How do I ask the assistant to use a browser?

Most of the time you do not ask at all. You describe the job, and the assistant works out for itself whether the job needs a browser. You never have to say the words "browser" or "computer".

The line it draws is between reading and working. If someone could answer your question by reading a page, you get a straight answer. If someone would have to work the site — click through it, type into it, filter, sort, sign in, page through results, fill something in — it opens a browser.

Requests that send it to a browser

These are the shapes that reliably get a browser, written the way you would actually say them:

• "Book a table for four on Friday at 7 — here is the restaurant's site."
• "Check where my order has got to on the courier's tracking page."
• "Find three flights from Manchester to Lisbon on the 14th and compare the prices."
• "Fill in the application form on this page with the details below."
• "Go through the invoices in my supplier portal and list anything still unpaid."
• "Download my latest statement — it is behind the login on my energy supplier's site."
• "Renew my parking permit on the council website."
• "That answer looks out of date — go and actually look at the page."

That last one matters more than it looks. If an ordinary search or page fetch comes back thin, stale, blocked, or the page clearly needed to load properly first, the assistant will go and open the site itself. You can also just tell it to.

What to put in the request

The more of the job you hand over up front, the less it has to stop and check with you. Worth including:

• What "done" looks like. A booking made, three options compared, a list of unpaid invoices — say which.
• The details it will need to type: names, dates, reference numbers, addresses, the options you want chosen.
• Whether it should go through with it. It stops before anything that would change something — buying, paying, submitting, sending — unless that is exactly what you asked for. "Fill this in" gets you a filled-in form and a question before anything is sent. "Fill this in and submit it" asks for the submitting too.

You will still do a few bits yourself. It will not type a password, a passcode, a security code or a card number, and it never attempts a "confirm you're human" check — so expect to step in at those points. It stops, leaves the browser open on the page, and asks you.

Pointing it at particular sites

If you only want it looking in one place, say so: "only look on the supplier portal". That restriction is enforced, not just a request it tries to honour. If the task would take it somewhere else — including a page the site itself redirects to, or a new tab — it stops rather than going there, tells you which sites it is allowed on and which one it stopped at, and invites you to add it.

If you would rather be explicit

Type /computer in the message box and choose Use a browser. It puts "Open a browser and " into the box for you, and you finish the sentence.

How to tell it has started

There is no approval card and no confirm step for an ordinary browser task. It opens and gets on with it. Here is the order things happen in:

1. The assistant replies straight away to tell you what it has set going.
2. The Computer panel slides in from the right-hand edge on its own. You do not have to open it.
3. The status label in the panel header reads "Starting", then "Working".
4. A card appears in the conversation, headed "Opening the browser" and then "Working in the browser", with a Watch button.
5. Once it is properly going, the status label and the card headline stop saying "Working" and start narrating the actual step instead.

The narration is plain English, and it is the quickest way to see what it is up to:

• "Opening" and the address it is going to, and "Reading the page".
• "Looking for" and what it is searching for, "Taking a look", "Looking closer".
• "Clicking", "Typing", "Filling in a field", "Pressing", "Hovering".
• "Scrolling", "Scrolling to it", "Waiting".
• "Opening a tab", "Switching tab", "Closing a tab", "Checking its tabs".

Under the headline, the card carries a running count of actions and the address the browser is currently on.

Note: signing in to a site is the one browser thing that always asks first. You get a confirmation card with Cancel and Approve, and nothing opens until you approve it.

While it is working, the chat waits

A browser task is not something you chat over the top of. While a run is going, the send button is replaced by a stop button and the message box reads "Working in the browser…". If you try to send anyway you will see "Still working in the browser — one moment."

You do not have to wait around, though. The task runs on its own, and the assistant comes back into the chat with the outcome as a new message when it is finished — you will not have to ask how it went. The message box goes back to normal once the run has finished.

Asking for a second browser task while one is running is refused for the same reason. The assistant tells you it is still working in the browser and to ask again once that has finished, and when you do it carries straight on in the same browser.

If it does not use a browser and you wanted it to

If your question has a plain answer, you get a plain answer — and that is usually quicker. If you would rather it went and looked at the live site, just say so. "Go and actually look at the page" is a request it acts on.