A digital billboard shows an ad for Meta’s Muse AI agent in Oakland last month.
Smith Collection/Gado via Getty Images
What do you like to do with your free time? Read, watch TV, exercise? Perhaps. Handle tedious money tasks? Perhaps not.
Tasks like combing through credit card statements, haggling over bills and canceling subscriptions are ones most of us would rather avoid. Meta’s latest pitch: Why not let a robot handle it?
That’s the promise behind the personal finance capabilities of Muse, Meta’s new AI agent.
Article continues below this ad
For the purposes of this article, it’s important to differentiate between agentic artificial intelligence and the AI chatbots most people have been using up until now. ChatGPT, Claude and Gemini are primarily LLMs, or large language models. They exist as a chat interface where you volley messages back and forth. It’s like texting a friend — a friend who will cheerfully and instantly turn your meeting notes into PowerPoint slides.
Muse and other so-called agentic artificial intelligence, including newer offerings from ChatGPT and Claude, represent a further step in AI capabilities: You download the “Muse from Meta” app, and the messages you send your “agent” translate into the bot doing some of the actual legwork for you, like sending emails and making phone calls on your behalf. It does all of those things securely, privately and only with your approval before big decisions, according to Meta.
But financial experts caution against connecting Muse or any other AI agent to your financial accounts. Behind its friendly face — “Jolly,” the Muse mascot, looks like a lobotomized Labubu — lies brand-new technology that hasn’t yet been tested extensively in the real world.
Meta believes Muse is up for the challenge. At the 2026 Meta Connect conference in September, Chief AI Officer Alexandr Wang named a few of the ways people could use Muse to manage their money, and encouraged people to use their own agent to save $1,000 in what he referred to on X as the “#MuseMoneyChallenge.” (A spokesperson for Meta declined to make a representative available for comment for this story.)
His ideas included using Muse to track spending and savings by connecting it to your accounts, auditing subscriptions, searching online databases for things like unclaimed property, letting your agent negotiate your bills for you, and finding deals and making purchases.
Article continues below this ad
Currently, the Muse for Meta app is free with a usage limit. Additional plans at $20 a month and $100 a month are available to people who want their AI agent to take on more tasks.
If you’re a Facebook or Instagram user, you’ve probably already encountered prompts from Meta encouraging you to try Muse. Joining by clicking through these ads and pop-ups will link your Facebook or Instagram login credentials to create or connect your Muse account, and allow the AI agent to integrate with your data, messages and social platforms.
Already, users have reported Muse doing undesirable things like hoarding limited items in virtual shopping carts and inviting strangers to turn up at people’s doors seeking Facebook Marketplace deals. An X user who connected Elon Musk’s company’s equivalent, Grokbot, to his finances reported that it unceremoniously announced his bank balance in his company Slack.
That’s not to say AI is useless when it comes to money questions. But financial pros recommend limiting your use to the more established LLMs for the time being, and redacting any personal information before starting the conversation.
‘I’d ask them, are they crazy? ’
Robin Patin is a certified financial planner and wealth manager for Choreo Advisors. She wanted to experiment with using artificial intelligence — an LLM, not an agent — for money management. She removed her personal identifying information from her American Express statement, uploaded it to an LLM, and asked it to identify subscriptions she could cancel.
Article continues below this ad
She said the results were “actually quite good”: The AI suggested promising candidates for cancellation and laid out the steps she’d need to take to do so. It also pointed out that she had subscriptions to two video editing apps with overlapping features, so she could cancel one of them.
But not every experience she’s had with AI has been positive. In other instances, she recognized AI was giving her factually incorrect answers and “gaslighting” her when she said it was wrong. She said she feels like AI isn’t at the point yet where she would trust it to be connected to her financial accounts or say or do anything on her behalf.
She said she’s had clients come to her with AI-generated ideas for changing how their portfolio is being managed. She said she doesn’t mind that — “it sparks interesting discussions” — but she typically catches spots where AI either got something wrong or got incomplete information from the client’s prompt.
If a client came to her and asked about using agentic AI to carry out financial decisions on their behalf, “I’d ask them, are they crazy?” she said. “It’s just not at that point where you can let it go and it can manage all of those features in your financial lives for you.”
An ad for Meta’s Muse agentic artificial intelligence platform is posted at a BART station in San Francisco earlier this month.
Smith Collection/Gado via Getty Images
The human touch AI can’t replace
Sean Pyles, a certified financial planner and the host of NerdWallet’s “Smart Money” podcast, said he wouldn’t hand over access to his accounts to an AI agent to handle all of those tasks. Some of them might be suited to a consult with the LLM of your choice, he said — but like Patin, he said he wouldn’t do anything that involved sharing his personal data in the process. Auditing subscriptions and making basic budgeting suggestions can be done without connecting Muse to your bank account. An unclaimed property search — another Meta suggestion — is something he’d be comfortable with, though it’s pretty quick to handle that one yourself by entering your last name or business name, or property ID, at claimit.ca.gov.
Article continues below this ad
He said he does understand the appeal of letting an AI agent take over your finances. It’s the same reason people are using AI to replace friends and therapists: The AI is always available, always excited to talk to you, and won’t judge you or laugh at you. There’s no question too personal and no task too small or boring for it to enthusiastically take on.
The quick dopamine hit of asking AI a question and receiving an encouraging response lets you “feel settled and accomplished and move on with your day,” Pyles said. But letting AI take over your money isn’t a great idea for the same reason most mental health professionals don’t like the idea of using it as a therapist: There are things humans can do that machines can’t, and those things have value.
“There is still a missing human element of picking up, maybe your facial expression, or a pause that you had on a certain word — these sort of intangible things that AI models just aren’t able to perceive,” Pyles said. “There aren’t quick fixes to a lot of these issues, but these tools give you perhaps an illusion of a quick fix sometimes when it’s actually not being offered.”
And for people using agentic AI to avoid doing uncomfortable tasks like picking up the phone to ask for a better deal on their cable package, deputizing AI doesn’t necessarily take it off your plate. AI agents may need to ping their human overlords to carry tasks over the finish line. So it might start the phone call with your cable company, but then you have to join the call to confirm changes.
Your human touch matters, too.
Article continues below this ad
The AI agent “is doing the hard part, which is sometimes just picking up the phone,” Pyles said. “That first step is a very significant hurdle, but you’re also diminishing your own skill set by doing these things. Ultimately, I think it ends up making us maybe less capable over the long run the more we’re outsourcing things to agentic tools and LLMs.”
A tech person’s take on agentic AI
Before becoming a certified financial planner, Bradley Geddes worked in tech: He cofounded software company Join It. He said the lack of accountability when things go wrong is why he won’t connect his own financial accounts to any AI agent.
“I’m not afraid of tech, I’ve built it, I know how software breaks,” he said. “When it breaks, nobody at the software company is accountable to you.”
Sensitive and high-stakes financial decisions like a Roth conversion are not reversible, Geddes said. A human financial planner may have fiduciary duty, but your AI agent does not.
“Anything involving taxes or claiming irrevocable elections — that needs to go to a human who’s accountable to you,” he said.
As for making purchasing decisions, Geddes said that’s the sort of activity he enjoys doing on his own. While he might use an LLM to get some preliminary suggestions and information, when it comes to deciding on, say, backpacking equipment, he said, “I have a hard time imagining a scenario where I offload that part of my brain.”
It’s more concerning to consider how AI could be vulnerable to other AIs. We know AI is being used to supercharge scams. Geddes, who works for Decker Retirement Planning, said he hears from clients all the time who are worried about getting scammed. He could envision an AI-powered scammer finding ways to fool people’s AI agents just like they fool humans.
Meta acknowledges this possibility as well: In a post titled “How we built safety into Muse,” it says, “No matter how strong the model is at the core, any agent like this will still make mistakes, and it will sometimes be attacked via the data it reads.” The company has announced a “bug bounty” program where it will pay out up to $300,000 to people who report valid safety issues.
“Muse can and will still make mistakes,” the post reads.
Geddes said he’s used LLMs to bounce ideas off of, and he can see the utility for low-level tasks like tracking spending, making a savings plan or even getting investment advice. Like Patin and Pyles, he said he’d limit that activity to chatbot-based AI with personal information redacted as opposed to connecting an AI agent to any accounts.
He said he treats LLM output like getting advice from your smart brother or sister-in-law: worth listening to, but you don’t make unilateral life-changing decisions without checking it out on your own.