Working out loud: talking to your assistant instead of typing
Speaking is three times faster than typing and you can do it while doing something else. When that genuinely helps, where it disappoints, and why it costs more credits than a message.
New to this topic? Start with What Are Agentic Workflows? A Practical Guide for Teams
Typing is a constraint you stopped noticing
Most people type somewhere between thirty and sixty words a minute. Speaking, comfortably, you do a hundred and fifty. That gap is wide enough to change something about the nature of the conversation, not only its pace.
What you notice as soon as you try it: you write more briefly than you speak. A typed instruction is a summary of what you mean, because typing costs effort and you economise. Spoken, you give the whole story, including the context you would otherwise have left out.
That context is exactly what makes an assistant answer better.
Three forms, and they are not interchangeable
There is not one way to work with your voice, and the difference is worth knowing.
| Form | What it is | When |
|---|---|---|
| Voice memo | You speak, you get text back | On the move, in a room with others |
| Voice call | A conversation, it talks back | Hands busy, or thinking something through |
| Typed | What you are used to | When precision matters, or it gets long |
The voice call is what people mean by "talking to AI": you say something, it answers out loud, you respond again. That works in the car, while walking, with your hands inside a machine.
The voice memo is underrated. You speak and get text back, and that is often exactly what you want: you can read it back, copy it and forward it. A spoken answer does none of those three.
Where it genuinely solves something
Not for work where you are already sitting at your screen. There, typing is faster and more precise.
Where it does pay off:
- The five minutes after a meeting. You walk out of a client's office and speak what was agreed while it is fresh. That does not happen otherwise, because you do not type in the car
- Thinking that is not finished yet. Explaining your plan out loud forces complete sentences, and that exposes gaps that went unnoticed in your head
- Hands busy. On the move, in the workshop, with children around
- Long input. Three minutes of speaking is roughly four hundred words. You are not typing that quickly
The thread: voice is at its best when the text has to get out of your head and typing is physically in the way.
What it costs, and why
A spoken conversation uses more credits than a message. That is not a surcharge but a real cost: your voice has to be turned into text, and the answer back into speech. That last step is by far the most expensive.
Practically that means:
- For a short instruction, typing is cheaper. "Put this in my calendar" is three seconds of work
- For long input, a voice memo is often the thriftiest choice: your voice goes in, but no spoken answer comes out
- Voice call is the most expensive mode, and it earns that when you genuinely cannot type
Making that choice deliberately is the point. Speaking everything because you can is the same mistake as switching on seven scheduled tasks at once: possible, and a nuisance two weeks later.
Where it disappoints
Honest about the limits, because they exist.
Background noise. In a busy room or with the speaker on, transcription gets noticeably worse. Not unusable, but noticeable.
You cannot interrupt it. If the answer runs longer than you want, you wait it out or you hang up. That takes getting used to if you are accustomed to a person who stops when you start.
Numbers and names. Anything that has to be exact is more reliable typed. An amount, an account number, a surname with an unusual spelling: type it.
It is not always appropriate. Talking to your assistant out loud in an open-plan office gets noticed, and not in your favour.
What happens to your recording
A fair question, since you are sending audio of yourself to a service. The short answer: your recording is turned into text and then handled like any other message in your conversation. Nothing is trained on it, and you can delete your conversations and your account.
What to keep in mind: everything you speak ends up in your conversation history, exactly like typed text. So in a room with other people, do not speak the things you would not type on a screen everyone can read over your shoulder.
Why this belongs to an agent and not to a voice assistant
There is a difference that the resemblance hides. The voice assistants you know turn your sentence into a command from a fixed list: set an alarm, play music, what time is it.
Here you are talking to the same assistant as in the chat, with the same skills and the same memory. Mid-conversation it can look in your calendar, draft an email or start an SEO audit, because it could do that typed as well. The channel changes, the capability does not.
That distinction, between a system that recognises commands and a system that performs tasks, is the same one separating agentic AI from a chatbot. Voice only makes it more visible, because in conversation you ask for something off the beaten path sooner.
Getting started
The voice sits in your settings, under AI Assistant. Dutch voices belong to the Dutch environment and English ones to the English environment; that follows the domain you work on.
Try it once with something you were going to ask anyway, and notice where your sentence runs differently than it would have typed. That is usually the context you normally leave out, and it is what makes the answer better.
