From Green Screens to AI Agents

In 2025, global hotel revenue exceeded $800 billion. Behind every booking — every confirmation email, every "your room is ready" notification — lies a labyrinth of technology, intermediaries, and protocols that has been accumulating for over six decades.

This is the story of how that labyrinth was built, layer by layer, from teletype terminals to AI agents. It's a story of failed deals that spawned empires, of XML schemas that became industry religions, and of a $133 million acquisition that created a $170 billion company.

It's also a story about what comes next. Because for the first time since the internet disrupted travel in 1996, a technology shift — agentic AI — threatens to disintermediate the disintermediators.

Before the internet, before websites, before even personal computers — there were terminals. Green phosphor screens connected to mainframes, operated by trained travel agents who spoke in cryptic commands.

The story begins in 1946, when American Airlines realized it was drowning in paper. Booking a flight required a physical card for every seat, updated by hand. The airline partnered with IBM on a research project that would take 14 years to bear fruit — but when SABRE went live in 1960, it was revolutionary.

SABRE could process a reservation in seconds. It knew every available seat on every flight. And it gave the agent using the terminal enormous power: whoever controlled the terminal controlled the booking.

This led to the CRS Wars of the 1970s and 80s. United built Apollo. TWA built PARS. The airlines realized that if your CRS was on the agent's desk, your flights showed up first. It was distribution as competitive weapon.

But hotels had a problem. The GDS systems spoke different languages — Sabre, Amadeus, Galileo, Worldspan each had proprietary formats. A hotel chain that wanted to be bookable through all four had to maintain four separate integrations. In December 1988, sixteen of the world's largest hotel chains each contributed $100,000 to fund a solution: THISCO (The Hotel Industry Switch Company). They recruited John F. Davis III, a Texas entrepreneur, to build the "UltraSwitch" — a computer that translated hotel reservation data into each GDS's proprietary format in real time. By the mid-1990s, THISCO's switch handled 70% of all electronic hotel reservations. Davis also launched TravelWeb in 1994 — the first website to book hotel rooms in real time — and merged everything into Pegasus Solutions, which IPO'd in 1997 for $53.4 million.

This architecture — hotel → PMS → CRS → switch → GDS → terminal → agent → traveler — defined the industry for decades. Every layer took a cut. Every layer added complexity. And every layer would eventually be disrupted.

The revolution started with a failed deal. In 1994, Sabre — the original GDS — sat down with Microsoft to discuss building an online travel agency together. Both wanted control. Neither would compromise. The talks collapsed.

Microsoft built Expedia. Sabre built Travelocity. And a race began that would reshape a $500 billion industry.

The timing was perfect. In 1995, Delta Airlines capped travel agent commissions at $50 per ticket — the first signal that airlines wanted out of the agent-commission model. Other airlines followed. Suddenly, travel agents faced an existential threat, and consumers needed a new way to book.

The OTAs filled the gap. Expedia launched in 1996 with Worldspan GDS connectivity. Travelocity launched the same year, powered by Sabre's own inventory. Priceline arrived in 1997 with Jay Walker's audacious "Name Your Own Price" model.

But the most important launch of this era happened quietly in Amsterdam. In 1996, a small company called Bookings.nl began listing Dutch hotels online. Two years later, Active Hotels in the UK pioneered direct connectivity to hotel property management systems, bypassing the expensive GDS infrastructure entirely.

Priceline — then struggling with its consumer brand in the US — spotted both companies. In 2004, they acquired Active Hotels for $161M. In 2005, they acquired Bookings B.V. for ~$133M and merged it with Active Hotels to create Booking.com.

That $133 million bet on a Dutch startup would become the most profitable acquisition in travel history, creating a company now worth over $170 billion.

By 2005, the online travel landscape was a mess of competing brands, overlapping services, and aggressive M&A. Over the next fifteen years, it would consolidate into one of the tightest duopolies in tech.

On one side: Booking Holdings (originally Priceline Group), which grew through strategic European acquisitions — Booking.com, Agoda, Kayak, OpenTable. Their model was commission-based: hotels pay 15–25% per booking, but Booking.com drives massive demand.

On the other side: Expedia Group, which grew through US-centric acquisitions — Hotels.com, Hotwire, Travelocity, Orbitz, Trivago, HomeAway/Vrbo. Their model was hybrid: merchant (buy wholesale, sell retail) and commission.

The 2015 watershed deserves special attention. In a single year, Expedia acquired both Travelocity (from Sabre, ~$280M) and Orbitz (from Travelport, ~$1.6B). Both sellers were GDS companies, and both sales were admissions of defeat: the GDSs had tried to play in the consumer space and lost.

Meanwhile, a new layer of technology was emerging: channel managers. Companies like SiteMinder, D-EDGE, and RateGain built platforms that let hotels push rates and availability to multiple OTAs simultaneously. The channel manager replaced the switch as the hub of distribution — and in 2018, RateGain literally acquired DHISCO (the successor to THISCO/Pegasus), folding the switch into a channel management feature.

By 2020, the architecture was clear: Hotel → PMS → Channel Manager → [Booking.com, Expedia, Airbnb, Google, GDS]. 74% of US hotel bookings flowed through this stack. Hotels paid 15–25% of revenue in OTA commissions. And the technology that was supposed to liberate hotels from intermediaries had created the most powerful intermediaries the industry had ever seen.

In November 2024, Anthropic quietly released something called the Model Context Protocol (MCP). In the noisy landscape of AI announcements, it was easy to miss. But for anyone who understood hotel distribution, MCP was a bombshell.

MCP created a universal standard for AI agents to discover and interact with external tools and data sources. For hotels, this meant something specific and radical: a property could publish an MCP server exposing functions like getAvailability(), getRates(), and createBooking(). Any AI agent — Claude, ChatGPT, Gemini, Perplexity — could then search, compare, and book hotel rooms directly. No OTA intermediary. No 15–25% commission. No loss of guest data.

By mid-2025, the dominoes were falling fast. OpenAI, Microsoft, and Google all adopted MCP. Governance moved to the Linux Foundation. Perplexity AI launched native hotel booking through Selfbook, covering ~140,000 hotels at zero commission. Google announced AP2 (Agent Payments Protocol) for secure AI-to-merchant payments. Even the OTA giants pivoted: Booking.com and Expedia launched as ChatGPT app partners.

And then in January 2026, Google announced UCP — the Universal Commerce Protocol — at the National Retail Federation conference. UCP aimed to standardize how AI agents discover, evaluate, and purchase any product or service, including hotel rooms.

The parallels to 1996 are striking. Just as the internet gave travelers a way to book without calling a travel agent, AI agents are giving travelers a way to book without visiting an OTA website. Just as OTAs disintermediated travel agents, AI agents may disintermediate OTAs.

But there's a crucial difference. The OTA era replaced one set of intermediaries with another. The agentic era has the potential to remove the intermediary layer entirely — connecting hotels directly with travelers through standardized protocols.

The hotel technology stack was designed for a world where distribution meant pushing inventory to intermediaries. The PMS managed rooms. The CRS managed rates. The channel manager pushed both to OTAs. And the OTAs owned the guest relationship.

But if AI agents can discover and book hotels directly — through MCP servers, llms.txt files, and Schema.org markup — then the entire stack collapses into something much simpler: a property operating system that serves as the single source of truth. And if all else fails? The AI agent can just call the front desk. Voice APIs like Whisper and Twilio mean an agent can literally pick up the phone, speak to the reservation desk in natural language, and book the room the old-fashioned way — except the "caller" is software running at scale. We spent sixty years building a digital distribution stack so complex it requires a switch, a GDS, a channel manager, and an OTA. Turns out an AI with a phone number can bypass all of it.

This is the ADAPT thesis: a unified set of open protocols that let properties natively speak the language of AI agents. Not a PMS bolted onto a channel manager bolted onto an OTA integration. Instead, a system where the property's availability, rates, and booking capabilities are published directly to the agentic web through open standards.

The implications cascade. Revenue management algorithms that respond to real-time agent demand signals. Guest profiles that stay with the property, not the OTA. Commission structures that drop from 15–25% to near zero. And a distribution model where the AI agent's job is to find the best option for the traveler — not the highest-commission option for the intermediary.

And the payment layer is evolving in parallel. Programmable commerce — where booking terms, cancellation policies, and settlement conditions are encoded directly into transactions — eliminates the reconciliation nightmares, virtual credit card complexity, and 30-day settlement delays that define today's OTA payment stack. When an AI agent books a room, programmable settlement clears payment instantly, with terms enforced by protocol rather than by trust in an intermediary.

The green screen gave way to the website. The website gave way to the app. The app is giving way to the agent. Each transition eliminated a layer of intermediation and put more power in the hands of the end user.

The next generation of property technology won't just manage buildings — it will make them natively discoverable, bookable, and settable by AI agents through open protocols and programmable commerce. That's not a feature. That's the foundation of a new architecture for hospitality.