaluators preferred it over the previous dictation 44.7% to 17.6%.

Look at what that compares. It’s the new Apple dictation against the old Apple dictation, and the old one was terrible. Nobody has published a head-to-head against Handy or the paid tools, so treat “Apple wins” as unproven.

Two other catches. It ships off by default and buried in Settings, so most people won’t find it. And it needs an M3 Mac or newer with at least 12GB of RAM, or an iPhone 17 Pro or Air. An M1 or M2 Mac gets nothing.

So use Handy now. It’s free, it runs on the hardware you already own, and it works today. When macOS 27 ships in September, go check whether Apple’s version is better on your machine. If it is, delete Handy and use the built-in one.

Starting this week

1. Install Handy and pick a hotkey you can hold without thinking about it. I use ⌃ ⌥ ⌘ (Control, Option and Command) held together. Nothing else on my Mac wants that combination, and my hand finds it without looking.
2. Use it for one thing only. Slack replies, or first drafts, or prompts. One.
3. Let the first week be bad. You’ll trail off, restart sentences, and say “um” into permanent text. That stops after a few days.
4. Stop correcting yourself out loud. Say it wrong, keep going, fix it in the cleanup pass.

The last one is the habit that matters. The rest is setup.

If you’re running the AI Chief of Staff project, this is what makes the capture step survive a real week. A thought you have to stop and type is a thought you skip; one you can say in fifteen seconds gets filed.