Meta's artificial intelligence (AI) agent Muse launched Tuesday, giving users a range of capabilities, including travel booking.
Some flight booking is enabled by U.K.-based Duffel, which provides access to live travel systems through its API. Users ask Muse to plan a trip, with Duffel then presenting options, making bookings and managing any changes. While the integration currently offers flight bookings, accommodations will be offered in the future, Duffel CEO Steve Domin told PhocusWire.
Users can also use Muse and book with airlines directly. When asking Muse why a traveler would book through Duffel, the agent states that it "is usually the smoother route," but in the event someone wants to book through an airline website, Muse handles the booking in its browser and prompts the user to step in for logins or approvals. Muse can also add loyalty information on behalf of a passenger.
In the use case below, PhocusWire asked Muse to surface all nonstop options from Atlanta to London, with it pulling Duffel's inventory and options from airline sites.
When using the Duffel route, the booking is passed to Duffel.
"When the transaction is executed, the booking is confirmed by the user, it goes through our rail and is issued under our agency, and then we'll also provide all the post-booking support," Domin said, noting that in most cases, the suppliers are the merchant of record.
To add bags or seats after booking or make changes to details such as passenger names, travelers have to go through the airline, according to the Muse agent. Day-of changes also have to go through the airline.
When booking with Duffel, payment goes through the Stripe Link wallet setup, with users approving one payment and Stripe Link supplying a "one-time virtual card" to the airline.
Users can also opt to take over the screen when it's time to pay and enter their own card details on the airline's payment page. Muse notes that it can also work with Apple Pay, Google Pay, PayPal or Shop Pay when bookings are made directly through an airline website instead of through Duffel.
Duffel monitors flight status and notifies users about cancellations, offering self-service capabilities inside Muse. In the event something cannot be addressed through self-service, users are handed off to one of Duffel's agents, which are available 24/7, Domin said.
Fare rules and booking restrictions, however, fall to Meta, Domin said.
"This is all handled by the Meta team and the Muse product. They've put safeguards in place to make sure that at all times throughout the flow, the user is confirming what they are booking and has a full understanding of what they are booking," he said.
Muse is available for users in the U.S. on iOS, Android and muse.ai, and will soon be integrated with its AI glasses. Users are able to message the agent in a separate Muse app or via WhatsApp. Meta also noted that Muse is available for free with a usage limit. After reaching the free limit, users can upgrade to a paid subscription or wait for the free limit to refresh.
Actually innovative?
While Duffel was originally considered an aggregator and recognized for its direct connection to suppliers, it is attractive to companies like Meta that want to add travel but don't want to become an online travel agency, according to Domin.
"We now hold agencies in several countries, we do support, we negotiate deals with suppliers and settlement globally, and we do a ton of things that a travel agency would normally do."
Magpie founder and CEO Christian Watts said this is likely a good move for both Meta and Duffel.
"In some respects, you could argue that agentic commerce is supposed to work exactly like this: a deep integration with a specialist company that can support the transaction from search through booking and servicing, reducing the number of potentially problematic integrations the agent has to manage," he told PhocusWire.
"If executed well, this could be one of the first genuinely deep agentic-commerce products from a frontier AI company. The significant difference is that Muse isn’t merely recommending a flight or sending the customer elsewhere—it can potentially understand the request, compare live options, complete the booking and help manage it afterward."
According to Mario Gavira, CMO of Travelier and angel investor, Meta sets itself apart by offering something tangible for the everyday consumer in the early phase of agentic AI, but it's not necessarily anything novel.
"All personal AI agents will inherently manage all types of everyday digital tasks, so I don’t see Meta Muse having any edge over the other competitors. I do expect that strictly adhering to requested tasks, maintaining the right level of human oversight and efficient token consumption will separate the winners from the losers."
Ira Vouk, founder of AI Hospitality Alliance and Hospitality 2.0, agreed, noting that Muse isn't offering anything "revolutionary."
"Most of those functionalities are also available within frontier apps now as well; they're just not as easily accessible there," she said.
Where Meta has a real advantage is with its large customer base and its partnership with payment processing platform Stripe.
"Meta's partnership with Stripe to leverage single-use virtual cards is what I see as a smart move. This creates the right security barriers to allow agents to make high value transactions, which is of course critical in travel," Gavira said.
However, while Meta is offering this tool to its large customer base, it's failed to "win over users" with AI chatbots in the past, Gavira said.
"I think their poor privacy record makes it very difficult to trust them enough to let their agents act autonomously across our digital lives."
Further, it's not solving industry issues like connecting to availability rates and inventory for hotel booking, Vouk said.
"Nobody has solved for that yet but Google."
And while Watts pointed out that Google would probably like to align itself with one company and "go fully agentic," it also has to remain somewhat neutral thanks to its many partners.
Gavira, however, said other large players are likely to follow in Meta's footsteps.
"I have no doubt that Google and OpenAI will follow quickly; they already own the agentic foundations with Gemini and Codex," he said.
Watts added, "I'm sure they are all watching. The only problem is: What are they watching? The Meta LLM might finally be competitive in intelligence, but are any customers actually using it?"