Best AIPersonal Assistants
HomeResearch & guides

How to Use ChatGPT Agent for Errands (Now ChatGPT Work)

Published by 3Gen Internet Corporation, maker of OpenAssistant, Skej and AGNT. Comparisons use public documentation, without hands-on testing.

A simple errand route made from a paper path linking a shopping tote, appointment calendar, reservation card and checked list, with a compass guiding the sequence.

The steps below are taken from each vendor’s help pages. We have not run them.

If you came here looking for ChatGPT agent, start with the news: OpenAI’s help page for it now says ChatGPT agent “is no longer available.” Errands that need a web browser now run in ChatGPT Work, which can open its own cloud browser to click, fill in forms and sign in to sites for you. (OpenAI: ChatGPT agent, OpenAI: cloud browser)

This guide shows how to run an errand in ChatGPT Work, how to keep it on a short leash, and one alternative: Claude in Chrome from Anthropic. If you would rather text an assistant than open a browser, OpenAssistant (made by 3Gen, which publishes this site) says it browses websites on your behalf to book travel, make reservations and get things done online.

What “errands” means here

OpenAI’s own examples of cloud browser tasks include checking restaurant availability, asking businesses for quotes, finding flights, comparing local stock, tracking a package with details from a connected email account, finding a DMV appointment and preparing the booking for your approval, comparing utility plans, and saving apartment listings. (OpenAI)

OpenAI is also clear that it can’t finish everything. Some websites block automated browsers, some sign-in or payment steps aren’t supported, and you may have to take over or do the last step yourself. (OpenAI)

Before you start

Step 1: Decide which sites it may open

Do this once, before your first errand.

  1. Open Settings > Cloud browser.
  2. Pick a default for website access:
    • Always ask: you approve every new site yourself.
    • Auto approve: ChatGPT reviews each address and only stops to ask if a site looks unsafe.
    • Always allow: every site is allowed. OpenAI says this is not recommended.
  3. Optionally allow or block specific websites. Those choices override the default.

(OpenAI)

Our suggestion for a first week: Always ask. You’ll see every site it wants to visit, which teaches you fast how it works. Note that allowing a site is not the same as approving an action. ChatGPT still asks before things like confirming a booking or paying, whatever you pick here. (OpenAI)

Step 2: Start the errand in Work

  1. Open ChatGPT on the web or in the mobile app and select Work. On mobile, the switch between Chat and Work is a dropdown at the top of the screen. (OpenAI) OpenAI’s cloud browser steps cover web and mobile; in the desktop app, Work may use the app’s own built-in browser instead. (OpenAI)
  2. Describe the outcome you want. Name the website, give the details, and state your limits. (OpenAI)
  3. Approve website access if it asks.
  4. Watch progress in the conversation and answer any questions or confirmation requests.
  5. Check the result and the sources it used before you rely on it.

You don’t switch the cloud browser on yourself. ChatGPT decides when to use it, and if a connected app or plugin can do the job directly, it may use that instead. (OpenAI)

A prompt that gives it what it needs (our own example wording, not from OpenAI):

On the DMV site for my state, find the earliest license renewal appointment within 10 miles of 94110 on a weekday morning in the next three weeks. Prepare the booking but do not confirm it. Show me the date, time, address and anything I need to bring.

The task keeps running after you close the tab or put your phone away. It pauses when it needs you. (OpenAI)

Step 3: Sign in to a site the safe way

When the errand hits a sign-in page, ChatGPT pauses and asks you to sign in.

  1. Check the website address and the preview of the sign-in form it shows you. You can also inspect the live site before going on.
  2. Type your username and password into the secure sign-in form, plus any two-factor code.
  3. Let it continue.

OpenAI says credentials entered in that form go straight to the remote browser, the model can’t see them, and ChatGPT doesn’t store them. A separate review model checks each sign-in request for signs of phishing before showing it to you. The sign-in then persists for later tasks until it expires. (OpenAI)

Never paste a password, security code or card number into the chat itself. OpenAI says the same. (OpenAI)

The cloud browser is separate from your own browser. It doesn’t see your open tabs, history, saved passwords or existing sign-ins. (OpenAI)

Step 4: Approve, or take over

Step 5: Make it a repeat errand (optional)

Some errands are really “keep checking until something changes.” ChatGPT handles those as scheduled tasks.

(OpenAI)

Step 6: Clean up after sensitive errands

Open Settings > Cloud browser > Browser data and clear data for one site or for all of them. Clearing a site signs you out of it, so you’ll sign in again next time. (OpenAI)

The alternative: Claude in Chrome

Claude in Chrome is Anthropic’s browser extension. Instead of a remote browser, Claude works inside your own Google Chrome on your computer.

Who can use it: all paid Claude plans (Pro, Max, Team, Enterprise), on desktop Google Chrome only. Anthropic says it doesn’t support other Chromium browsers or mobile. It is in beta in the Chrome side panel. (Anthropic)

Install it:

  1. In Google Chrome, open the Claude listing in the Chrome Web Store and click Add to Chrome.
  2. Sign in with your Claude account when asked.
  3. Pin it: click the puzzle piece icon, then the thumbtack next to Claude.
  4. Grant the permissions it requests.
  5. Click the Claude icon in the toolbar to open the side panel.

(Anthropic)

Choose how much it does on its own. A drop-down on the chat input has three modes: Manually approve, Automatically approve, and Skip all approvals. In the newer side panel the default is Automatically approve. Even with sites always allowed, Claude still asks before downloading files, entering sensitive information or granting authorizations. You can review sites from the extension icon, three dots, Extension settings, Your approved sites. (Anthropic: permissions)

What it won’t do. This matters for errands. Anthropic lists actions Claude in Chrome refuses in every mode, including making purchases or financial transactions, creating accounts, and handling card or ID details. (Anthropic: permissions) So it can research, compare and fill in forms, but you complete any payment yourself. Anthropic also advises against using it on financial accounts, medical information or other people’s personal data. (Anthropic: built-in browser)

Repeat errands: you can schedule saved shortcuts from the clock icon in the upper right of the panel, to run daily, weekly, monthly or yearly. (Anthropic)

Which one to use

ChatGPT Work with cloud browser Claude in Chrome
Where the browser runs OpenAI’s remote computer Your own Chrome on your desktop
Keeps working when your computer is off Yes, per OpenAI No. Claude can only act in the browser while Chrome (and, for some tasks, Claude Desktop) is open
Phone Web and mobile Runs in desktop Chrome. On Max and Team (rolling out to Pro) you can start or steer a task from the Claude web or mobile app, but the browser work happens in Chrome on a connected computer
Sign-ins Secure form, sessions persist until they expire Uses your own browser
Can it pay or book Asks you to confirm bookings and payments Purchases are refused in every mode
Plans Paid, not Free or Go Paid (Pro, Max, Team, Enterprise)

(Sources: OpenAI, Anthropic, Anthropic permissions)

Pick ChatGPT Work if you want errands to run from your phone and finish while you’re away, including the booking step after you confirm it, where the site supports it. Pick Claude in Chrome if you’d rather watch the work happen in your own browser and are happy to make any payment yourself.

Check it yourself

This section is advice for you. We have not run these checks.

Run the same three errands in whichever tool you’re trying, and keep notes:

  1. Look-up errand: “Find three plumbers within 5 miles of my zip code with openings this week and list their quote request pages.” Good sign: all three links work and match the area.
  2. Form errand: “Fill in the quote request form on [site] with these details, then stop before submitting.” Good sign: it stops before submitting.
  3. Watch errand: “Check [product page] daily and tell me if the price drops below $X.” Good sign: a scheduled task appears and you get a notification when it runs.

FAQ

Is ChatGPT agent still available?

No. OpenAI’s help page says ChatGPT agent is no longer available. Errands that need a web browser now run in ChatGPT Work, which can open its own cloud browser.

Which plans include the cloud browser?

Paid ChatGPT plans in supported regions. It is not on Free or Go, and availability can vary during rollout.

Is it safe to sign in to a site?

OpenAI says credentials typed into the secure sign-in form go straight to the remote browser, the model cannot see them and ChatGPT does not store them. Never paste a password, security code or card number into the chat.

Will it pay or book without asking?

No. Before an action that is hard to undo, such as confirming a booking or making a payment, ChatGPT asks you in the chat.

Can I run a repeat errand?

Yes. Ask in the chat or open Scheduled to create, edit, pause or delete tasks. Monitoring tasks check for a change and notify you. Active task limits are 3 on Free and Go, 5 on Plus, 10 on Business and Edu, and 15 on Pro and Enterprise.

Try OpenAssistant

Prefer to text one assistant for errands and keep the browser out of it? Try OpenAssistant. It also handles your calendar, email and follow-ups, and you can see plans on the pricing page.

Sources