Real Agent Cases

Find furniture on Facebook Marketplace for a new apartment with Meta's Muse, while Dots got the accounts locked (partial)

Posted
by Mary Randolph (Scientific American)
Last checked
Method:
How we verify

Evidence: Article or interviewA newsletter, journalist or interview account of what happened.

PartialThe source says it only partly worked, or had a significant problem a person had to deal with.

What went wrong: Muse found a dining table; logging Dots into Facebook got the poster's Meta accounts temporarily locked.

Outcome

Scientific American's Mary Randolph was moving apartments and tested Meta's Muse (named Mark) and OpenAI's Dots (named Sam) for a week. Muse logged in to their Facebook account, searched Marketplace for couches, lamps and rugs, and could draft and send messages to sellers with their approval, managing the replies while they did other things. They now own a Muse-sourced dining table. When they gave Dots their Facebook login to do the same, Meta flagged it as a bot and temporarily locked their Instagram, Facebook and WhatsApp. Later, Muse pushed them seven times in three days to join Target's membership program and buy a lamp on sale.

Agent

Muse, ChatGPT Dots

Prompt Recreated prompt

The source didn't share its exact prompt, so we wrote this from what it describes.

Before you start, ask me 1–3 short questions about anything important that's missing below. Keep them quick. Then do the task, checking in with me at the key points.

Goal: Find second-hand furniture for my new place and help me contact sellers.

What I can give you (ask for whatever's missing):
- What I need, with sizes and budget: [LIST]
- My area and how far I can collect: [DETAILS]
- Style I like (photos or words): [DETAILS]

Rules:
- Search the listings and shortlist the best few for each item, with price, condition, size and distance.
- Draft a short message to each seller, and show it to me before anything is sent.
- Don't sign in to accounts from more than one assistant or device at once, as sites may lock the account.
- Don't push memberships, sign-ups or items I didn't ask for.
- If a site asks me to sign in, pause and let me do it myself (including any CAPTCHA or two-step code); never guess or store passwords.
- Ask for my OK before you pay, book, send, submit or cancel anything.
- Check facts that can change (prices, opening hours, dates, availability, rules) against the official source, and tell me anything you couldn't verify.

Output: A shortlist table with links and draft seller messages.

You'll know it worked when: I pick up the furniture I need at a fair price.

Try this in:

This prompt has blanks to fill in, like [LIST]. We'll copy it instead of filling it in, so you can edit it before you send.

Opens in a new tab. We don't see what you send.

These links are only a shortcut. We're not affiliated with OpenAI, Google or xAI, and your results may not match the source.

Optional stronger version

Our rewrite with extra checks. The poster didn't test it.

Before you start, ask me 1–3 short questions about anything important that's missing below. Keep them quick. Then do the task, checking in with me at the key points.

Goal: Find second-hand furniture for my new place and help me contact sellers.

What I can give you (ask for whatever's missing):
- What I need, with sizes and budget: [LIST]
- My area and how far I can collect: [DETAILS]
- Style I like (photos or words): [DETAILS]

Rules:
- Search the listings and shortlist the best few for each item, with price, condition, size and distance.
- Draft a short message to each seller, and show it to me before anything is sent.
- Don't sign in to accounts from more than one assistant or device at once, as sites may lock the account.
- Don't push memberships, sign-ups or items I didn't ask for.
- If a site asks me to sign in, pause and let me do it myself (including any CAPTCHA or two-step code); never guess or store passwords.
- Ask for my OK before you pay, book, send, submit or cancel anything.
- Check facts that can change (prices, opening hours, dates, availability, rules) against the official source, and tell me anything you couldn't verify.
- Warn me about common resale scams, like deposits before viewing, and suggest safe pick-up arrangements.

Output: A shortlist table with links and draft seller messages.

You'll know it worked when: I pick up the furniture I need at a fair price.

Before you finish: re-read my goal and constraints, check your answer against each one, and list anything you assumed or couldn't confirm.

Source

https://www.scientificamerican.com/article/i-gave-metas-muse-and-openais-dots-my-passwords-for-a-week/

  • PCWorld's Ben Patterson, a Manhattan theatre lover who can't afford Broadway every night, used Muse to hunt for ticket bargains. Asked how much tickets to Verdi's Macbeth at the Met were, it first gave a general web answer, but when asked to check the official site it opened a browser and found a $71 seat (including a $10 service charge) for a weekday evening, logged in and teed it up for checkout. Patterson cancelled, then had it go through every remaining performance of a limited-run Off-Broadway play, finding which had seats and the best price, $92.41. For Hamilton in the next 30 days it found a rear-mezzanine seat for a Saturday matinee at $222 all-in. Patterson didn't buy this time.

    Muse · Shopping · Source date:

  • For PCMag, Ruben Circelli asked three agents to find J. Kenji López-Alt's no-stir risotto recipe, research the best version of each ingredient at the local Wegmans and add them to an Instacart cart. ChatGPT's Dots, once Instacart was connected, found the recipe, researched the ingredients and filled the cart without issue. Meta's Muse got there slowly, needed Circelli to log in and enter a verification code, and added saffron threads that aren't in the recipe. Gemini's Spark found the recipe and proposed ingredients but couldn't access the account, tried to take over Chrome and froze.

    ChatGPT Dots, Muse, Gemini Spark · Shopping · Source date:

  • Thy Toeung needed to review their car insurance before their son turned 16. They made a data sheet listing every driver, vehicle and the cover they wanted, and asked Claude, ChatGPT, Gemini and Muse to act as their insurance-shopping agent and get real quotes from six reputable insurers. Gemini's auto-browse explained it couldn't use the websites and handed back a comparison table to fill in. ChatGPT took 2 hours and 279 messages across 12 attempts and got no quotes; Toeung believes sites such as Allstate, Nationwide and Liberty Mutual block AI agents. Claude ran for 3.3 hours and got 5 real quotes. Muse, on its free tier, used several sub-agents and got 5 quotes in about 20 minutes. Toeung switched insurer and got about twice the cover for around 40% of the previous price.

    Claude, ChatGPT, Gemini, Muse · Shopping · Source date: