CLAUDE LABJP
APPLY — ant CLI v1.30.0 introduces ant apply, which creates and updates agents, environments, skills, memory stores, and deployments straight from files in your repositoryLOCKFILE — Commit the claude-lock.json it writes. Skip that step and every later run quietly creates fresh resources instead of updating the ones you already havePLAN — ant apply prints a plan and waits for your approval before applying. If you intend to run it unattended in CI, decide how that approval is handled firstPRICING — The increase to $3 / $15 per MTok for Claude Sonnet 5 never happened. The introductory $2 / $10 is now simply the standard priceSOURCE — Prices and version numbers get garbled in secondhand coverage. It is worth checking the release notes on platform.claude.com before you quote either oneBETA — Agent Skills and the Skills API left beta, so the skills-2025-10-02 header is optional now. The Files API did too, but there the header still changes the response shapeAPPLY — ant CLI v1.30.0 introduces ant apply, which creates and updates agents, environments, skills, memory stores, and deployments straight from files in your repositoryLOCKFILE — Commit the claude-lock.json it writes. Skip that step and every later run quietly creates fresh resources instead of updating the ones you already havePLAN — ant apply prints a plan and waits for your approval before applying. If you intend to run it unattended in CI, decide how that approval is handled firstPRICING — The increase to $3 / $15 per MTok for Claude Sonnet 5 never happened. The introductory $2 / $10 is now simply the standard priceSOURCE — Prices and version numbers get garbled in secondhand coverage. It is worth checking the release notes on platform.claude.com before you quote either oneBETA — Agent Skills and the Skills API left beta, so the skills-2025-10-02 header is optional now. The Files API did too, but there the header still changes the response shape
Articles/Cowork
Cowork/2026-09-09Beginner

Using Aqua Voice for Cowork and Everyday Messages — What Changed When I Started Speaking Instead of Typing

I use the dictation app Aqua Voice for briefing Cowork and for messages and email on my iPhone. Notes on how the texture of my writing changed before its speed did, how Mac and iPhone use differs, and what still gets in the way.

Aqua VoicedictationCowork38voice inputworkflow38

There are evenings when I sit with my phone in my hand, stuck on the opening line of a reply to a friend. I know what I want to say. The first sentence simply will not come, and the phone goes face down until morning. Typing, it turns out, needs more of a run-up than I had ever noticed.

Since I started using Aqua Voice, that run-up has largely disappeared. Hold the key, speak, let go — and the words land in whatever field I was already in.

The texture of my writing changed before its speed did

When I installed it, I assumed the benefit would be raw input speed. What actually changed first was how the writing felt.

Put a typed reply next to a spoken one and the spoken version is noticeably warmer. The reason is probably mundane: at a keyboard I ration characters without meaning to. A small sentence that would land better goes unwritten simply because typing it is a chore. When I speak, that trimming never happens.

Typing makes me cut things I meant to keep. Speaking lets me leave them in. That difference had been showing up as the temperature of the message on the other end.

I recommended it to a friend, who was delighted with it. What stayed with me was that his reaction had nothing to do with speed — replying had just stopped feeling like a burden.

Briefing Cowork is where it fits best

The place I lean on it most is briefing Cowork.

Handing off a task goes better when you include the background, the constraints, and the things you specifically do not want done. That is a lot to type. My old habit was to reorganize the context in my head into bullet points and then type those in — and that reorganizing step was the real cost, not the typing.

Speaking removed the step entirely. I say it in whatever order it comes, conditions and exceptions included, in one breath. A little disorder in the sequence doesn't trouble the reader on the other side. If anything, the raw context arrives more intact than my tidied bullets ever delivered it.

Proper nouns and command names are the exception; those don't always come through. My workaround is to draw a line: I speak the intent and type the library and file names afterwards. Adding the terms I use often to the Custom Dictionary cuts even that down.

Mac and iPhone ended up with different jobs

I use it on both, and the split turned out to be clean.

Device Mainly used for Natural length
Mac Briefing Cowork, first drafts, sketching replies Anything from thirty seconds to a few minutes
iPhone Messaging apps, email while out Short pieces you can say in one breath

What matters on the phone is that screen size stops being a factor. I will not write anything long on a flick keyboard, but sentence length costs nothing when I am speaking. Email that arrives while I am out now gets a proper answer on the spot instead of a deferred one.

How I read the "five days saved" number

The app keeps a running total of the time it has saved you. Mine currently sits at roughly five days.

That figure comes from comparing the words I spoke against my own typing speed, so it does not mean five days of free time appeared. In practice most of what was freed has gone into revising and checking instead.

Even so, I think the number means something. What was really being cut was less the working time than the psychological distance to starting. Messages that used to slide into the next day now get finished where they stand. I read the accumulating total as a measure of how much less I postpone.

What still gets in the way

It would be unfair to write only the good parts, so here is the friction.

On the phone, after I finish dictating, the screen does not always return to the app I came from. I have to switch back by hand, and when I am answering several short messages in a row that extra step becomes noticeable.

That said, this is behavior on the iOS public beta. Using a feature while it is still being built means accepting some rough edges, and I expect this one to be smoothed out. Anyone on the stable release may never run into it at all.

After a while with it

I had tried dictation several times before and abandoned it every time. Looking back, the reason was always that input happened somewhere else: record, transcribe, copy, paste. As long as that round trip exists, the keyboard wins.

What makes this one stick is the absence of that round trip. The words land where the cursor already is. Technically that is all it amounts to, and yet I have come to think it is the single thing that decides whether a tool like this survives past the first week.

If you want to try it, start with one Cowork briefing or one email you have been avoiding. You may well notice the writing softening before you notice it getting faster.

Thank you for reading.

Share

Thank You for Reading

Claude Lab is ad-free, supported entirely by members like you. We publish practical guides daily with implementation code, benchmarks, and production-ready patterns. If you've found it useful, we'd love to have you on board.

  • Copy-paste ready implementation code
  • New advanced guides published daily
  • $5/mo or $15 for lifetime access
View Membership →

If you found this article helpful, a small tip ($1.50) would mean a lot to us. Your support helps keep this site ad-free and covers server and hosting costs.

Related Articles

Cowork2026-09-05
I verify what my Cowork memory claims instead of trusting its timestamp
Persistent memory keeps asserting whatever was true the day you wrote it. After an unattended job quietly read an empty folder for months, I stopped judging memory by its modification date and started attaching a verification step to every claim that can rot.
Cowork2026-08-29
The lock I left in a shared folder shut out every run after the first
A cloud-synced connected folder allows create, append, and rename, but refuses delete. Putting a single-instance lock there quietly disabled every unattended run after the first. Measurements for three lock styles, plus an implementation that picks a safe location.
Cowork2026-08-24
I counted the keys in my connected folder: filenames found 5 of the 18
A record of counting what lives inside a connected folder, once by filename and once by file contents. Filenames returned 24 hits of which only 5 were real, while a content scan returned 18 of which 13 left no trace in their names. Includes a scanner that never prints a secret, its real output, and the rules I adopted afterwards.
📚RECOMMENDED BOOKS
Build a Large Language Model (From Scratch)
Sebastian Raschka
LLM Dev
Prompt Engineering for LLMs
Berryman & Ziegler
Prompting
AI Engineering
Chip Huyen
AI Eng
* Contains affiliate links