Aerial view of the Nairobi skyline from the KICC rooftop at golden hour · Lebu Ayiga · via Wikimedia Commons · CC BY-SA 4.0

The AI agent will book your room. It will still buy it from an OTA.

By Onyango George (OG)

Expedia announced this week that it was joining Meta's new travel agent. Its shares closed down almost 8 per cent the next day, and Booking, Airbnb and TripAdvisor fell with it. The market understood something East African operators should: in the agent era the commission does not disappear, only the shopfront does.

Expedia spent this week doing what a company does when it has decided which side of a technological shift it wants to be on. It announced that travellers would soon be able to plan and book hotels and trips directly inside Meta's new personal agent, Muse, without visiting an Expedia site at all. Expedia would remain the merchant of record. The transaction would stay its transaction. Only the window the traveller looks through would change.

The market read the announcement and sold. Expedia closed down almost 8 per cent on Wednesday. Booking Holdings fell 5 per cent to a session low of $155.37, within touching distance of its 52-week low, and is down roughly 29 per cent on the year. Airbnb fell 6 per cent. Meta rose. Measured from Muse's launch in the first week of September to Wednesday's close, the damage is heavier still: Expedia down 13.1 per cent, TripAdvisor 13.8 per cent, Airbnb 17.8 per cent, Booking 19.3 per cent, against Meta up 20.6 per cent.

For a lodge owner in Laikipia or on Zanzibar's east coast, the temptation is to read that as good news. The middleman is finally being disciplined. The commission is under threat at last. That reading is almost exactly backwards, and the reason is visible in the plumbing rather than the share price.

Look at what Muse actually does with a trip. For flights it runs on Duffel, the London booking infrastructure company, which since 9 September has given the agent live, bookable inventory across more than 500 airlines. That is a genuine pipe, direct to the carrier. "Travel is the ultimate stress test for an AI agent," said Steve Domin, Duffel's co-founder and chief executive. "It's dynamic, complex and deeply personal and when something goes wrong, the consequences are real."

For hotels there was no pipe at all. As Adriana Lee, Skift's travel technology reporter, established when she examined the launch, Muse shopped rooms by opening a browser and going to consumer websites, Expedia and Hotels.com among them, with no defined commercial relationship with the supplier and the agent rather than any travel integration deciding where the search went. Now Expedia is formalising that path. The arrangement makes official what the robot was already doing informally: rooms in East Africa are sold through an aggregator, and the aggregator keeps the customer.

So the asymmetry holds. The flight to Jomo Kenyatta gets cheaper to distribute. The room does not. What the agent removes is not the intermediary but the property's last sightline to the guest.

Alejandro Berdonces Corroto put the underlying problem plainly in an analysis for Hospitality Net in August, updated on 7 September as the Muse details firmed up: "no settled commission structure exists for when an AI agent executes the booking on the traveller's behalf." His reading of where that vacuum leads is the argument East African operators should be having this month and mostly are not. "OTAs are positioned to replay inside AI assistants the same move they made in search: absent specific regulation, independent hotels will spend years paying, again."

That is the mechanism, and this week supplied the proof. An agent that books through Expedia is not disintermediating Expedia. It is doing Expedia's customer acquisition at machine speed and at no cost, and the commission on the resulting booking is the commission that was always there. The only variable that has moved is the cost of being shopped.

Berdonces Corroto's sharpest line goes to what commission was ever for. It "was never really a fixed percentage so much as the price paid for not controlling the relationship with the end guest." By that definition an agentic booking is the most expensive kind a property can take. The lodge never sees the traveller, never sees the query, never learns which twelve properties it lost to, and pays the full rate for the privilege.

There is a second cost, less discussed and arriving sooner. Lee's reporting flags that Meta's surfaces reach 3.6 billion daily users and that an agent operating at that scale could sharply inflate look-to-book ratios and raise infrastructure costs for suppliers, noting the agent's habit of repeatedly re-searching when a budget is not met. Muse reached the top of the US App Store this week and took 2.5 million American downloads in a fortnight. For a 24-room camp whose channel manager prices by transaction volume, that is not an abstraction. It is a line item that grows while the booking count does not.

None of this lands in a vacuum, and the contrast with what Kenya has actually bought is instructive. On 29 May the Ministry of Tourism and Wildlife signed a partnership with Google built on four components: a Tourism Pulse data hub on Google Cloud tracking search trends and brand sentiment, a Gemini trip planner that builds itineraries from a traveller's description, a digital skilling programme for tourism SMEs, and targeted advertising in priority markets. Rebecca Miano, the Cabinet Secretary, framed the deal as positioning Kenya as Africa's leading AI-first tourism destination, against a target of roughly 5 million international visitors from 2.7 million in 2025.

Read that alongside this week and a gap opens. Kenya's deal sits almost entirely on the inspiration layer, the part of the funnel where a traveller decides whether to come. Muse and its competitors operate on the layer where the traveller decides who gets paid. Those are different layers, owned by different companies, and only the second one carries margin.

The gap is not theoretical, because the booking layer was already contested before Meta arrived. Google confirmed on 7 August that it was testing agentic hotel booking in the United States through Ask Maps, returning real prices and availability rather than ranges, with Amadeus as technical partner, and was explicit that it would not itself become the commercial operator of the transaction. Julie Farago, its vice-president of travel engineering and local search, had set out the ambition in November 2025 as letting users "describe what they're looking for, compare options, then refine their criteria". In every one of these designs the property is represented by somebody else's connectivity.

The first-order cost of that is already on the record. "I do believe that some of the changes that were made in the display at Google definitely put some pressure on SEO," Glenn Fogel, chief executive of Booking Holdings, told analysts on 4 August, attributing the weakness to AI Overviews. If the company spending billions a year on paid search is losing organic ground to AI summaries, the East African lodge relying on a WordPress site and a decade of blog posts lost it some time ago.

Not everyone believes the agent wins, and the sceptic's case is worth taking seriously even though it does not help. On Wednesday, as her own company's shares fell on its Muse announcement, Ariane Gorin, Expedia's chief executive, told Skift: "I certainly personally don't believe that a large portion of bookings are going to be an end-to-end conversational experience." Expedia has abandoned its own single unified agent for specialised point agents handling discrete parts of planning, and now integrates with Muse, Google's AI Mode, ChatGPT, Claude and Alexa, judging each case on its merits. Its AI tools have not yet lifted booking volume, though the company reports 60 per cent more traveller intent data, higher engagement and lower cancellation rates among travellers who book after using them.

Gorin is probably right about the near term. She is also, read from Nairobi, making the case for the OTA rather than against the agent. Her stated value proposition is fulfilment, customer service, payments, supplier connectivity, inventory breadth and bundling: precisely the reasons a booking should route through Expedia rather than to the hotel. Fogel has made the mirror argument, observing that the vast majority of Booking.com's room nights come from independent hotels while the ten largest global chains account for only a low double-digit percentage. East Africa's supply is overwhelmingly independent. When the platforms describe whom they protect, they are describing us, and the protection has always carried a price.

The demand, meanwhile, is not the problem. UN Tourism's barometer, published on 15 September, put global arrivals up 0.4 per cent in the first half of 2026, 690 million travellers, three million more than the same period last year. Africa grew 4 per cent, the strongest of any world region, with sub-Saharan Africa at 6 per cent. The visitors are coming. The contest is over who stands between them and the room.

Very little of the useful response involves buying software. Three things are worth doing while the commission question is still open. The first is to know which listings the agents are reading. Muse shops consumer OTA sites, not your booking engine, and Expedia's tie-up entrenches that. If your channel manager pushes clean, current rates and availability to the specific sites an agent browses, you are visible; if a room type is stale on one of them, you are invisible to a shopper you will never know existed. That is a connectivity audit, not a purchase, and it costs an afternoon.

The second is to start measuring look-to-book rather than conversion alone. If your channel or booking engine contract prices by transaction or by API call, agentic shopping transfers cost to you before it transfers revenue. Take the baseline this quarter, while the volumes are small and the rollout is still American. You cannot negotiate a term you have not measured.

The third is to treat your own site as machine-readable infrastructure rather than a brochure. When an agent does look direct, it reads structured data: rates, availability, location, policies, in a form software can parse. Most East African lodge sites carry none. It is cheap, it is durable, and unlike everything else in this story it sits entirely within an owner's control.

One detail deserves a closing moment. Muse is live in the United States only, on iOS, Android, muse.ai and WhatsApp. That last surface is the channel through which an enormous share of East African properties already take enquiries, confirm transfers and chase deposits. The agent Meta has built runs there. The rollout has not reached us, but the shopfront has been installed in our market for years, and Meta owns both ends of it.

The market's verdict on Wednesday looked like a judgement on middlemen. It was not. It was a judgement on who owns the guest, and Expedia had just told everyone the answer. For a camp in the Mara the practical translation is narrower and harder: a new intermediary has arrived, it works for the traveller, it buys from the platform you already pay, and it does not sleep.