Instinct's 50% travel mix is small in dollars — but it proves the agent-displacement thesis [Booking Holdings], [Expedia] can't ignore

On September 28, 2026, Instinct — an invite-only personal AI agent with 14 employees — disclosed that more than 50% of the transactions flowing through its platform are travel-related. Founder Noah Shinn, on a podcast released the same day, put annualized transaction volume at $1B or more and said the platform is growing at roughly 10% per day. Reuters reported hours earlier that Instinct had closed a $1B Series C led by Sequoia, Benchmark and Coatue at a $10B valuation — a four-fold mark-up from a $2.5B round less than a month earlier.

The instinct-level math is unflattering for Booking Holdings and Expedia only at the margin: Instinct's $500M-plus in implied travel GMV is a rounding error against [Booking Holdings]' $51B in Q2 gross bookings. The number that matters is the 50% concentration. Travel is over-represented in agent transactions by roughly 3x its share of consumer spending — confirming the structural argument that travel is the wedge category where delegated AI buying will arrive first.

Booking Holdings and Expedia are not losing meaningful volume to Instinct yet. Booking Holdings's AI-driven referrals stayed under 1% of room nights in Q2 2026 even as agent platforms multiplied, per the company's August 4 earnings call. Expedia lifted full-year gross bookings guidance to $129.5B-$130.8B (8-9% growth) and reported Q2 gross bookings up 12% to $33.9B, with B2B bookings up 21% versus B2C up 8% — the B2B channel is the agency's defense, not the consumer AI surface.

The compression vector is take rate, not volume. Both OTAs already showed stress on the agency side in Q1 2026: Booking Holdings's merchant revenues jumped 27% YoY while agency revenues fell 2%, per HospitalityNet — a forced mix shift as the company leans into inventory ownership. If agents like Instinct extract even 100 basis points of take-rate concession from OTAs that want to remain visible inside AI surfaces, Expedia's $4.3B quarterly revenue base and Booking Holdings's $7.35B base both face a structural margin drag of 7-10% before any volume is lost.

The supply side has been waiting for agent compression of OTAs for a decade. Marriott International and Hilton Worldwide have both flagged AI agents as a potential accelerator of direct booking in 2026 10-Ks — language that did not appear in earlier filings. Hilton Worldwide launched its AI Planner on hilton.com in March 2026; Marriott International put Ask Bonvoy in front of 283 million Bonvoy members in beta this year. Both are founding partners in Alphabet's AI Mode hotel booking tool, which launched in the US on August 27, 2026 alongside Booking Holdings, Expedia, Trip.com and Choice Hotels.

The economic logic is straightforward: OTAs charge hotels 15-25% commission on every room night. A direct booking through an agent that connects to the hotel's own inventory saves the supplier that commission while preserving the agent's consumer relationship. Marriott International's Q2 2026 revenue of $7.07B and Hilton Worldwide's $3.34B already absorb some of that commission cost; if even 5 percentage points of OTA mix shifts to direct over the next 24 months, the implied operating income lift for the two chains combined is roughly $1B annually at current mix — paid for almost entirely by OTA margin.

Two tech platforms launched travel-relevant agent surfaces within weeks of each other. Alphabet added hotel booking inside AI Mode in the US on August 27, 2026, with Booking Holdings, Expedia, Marriott International, Hilton Worldwide, Trip.com, Choice, IHG, Priceline and Wyndham as launch partners — but the hotel or OTA remains merchant of record and Google positions itself as a transaction facilitator, not an OTA. Meta Platforms launched its Muse AI agent on September 9, 2026, and the September 22 follow-on commentary triggered the steepest single-day selloff in travel stocks since 2022: Booking Holdings fell 3.8-3.9% and Expedia fell 3.7% on the day, per AIWeekly and The Motley Fool.

The structural risk that emerged in Alphabet's launch documentation is that the platform shares only the minimum data needed to close a booking — name, payment, occupancy, property — and withholds inspiration/decision context even in aggregate. That means Booking Holdings and Expedia get the transaction but lose the customer journey they used to monetize through advertising, alternative-accommodation cross-sell and loyalty enrollment. The OTAs become a back-end inventory layer whose pricing power deteriorates as the agent layer consolidates around Alphabet and Meta Platforms.

Booking Holdings is down roughly 18.6% in September 2026 — its worst month since June 2022 — yet its Q2 earnings showed room nights up 5%, gross bookings up 9% and adjusted EPS up 15% to $2.54. The disconnect is sentiment, not fundamentals. BTIG reiterated Buy on September 24 with a $250 price target on the view that AI-agent fears are overdone; Morgan Stanley initiated Overweight with a $230 target on September 17 calling AI 'a new acquisition channel.' The risk into Booking Holdings's Q3 print (expected late October) is not the quarter itself but whether Q3 referral data moves materially off the <1% base.

For Expedia, the Q2 print was unambiguously strong: gross bookings +12%, revenue +14%, adjusted EBITDA +23%, GAAP net income up 166%. The stock has nonetheless tracked Booking Holdings on AI fears, with the September 22 Meta-Muse selloff hitting Expedia harder than the underlying numbers warrant. Expedia's full-year guidance lift to $129.5-130.8B bookings implies the company sees its own B2B channel (already +21% in Q2) as the right defense — letting third parties including AI agents call into Expedia inventory, rather than fighting for direct consumer mindshare.

The bull case for Booking Holdings and Expedia rests on three defensive moats that the Q2 results already show operating: (1) the merchant model shift, which gives Booking Holdings inventory ownership and 20-30% gross take versus 10-15% on agency; (2) the B2B pivot at Expedia, which lets third-party AI agents connect into Expedia's supply — making Expedia the rails rather than the storefront; and (3) the Connected Trip at Booking Holdings, which already represents low-double-digit percent of transactions and bundles ancillary services that an agent cannot easily replicate. Airbnb is the structural outlier — its 60% AI-coded engineering base (May 2026 disclosure) and direct-to-host model mean it has always been less dependent on agent intermediation.

The bear case is structural. Instinct's 50% travel share is not the number that matters — its 10% daily growth rate is. Even at 1% daily, a $1B annual run-rate compounds to $30B within 18 months; Instinct's stated growth is ten times that. By Q4 2027, the question is not whether AI agents carry meaningful travel GMV but whether Booking Holdings and Expedia have been forced to share20-30% of their take rate with the new agent platforms to remain visible — the same dynamic Google extracted from publishers in the 2010s. The early signal will be Booking Holdings's Q1 2027 disclosure of agency revenue mix (currently ~40%) and Expedia's B2B take rate progression.