2026, the year travel AI went into production
In the agentic booking timeline I followed the eighteen months in which AI assistants learned to search, compare and start booking travel. That story was told from the traveller's side. This one comes from the other side of the desk, and it is the one almost nobody has laid out in order: between March and July 2026, the industry's infrastructure changed phase.
The line that captures it best came from Evantia Giumba, managing director of Amadeus France, on 2 July 2026, presenting Amadeus Max — a set of assistants for travel professionals able to resolve complex queries in natural language, analyse fare rules and understand a booking's history: what separates 2026 from 2025 is the transition to large-scale production; the challenge is no longer to experiment, but to industrialise safely.
Five moves from those months explain what that means in practice. None of them is spectacular on its own. Together, they are a change of era.
1. AI stops being a paid extra
On 16 June 2026, at HITEC, Oracle presented OPERA Cloud Assistant: AI embedded in the workflows of the world's most widely deployed hotel PMS. Intelligent room assignment cross-referencing reservation, preferences and history; rate descriptions generated automatically so they say the same thing on every channel; operational guidance in natural language for the employee who can't remember how to pull a report; built-in translation for 230 countries and territories.
The detail that matters is none of those functions: it is that it arrives at no additional cost for OPERA Cloud customers. Wyndham, with more than 2,100 properties running on that system, appeared as the anchor client.
When the basic AI layer comes included in the software you already pay for, the market for standalone tools doing exactly the same thing evaporates. For a small business that was weighing up buying a separate assistant, the first question of 2026 is no longer “which one do I buy?” but “what does the system I already have bring with it?”.
2. From the closed agent to the agent builder
In March 2026, Canary Technologies launched its Hospitality AI Agent Studio: instead of selling you a finished agent, it sells you the workshop to build one. Templates for front desk, concierge or central reservations, or from scratch, for internal use or guest-facing.
It is the same move enterprise software has been making for two decades — from the closed product to low-code — and its arrival in hospitality means one concrete thing: the vendor has accepted that every hotel wants its agent done its own way and that there is no single template that fits everyone. Read properly, it is good news for the hotelier with judgement and a trap for the one without: a tool that lets you build anything also lets you build a disaster.
3. Data first, agents later
In April 2026, Minor Hotels — parent company of NH, so this is felt in Spain — announced it was building from scratch a global platform unifying guest, marketing and service data across more than 63 countries, with Google Cloud, Salesforce, OneTrust and Deloitte, and with full rollout planned within the same year.
Look at what they put first: not an agent or a chatbot, but the data platform the agents will run on later. They announced plumbing. It is the same doctrine the serious vendors in the Spanish market keep repeating, and the one that holds up the uncomfortable argument I already set out in the data gap: most hotel AI projects don't fail because of the model, they fail because underneath there are three different rate sheets and none of them is the right one.
The small-business version of that lesson doesn't cost millions: make your rates, policies and services say the same thing on your website, in your booking engine, on the platforms and on your Google listing. That is exactly the same project at another scale.
4. The agent that shows where every number comes from
On 9 June 2026, Lighthouse presented Ernest, an autonomous AI companion for commercial teams — revenue, sales, marketing, distribution — built on a data footprint of 80,000 hotels and more than 500 integrations. It does three things: it answers in natural language with competitive analysis, it models revenue scenarios based on the property's own decision history, and it executes tasks inside the stack under user-defined guardrails.
But the feature that deserves an article of its own is a different one: every recommendation links back to the source data in the PMS or the CRS. Full traceability.
That is not a technical indulgence. It is the condition for a management committee to let a language model touch prices. The diagnosis that came with the launch — that teams aren't short of dashboards or systems, but are slowed down by scattered data and repetitive analysis — describes a lot of people's daily reality with uncomfortable precision.
5. The adoption figures, held with tongs
A global Canary Technologies survey from March 2026, with more than 400 technology decision-makers from hotels, chains and management companies, produced three round numbers: 71% believe AI already has a significant or transformative impact on the industry, 82% expect to expand their use of it over the following twelve months, and 85% plan to devote at least 5% of their technology budget to AI tools in 2026.
Held with tongs, because Canary sells AI agents. And with the obligatory counterpoint: when you ask about data infrastructure instead of intention, the picture collapses: in Otelier's 2026 Hotel Operations Index, only 15% of hoteliers are very confident in the accuracy and timeliness of their operational data, and only 25% say they are ready to adopt AI. Between “we're going to expand our use” and “we can expand it” there is a chasm, and that chasm is the real work of the next two years.
The reading
The case with numbers that is still the benchmark is Wyndham's: on its July 2026 earnings call it said its AI guest engagement platform, Wyndham Connect, is now in more than 5,000 hotels, and its AI Concierge in around 1,500. It is set out in full, along with the other cases that survive a critical reading, in the hotel AI cases with ROI you can believe.
2026 also brought new results published by listed companies, and they all point to the same place: customer service. eDreams ODIGEO said on 22 April 2026 that its AI voice system now handles 90% of inbound phone enquiries in five languages, with 33% fewer transfers to a human agent and resolutions 15% faster. Airbnb, in its shareholder letter for the second quarter of 2026, put at nearly 45% the share of issues that start with its AI assistant and are resolved without a human agent, and at around 16% the year-on-year drop in its customer support cost per booking, partly thanks to that assistant.
What 2026 adds is not better demos: it is plumbing. Unified data, AI included in the system you already use, tools to build with instead of closed products, and traceability so that somebody dares to sign the decision off.
If you run a small travel business, the practical conclusion is almost disappointing in its simplicity: you are not behind for lacking agents. You are behind if your information is a mess. Tidying it up is boring, it makes no press release, and it is exactly what the big players with eight-figure budgets are doing.
