Direct hotel bookings could command 66% of reservations by 2030, an AWS executive told the third annual Destination AI Summit in Washington, D.C., framing artificial intelligence as the lever that pulls volume away from online travel agencies.

"It's blowing up everyone's distribution strategies," said Michael Leidinger, SVP and CIO at Hilton, summarizing the panel consensus that AI-search and machine-readable hotel content will reroute travelers away from third-party channels. Greg Land, hospitality lead at Amazon Web Services, anchored the projection: direct at 66% versus 34% for OTAs by the decade's end.

What did the PMS debate settle?

The most contested panel pitted hoteliers against their software vendors over who controls daily operations. Richard Valtr, founder of Mews, and Adam Harris, CEO of Cloudbeds, pitched the property management system as a full "hotel operating system" — payments, labor, revenue management and business intelligence bundled into one stack.

Jacob Messina, CEO of Stayntouch, drew the loudest applause by flipping the frame. "These guys want to tell you how to run your hotel," he said. "We're going to give you the tools you need to be successful." The audience voted him the winner.

Valtr doubled down on integration. "Most systems today look like Mews," he said, drawing audible groans from peers. He pointed to the autonomous back-office model of AI Hospitality Group, founded by moderator Sloan Dean. "I believe what Sloan is building is the future. Yes, jobs will be reconfigured."

Where are hotels deploying AI today?

A women-leaders panel translated the vision into tasks operators can run this quarter. Monika Nerger, a technology advisor to global luxury brands, described pushing PMS, sales and POS data into Anthropic's Claude to generate revenue recommendations beyond legacy reports.

"AI allows us to radically change the way we're using data without much investment," Nerger said.

Nancy Wolff, SVP of technology at Columbia Hospitality, used documented SOPs and prompts to surface Accounts Payable errors before month-end close. "How did they do this? They just rewrote their SOPs," she said.

Page Petry of PD Petry Consulting tied AI to F&B economics:

• Flagging invoice changes against historical vendor pricing
• Informing portion adjustments to hit food-cost targets
• Mining guest reviews to surface signature menu items

The panel pressed for enterprise deployment on licensed, private environments with consistent data governance.

How are hotel jobs being redefined?

Back-office automation produced the summit's bluntest forecasts. Keryn McNamara, CIO at Aimbridge Hospitality, walked through LIFT, the company's internal Labor Insight & Forecasting Tool that blends labor data with AI-assisted scheduling.

"I have to think about things I can take off my properties, take tasks out of my GM's hands, and give them time back to focus on the guests," she said.

Matt Schwartz, CTO at Sage Hospitality, downplayed front-of-house spectacle. "It's not a robot greeting you, it's a robot walking through the halls while you're asleep changing filters," he said.

McNamara predicted the front desk and most back-office functions will disappear. Alex Cabañas, executive chairman of Pyramid Global Hospitality, extended the staffing logic: "We're going to be hiring lobby hosts, not GMs."

What do owners want from AI?

Owners framed saved labor hours as re-deployable inventory, not pure cost savings. Raymond Martz, co-president and CFO of Pebblebrook Hotel Trust, demanded faster reporting from operators.

"We want real-time data on the bookings. Before, when we got pace data from the management companies, it was already two weeks old," he said.

Carlos Rodriguez, CEO of Driftwood Capital, sharpened the point: "When you save hours, you increase inventory. But when you do nothing with it, when you don't free people up to do other, more impactful things, you just have increased inventory."

Owners positioned AI as the only path to real-time visibility into asset-level performance and to lifting NOI faster than revenue growth alone.

What are the open risks?

Conference-goers privately traded "P(doom)" scores — personal estimates of AI's probability of causing an existential catastrophe — even while publicly scheduling deployments. Leidinger flagged the abuse risk of proactive trip-care agents that rebook disrupted itineraries.

"How do we protect ourselves from abuse of that?" he asked.

With no operator claiming certainty about 2029, the summit's working assumption was that the next 24 months of pilot deployments will set the benchmarks the rest of the industry inherits.