TL;DR: The future of voice computing is not a concept video; the pieces shipped. Speech recognition moved on-device (Infina runs NVIDIA's Parakeet model on the Apple Neural Engine, so it is private, fast, and works offline), and AI agents made long plain-English instructions the main input of a workday. Infina already closes the loop no other dictation app completes hands-free: from a couple of feet away, say "type" plus your prompt and it is typed, say "send" and Enter is pressed, say "open Claude Code" and you are in the next app, no keyboard at all. It costs $99 once (as of July 2026) with a 7-day refund.

The claims about what exists today are checkable on any Apple Silicon Mac.

Voice computing spent decades as a demo

Talking to computers has been "two years away" since the nineties. It kept failing for the same three reasons.

Recognition was not accurate enough, so fixing the transcript cost more time than typing it fresh. Everything round-tripped through a cloud server, which made it slow and made people rightly nervous about a live microphone. And there was nothing much worth saying: "set a timer" and "play music" do not change how work gets done.

All three broke at once, quietly, over the last few years.

What changed, part one: recognition moved on-device

Modern speech models are small enough to run on the chip already in your laptop. Infina ships NVIDIA's Parakeet model running on the Apple Neural Engine of any Apple Silicon Mac.

That one architectural fact fixes two of the three old failures. Latency collapses because audio never travels to a server, and dictation keeps working offline, on a plane, on hotel Wi-Fi, anywhere.

Privacy stops being a promise and becomes physics: by default, your audio never leaves your Mac, and no transcripts are stored. Accuracy crossed the useful threshold too, 95%+ for clear speech, so reading back a transcript stopped being an editing job.

The technical picture is in on-device dictation for Mac.

What changed, part two: AI gave us something worth saying

The other half of the shift has nothing to do with speech tech.

Work with AI tools is conducted in plain language. A day spent driving Claude Code, Cursor, or ChatGPT is thousands of words of instructions: describe the change, review the result, redirect, repeat. Natural language stopped being a toy input and became the primary one.

And language is the one input where voice beats hands decisively. Conversational speech is commonly cited at around 130 to 160 words a minute against roughly 40 to 60 typed; you do not need a study to confirm that, just a timer. We unpack why agents amplify this in AI agents need voice.

Keyboards were designed for an era when input meant code, commands, and precise text. Prompts are none of those. They are speech that happens to be typed.

What voice computing looks like today, concretely

Here is the state of the art on a Mac right now, not a roadmap slide.

Hold the Option key and talk; release, and your words appear in whatever app is focused: a terminal, an editor, a chat box. That is push-to-talk dictation, transcribed on your Mac.

Then there is the part that makes it computing rather than typing. Toggle Infina's hands-free mode (double-tap Cmd; it is labeled experimental and ships off by default) and the keyboard leaves the loop entirely:

  • Say "type" plus your words: "type refactor the auth flow but keep the tests green". They are typed into the focused app.
  • Say "send": Enter is pressed.
  • Say "open Notes", "open Cursor", or "open Claude Code": the Mac switches apps, and you repeat.

That loop runs from two or three feet away. You can stand up, pace, hold a coffee, and keep several agent sessions moving at once.

No other dictation app completes that prompt, send, and switch-apps loop hands-free in plain English. The walkthrough is in hands-free voice prompting, and the wider picture in hands-free computing on the Mac.

Where this goes: voice commands, keyboard edits

Here is our view of the next few years, labeled as exactly that: our view, not a promise.

The keyboard does not die. It gets demoted to what it is genuinely best at: precision. Renaming a variable, fixing one word, moving through text character by character. The keyboard becomes the editing tool.

Voice takes over the other job: intent at volume. Telling agents what to build, what to fix, what to try next, and moving between the windows where they work. Voice becomes the commanding tool.

The ratio between those two jobs is shifting fast. As agents get more capable, the human contribution becomes more describing and deciding, and less typing of the artifact itself. The instructions grow longer and more numerous while the hand-edits shrink.

If that holds, the bottleneck of the AI era is input bandwidth, and the cheapest bandwidth upgrade anyone can buy is the voice they already have.

What we are not predicting

A vision essay earns trust by its restraint, so here is ours.

We are not predicting the end of screens, ambient computing that whispers in your glasses, or a Mac you never touch. Reading remains faster than listening for output, so your eyes stay in the loop.

We are not claiming today's version is finished. Infina's hands-free mode is labeled experimental and ships off by default. The base product is English-only; the optional cloud add-on ($10 a month, with a 7-day trial) brings more languages and LLM-polished output through our cloud AI providers (Together AI and Groq).

It is Mac-only and needs Apple Silicon. And we make no roadmap promises here: everything above either ships today for $99 once (as of July 2026, with a 7-day no-questions refund, details on pricing) or is clearly labeled as opinion.

FAQ

What is the future of voice computing in one sentence? Voice becomes the primary way people command computers and AI agents, while the keyboard remains the precision tool for editing, a split that is already visible on Macs today.

Will voice replace the keyboard entirely? We do not think so, and we build voice software. Keyboards win at precise edits and exact syntax; voice wins at expressing intent at 130+ words a minute. The likely future is both, each doing the job it is best at.

Why is voice computing suddenly viable after decades of failure? Three things converged: speech models became accurate enough, they became small enough to run on-device (private and instant, with no cloud round-trip), and AI agents made long natural-language instructions the core of daily work.

Is voice computing private? It can be, when it runs on-device. Infina transcribes on your Mac's Neural Engine by default, so audio never leaves your device and dictation works offline. Cloud processing exists only as an optional paid add-on.

What can I actually do by voice on a Mac today? With Infina: hold Option to dictate into any app, and in hands-free mode say "type" plus your words to have them typed, "send" to press Enter, and "open" plus an app name to switch apps, all from a few feet away.

How much does this cost? Infina is $99 one-time as of July 2026, with every 1.x update included and a 7-day no-questions money-back guarantee. The optional cloud add-on ($10 a month, 7-day trial) adds polished output and more languages.

The bottom line

The future of voice computing stopped being futuristic the moment recognition moved onto the laptop's own chip and AI turned plain English into the workday's main input.

What remains is adoption: noticing that the sentence you were about to type is one you could have said, roughly three times faster, from across the room, while another agent works in the next window.

That is the bet Infina makes, and it is priced like a tool, not a vision: $99 once, refundable for 7 days. The future does not need your subscription.