TL;DR: In agentic coding, your agents write the code; you write the prompts. That makes prompt speed the ceiling on how fast you ship, and speaking is commonly cited at roughly three times typing speed. Infina takes it further than dictation: lean back a couple of feet from your Mac and run the whole loop by voice, say "type" plus your prompt and it gets typed, say "send" and Enter is pressed, say "open Claude Code" or "open Cursor" and you are directing the next agent, hands never touching a key. On-device on Apple Silicon, $99 once (as of July 2026), 7-day refund.
Iteration speed is prompt speed
The shape of software work changed. With Claude Code, Codex, and Cursor doing the typing of code, your output is no longer measured in lines written. It is measured in iterations: prompt, review, prompt again.
Each iteration has two human steps. Reading the agent's output, which is fast because you skim diffs well. And writing the next prompt, which is slow, because a good prompt is a paragraph of context, constraints, and direction.
That second step is the one you control. If you prompt faster, you ship faster, because everything else in the loop is the machine's problem.
The full argument for the category is in why voice prompting. This essay is the builder's version.
Idle agents are the new idle compute
Here is the uncomfortable arithmetic of a typed workflow.
A detailed prompt runs 100 to 150 words. At a typical 50 words per minute of typing, that is two to three minutes at the keyboard. Speaking the same prompt at a commonly cited 130 to 160 words per minute takes under a minute. No study needed; time yourself once.
Now multiply by your day. Twenty prompts is somewhere around 40 to 60 minutes of typing, versus roughly 15 to 20 minutes of speaking. Run three agents instead of one and the typed version stops scaling: while you type for agent A, agents B and C sit finished, waiting.
Every minute an agent waits on your fingers is idle compute you already paid for. You would never tolerate a CI pipeline that idles because someone is hand-typing the config. That is what a keyboard-only prompt loop is.
The hands-free loop: review one agent, dictate to the next
Dictation alone narrows the gap. But ordinary dictation apps still route the loop through your hands: hold a hotkey to speak, press Enter to send, Cmd-Tab to the next window. Twenty prompts a day is sixty small keyboard interruptions around the speaking.
Infina closes the loop entirely. Double-tap Cmd to switch on hands-free mode (it is experimental and off by default; we say so because it is true). Then, from two or three feet away:
- Say "type" plus your prompt: "type add rate limiting to the upload endpoint and update the tests". Infina types it into the focused window. No hotkey; the word "type" is itself the trigger.
- Say "send": Infina presses Enter and the agent starts working.
- Say "open Cursor" (or "open Claude Code", "open Notes"): you are in the next window, and the loop repeats.
No other dictation app completes that prompt, send, and switch-apps loop hands-free in plain English. That specific loop is the moat.
And it unlocks the real trick: pipelining yourself. Your eyes review agent A's diff on screen while your voice is already dictating agent B's next prompt. Reading and speaking are different channels; typing and reading are not. A keyboard forces those steps to run in series. Voice runs them in parallel.
That is the difference between owning agents and babysitting them. The multi-agent playbook is in voice prompting for multiple agents, and the terminal-specific walkthrough is hands-free Claude Code.
Why raw dictation is right for prompts
Infina's base product is deliberately raw: on-device transcription (NVIDIA Parakeet on the Apple Neural Engine) with fast, regex-level formatting. No cloud round trip, works offline, audio never leaves your Mac by default.
For prompting, raw is the correct default. Claude Code does not care if your prompt says "gonna" or has a comma splice. It cares that you gave it the file names, the constraint, and the goal. Shipping your words to a server to be beautified before they hit a terminal would just add latency to every iteration.
When you do want polish (human-facing emails, docs, more languages), the optional cloud add-on is $10 per month with a 7-day trial, using our cloud AI providers (Together AI and Groq). That is the polished-prose apps' whole pitch, sold as a $15-per-month-forever subscription; here it is an add-on you switch on only if you want it.
Hold Option for classic push-to-talk any time; hands-free and push-to-talk coexist.
The cost side of shipping faster
The loop math becomes money math quickly, and the long version lives in voice prompting ROI.
The short version: Infina is $99 one-time as of July 2026, every 1.x update included, no subscription. There is no free trial; there is a 7-day no-questions refund, which is a full week of real work on your real agents. Details on pricing.
If speaking your prompts saves you even 20 minutes a day, a one-time $99 against a builder's hourly value pays for itself in the first couple of weeks. That arithmetic is yours to run, not ours to inflate.
Honest limits
- Mac only, Apple Silicon required for the on-device models.
- English only in the base product; the cloud add-on covers more languages.
- Hands-free mode is experimental and off by default. Push-to-talk is the mature path while you build trust in it.
- Voice writes prompts, not code. Symbol-dense editing stays on the keyboard, where it belongs.
- A reasonably quiet room helps at two to three feet; a loud open office degrades any microphone.
FAQ
What does "prompt faster, ship faster" actually mean? In agentic coding, agents do the typing of code and you do the prompting. Prompt speed becomes the ceiling on iteration speed, so producing prompts faster directly raises how much you ship per day.
Is speaking prompts really faster than typing them? For most people, yes, by roughly three to one. Speech is commonly cited at 130 to 160 words per minute versus 40 to 80 for typing. Time yourself dictating one real prompt and typing the same one; the ratio is easy to verify.
Does Infina work with Claude Code, Codex, and Cursor? Yes. Infina types into whatever app has focus, at the OS level, so it works in any terminal, editor, or chat window with no per-app plugin. Say "open Claude Code" or "open Cursor" to switch between them by voice.
Can I really run multiple agents this way? Yes, and that is where voice pays most. You review one agent's output on screen while dictating the next agent's prompt, then say "send" and "open" plus the next app. A keyboard forces those steps into a series; voice overlaps them.
Do I need the cloud add-on for prompting? No. The $99 base product is built for prompting: raw, fast, on-device, offline-capable. The $10 per month add-on (7-day trial) adds LLM-polished output and more languages for human-facing writing.
How much does Infina cost, and is there a trial? $99 one-time as of July 2026, no subscription, every 1.x update included. No free trial; instead there is a 7-day no-questions-asked refund, so you can test it against your real workload risk-free.
The bottom line
Agents removed the typing of code from your critical path. The typing of prompts is what remains, and it is the slowest thing left in your loop.
Speaking removes most of it, and Infina removes the rest: the hotkeys, the Enter presses, the window switching. Say "type", say "send", say "open Cursor", and keep three agents fed from a couple of feet away.
Prompt faster, ship faster is not a slogan, it is the arithmetic of your own day. $99 once as of July 2026, and if a week of real use does not prove it, the refund is one email.