TL;DR:
Answering mode explains and shows. Doing mode (an agent) clicks, types and changes things in your apps.
Explaining is low risk because nothing on your computer changes. Doing saves effort but needs more care.
Use answering when you want to understand, and doing when the steps are routine and you can check the result.
A tax-style form with twelve boxes is open on your screen, and you have no idea what half of them mean. You could ask an assistant to explain it, or you could ask it to fill it in. Those sound similar, but they are two very different kinds of help, and knowing which you are asking for changes how much you should trust it.
The short answer
Answering help reads what you show it and tells you something: an explanation, a summary, a drawing, a list of steps. Nothing on your computer changes. You stay in charge of every click.
Doing help, often called an agent, takes control of an app, with a cursor of its own, and performs the steps: it opens pages, fills fields, moves files. That is explain mode versus agent mode. The agent takes a goal, and you check what it did.
Neither is better. They answer different needs, and many jobs use both: understand first, then act.
What happens step by step
In explain mode
You ask a question, usually about something on your screen.
The assistant looks at the screen at that moment.
It replies in words, a picture or a worked example.
You decide what to do and do it yourself.
In agent mode
You describe the outcome you want.
The assistant plans steps and looks at the screen as it goes.
It clicks and types in the app, step by step.
It stops to ask before steps that cannot be undone, such as sending or paying.
You review what it did.
Walk-through: explaining a form vs filling it in
Take a form with a field labelled "Date of birth", another labelled "Tax ID", and a checkbox reading "I confirm the information is true".
In explain mode you ask, "What does Tax ID mean on this form?" The assistant looks at the field and the text around it, explains in plain words what the label usually refers to, and tells you where the number is normally found. Nothing was typed. If it got the explanation slightly wrong, the cost is a moment of confusion.
In agent mode you say, "Fill in this form from my details." Now the assistant has to read your details from somewhere, decide which value belongs in which box, and type it. If it puts the day and month in the wrong order, there is a wrong date in a real form. The last checkbox is a promise made in your name, so a careful agent leaves that one to you.
Notice the difference in what you need to check. After the explanation, you check whether it makes sense. After the filling, you must check all twelve fields against the source. That is the extra review effort of doing.
When each fits
Situation | Better fit | Why |
|---|---|---|
You do not understand something on screen | Answering | Understanding is the goal, nothing to change |
You are learning how to do it yourself | Answering | You need the steps in your head, not on a finished page |
The same routine steps, every week | Doing | Boring for you, easy to check |
A one-off, high-stakes submission | Answering, then you act | You want to stay in control of the final click |
Moving between several apps to gather facts | Doing, with a check | Time-consuming but usually reversible |
What it can't do
Explain mode cannot finish the job for you. It tells you what to do and you do it.
Agent mode cannot be trusted blindly. It can misread a screen, click the wrong thing, or misunderstand your goal.
Neither knows your private context unless you provide it. A form that needs your details needs them from you.
Neither makes decisions that are yours, such as whether to sign something.
Privacy and control
The rule of thumb is that more power needs more control. With answering, the main question is what the assistant sees on your screen. With doing, it also matters what it can change. Look for these: a clear way to stop it at once, a prompt before it sends a message or pays, and a promise that it never types your passwords or card numbers.
DuckyHelper does the answering part today: you press a key in any app, ask out loud, and it explains while it sees your screen and draws on it. Its agent features, where it works in your apps with its own cursor, are early access and are built to ask first before sending a message or paying for anything.
Try it yourself
Open any form you have not filled in, a sign-up page will do. First ask an assistant to explain two fields you find confusing. Notice that you learned something and nothing changed. Then, only if you want to, try the doing version on a harmless form, stop before the final button, and compare every field with what you meant.
Frequently asked questions
Is explain mode always safer?
It changes nothing on your computer, so yes in that sense. You can still be misled by a wrong explanation, so check anything important.
Can I switch between the two in one task?
Yes, and it is often the best way: understand with answers, then hand over the routine steps with a check at the end.
Get started
