What Muse Is
Muse is an AI agent from Meta (META), sold in the app store like any other app. It works like a chatbot that can do tasks and connect to the internet. Over two weeks it was linked to bank accounts, emails, and a calendar, then let to run daily tasks. It ordered groceries, scheduled meetings, and made a podcast with two AI hosts. Some features worked well, some did not, some would never get used - but the overall verdict was that it ranked as one of the most useful AI products yet tried.
What Worked
Most chatbots, however smart, are just tools you talk to. Muse acts. It has a phone call feature that dials on your behalf using two set voices, Haley (female) and Brett (male). In one case it called a dental insurance company, said who the customer was, gave the member ID, passed a security question, waited on hold for 15 minutes, then handed the call to the operator - saving time.
Fed all the bank statements, it read through them, put them into spreadsheets, and found duplicate subscriptions and unwanted charges it could then cancel, saving about $40 a month. Rocket Money and Claude were tried for the same jobs and could not do them.
Muse also booked restaurant reservations and got a movie ticket at a local independent theater, both cases where you just fill out a form online and pay.
Why Meta Can Pull This Off
Instagram, Facebook, and WhatsApp are used by over three billion people daily. All that Instagram and Facebook data connects to Muse. You can tell it to open a recipe saved in your Instagram reels and order the groceries for it to a chosen store. The trade-off with these AI products: the more the company knows about you, the more useful the agent can be. Meta already knows a lot about its users and is already tied into many parts of daily life. Whether people find that intrusive on privacy and data grounds is a separate problem. A comparison was drawn to Tencent (WeChat), which some think could try to roll out something similar using the same kind of user base.
Where It Failed
Many websites are not yet built for agents to operate. Getting a ticket at an AMC theater, which needs signing into a rewards account, gave it trouble. It can sign in and pull a code from your email, but it sometimes got tripped up or flagged as a bot, forcing a manual step - entering the password or fetching a confirmation code by hand.
Some sites have no interest in letting agents in. Amazon (AMZN) is one. Amazon was used in the written piece, and around the time the piece came out, Amazon decided it did not want Muse agents shopping and buying on the site - it wants humans going through the site - so it banned Muse. This points to a wider problem: companies like Amazon that run their own agents will not want a rival agent from Meta (META), Google (GOOGL), or OpenAI operating on their sites smoothly.
Reader Reaction
Readers split into two groups. One worries about privacy - already uneasy about Instagram and Facebook, they see this as one more way Meta collects and makes money from personal data, now more personal than ever because it reads your email and could see text messages.
The second group, a concern I share, objects to automating parts of life just because the technology can. In one test Muse went on Facebook Marketplace and haggled to buy bowls and plates from Heath Ceramics. Many readers asked why you would not just do that yourself, since it removes human contact. I largely agree. Now that the experiment is over, many of those tasks would be better done by hand.
On Privacy and Security
Handing over the data was a leap of faith and will be a huge obstacle. Many people will never be comfortable with it, and that is fair.
The banking side is more secure than people think because it runs through Stripe and Plaid, another financial services company. You give your information to those services rather than photographing a credit card and sending it to Muse to key in by hand.
A personal email was connected, but not a work email, because work mail holds sensitive material and a third-party agent should not reach it. This will be a major problem for Meta, both because of its privacy record and because of how willing people are to trade privacy for the convenience an agent offers.
Going Forward
Question asked: will Muse keep getting used now that the experiment is over? Answer: yes, for all customer service needs, plus paying parking tickets. Sitting on hold to reach another company's AI agent for customer service is a chore worth handing off, even if people naturally want to do things themselves and talk to real people.
