2026 Marks the Start of the Travel Agentic Commerce Transition

In 20 seconds

If you’re heading to ITB Berlin, the headline is not “AI in travel”, we have had that conversation for years. The shift is that agents are starting to sit in front of travel commerce, and incumbents are now responding in ways that tell you what they believe the real battle is: execution, trust, and data, not chat interfaces.

​ITB’s agenda shows an industry moving from curiosity to consequence, with sessions focused on how AI reshapes business models, distribution, marketing, and the shopping and purchasing habits of travellers, including explicit attention on agentic AI. At Trusted Agents, we believe this is a signal that 2026 is the start of a 3–5 year transition rather than a single product cycle.

Those session themes also point to the next practical reality: to ride the wave, you need enterprise readiness in place before the propositions fully settle, meaning data foundations, security posture, and an agentic mindset that assumes customers will arrive via delegated software, with expectations around trusted identity and portable context, not just better marketing personalisation.

Do you block and throttle agents, entertain them via a controlled endpoint, build your own agent to defend the customer relationship, or prepare for agent-to-agent workflows, and whichever path you choose, are you ready to enforce delegation limits, verify who is acting on whose behalf, and safely use customer context without turning privacy and infosec into afterthoughts?

Enjoy ITB. If you want to leave Berlin with a clear “what we do next” plan rather than a notebook full of ideas, we run a 48–72 hour executive briefing burst during and immediately after ITB. It turns what you heard into a practical posture: what to pilot, what to govern, and what your teams need to build first. Email us at Shift@TheTrustedAgents.ai.

What we’re tracking this week

Incumbents are moving from “AI features” to “agentic rails” (trust + payment + execution)

• ​Sabre / PayPal / Mindtrip partnership​
• ​Amadeus acquires SkyLink​
• ​Expedia hiring for agentic AI, “not just shipping wrappers”

Bottom-up AI booking agents are exploding, but enterprise-grade safety is lagging

• ​DirectBooker on MCP as a new integration layer
• ​DIY travel agent workflow template (Telegram + APIs)​

Three questions to ask vendors at ITB

1. Connectivity: “Show me the exact way you expose availability, price, restrictions, and servicing actions to agents. Is it a documented tool interface, or are you still relying on screen scraping?”
2. Security: “Where do you enforce delegation limits and step-up approval, and what evidence do I get when something goes wrong?”
3. Customer experience: “If the customer arrives as an agent, what is the ‘happy path’ from intent to booking to change or refund, and where does it break today?”

ITB is the moment travel stops talking about AI and starts rebuilding for agents

What happened

ITB Berlin’s 2026 program is explicit: there is now an AI Track and Travel Tech programming that treats AI as a structural force in distribution, marketing, and commercial execution, not a novelty layer.

At the same time, the large incumbents are showing their hand, Mindtrip have announced a partnership to launch what they describe as an end-to-end agentic travel experience, with Sabre providing the travel transaction layer and PayPal bringing wallet, identity signals, and checkout. They say the product launches in Q2 2026, starting with flights and adding hotels later.

​Amadeus has acquired SkyLink, an AI-powered corporate booking solution designed to live inside enterprise chat tools (Teams/Slack style), and has signalled an intent to extend conversational capabilities beyond corporate travel into airlines, airports, and hospitality over time.

​Expedia, meanwhile, is hiring senior AI talent with a message that matters: they are “reimagining how travel works using the next generation of AI orchestration and reasoning”, which is a very different posture from bolting a chat UI onto a legacy funnel.

And for context on the demand side, Adobe has already measured a sharp rise in generative-AI-driven traffic to U.S. travel sites, up 3,500% year-over-year in July 2025.

What it signals in the agentic stack

Travel has always been a hard domain because inventory is dynamic and fragmented, and the intermediaries are not cosmetic. They are the stitching that makes the market legible. When agents become the front door, the stack shifts from “search and browse” to “request, evaluate, execute, service”. That makes three layers decisive:

• Cognition: models that can interpret intent and constraints.
• Execution rails: connectivity, booking, payment, servicing, refunds.
• Infosec rails: identity, delegation limits, policy, auditability, human-in-the-loop for high-risk actions.

This is why “enterprise assurance at agent speed” is not a slogan. It is the operating requirement. Read more here Five New Attack Vectors for Agentic AI​

What changes for enterprises

If you are an airline, hotel group, OTA, TMC, or a supplier-side platform, the roadmap gets more concrete:

1. Make offers machine-readable (availability, price, restrictions, ancillaries, policies) because agents cannot reliably reverse-engineer your commerce from HTML.
2. Treat connectivity as a product, not plumbing. In an agent world, connectivity determines inclusion in the decision set.
3. Rebuild servicing for agent loops (change, cancel, refund, disruption), because agents will compress time and increase volume, for good and for fraud.
4. Decide your posture: block, controlled endpoint, DIY agent, or A2A readiness, and communicate it internally so teams stop improvising.

Where it can go wrong

Agents amplify both conversion and abuse. If an autonomous system can search, price, book, change, and refund in a tight loop, attackers will try to exploit that loop faster than your traditional controls can react. And even without malice, the “confidence gap” risk is real: customers delegate before you have shared definitions, governance, and evidence-grade logs.

Practical next step

Before ITB conversations turn into abstract futurism, pick one journey (shopping to booking, or change/refund) and map the controls you would need if the “user” was an agent: identity, delegation, rate limits, step-up approvals, and audit trails. Then ask every vendor how they support those controls.

Read more

• ​Slow and Steady Wins the Race and You Still Need Speed​
• ​ITB Berlin Convention program​
• ​Sabre / PayPal / Mindtrip partnership​
• ​Amadeus / SkyLink acquisition​

People to meet at ITB

​Peter Marriott at Dataiera: Enterprise-grade infrastructure for agent-to-agent flight booking, designed to complete search to offer to payment, without every airline building a bespoke stack.

​Carl-Johan Holmen at The Visit Group: A practical view of the “connected trip” from ferry to hotel to entertainment, which is where agentic experiences either become seamless or fall apart.

​Martijn van der Voort at Murfee AI: A governance-first approach to connecting tools and safely automating complex travel workflows, including disruption handling and payments.

What's Underneath this Week's Headlines? Two stories that sharpen the lesson.

AI booking agents are multiplying fast, because the cost of building one just collapsed

​DirectBooker make a simple point that every hotelier should sit with: “AI is limited in its ability to answer traveler questions when information accuracy or detail is critical.” That is why the fight is moving toward structured, cached, real-time data and “agent-ready” endpoints, not prettier chat.

Now layer in the builder tools. You can buy or copy a travel planning agent workflow for the price of a team lunch. The n8n travel agent template shows the pattern: a chat surface (Telegram), a model, a search API, then stitched steps that look like a “planner”.

Then comes the inevitable startup spray. In a one-minute search you can find a long list of consumer-facing AI travel apps:
​searchspot.ai, nilgai.ai, selfe.ai, travelai.com, imean.ai, usevacay.com, roamaround.app, wonderplan.ai that perfectly illustrate how travel booking will be.

Most of these will not become enterprise-grade. That is normal. Early markets attract hundreds or thousands of experiments, because distribution is still cheap and the rules are still blurry. The question for ITB is not “which app wins”. It is: where will enterprise-grade execution come from, and how will suppliers avoid becoming invisible if agents prefer the cleanest, most reliable endpoints?

The agentic stack is filling out, but “safe enough for enterprise” is still the gating factor

If you want a simple map for the conversations you are about to have in Berlin, this is the one I am using:

1. Models are getting better and cheaper - Reasoning and tool integration continue to improve, and the industry is shifting from “bigger pre-training” toward more inference-time scaling and better agentic scaffolding. That matters because it makes agent behaviour more consistent and less expensive at runtime. Read more at: The State Of LLMs 2025: Progress, Problems, and Predictions​
2. Protocols: the rails for travel - In travel, “agentic” only becomes commercially real when protocols take you beyond discovery into booking, payment, servicing, disruption handling, and post-trip changes. Think of them as rails: they standardise how an agent connects to systems, how it hands work to other agents, how it checks out safely, and how it carries identity, authority, and customer context without turning everything into screen-scraping and brittle hacks. At Trusted Agents we offer a 30 minute briefing module on Agentic Protocols.
   • Tooling and system access - MCP and WebMCP turn airline, hotel, and TMC systems into structured tools an agent can call (availability, pricing, rules, changes) with predictable inputs and outputs. Their gap is important for travel: they are still weak at carrying user identity, loyalty, and intent across systems, which is why so many “AI travel assistants” punch out to legacy flows when it matters.
   • Coordination:A2A is the missing piece for end-to-end trips because travel is multi-party. A personal agent can hand off a task to an airline’s servicing agent, a hotel’s on-property agent, or a corporate travel policy agent, and exchange artifacts like forms, receipts, and approvals.
   • Commerce and payments:UCP, ACP, AP2, and AgentPay are different approaches to making checkout and settlement agent-safe, with auditable mandates, merchant controls, and human-present approvals. This is where refunds, chargebacks, fraud loops, and “who authorised this” get solved properly.
   • Trust and personal data:TAP, MyTerms, HCP, and Solid are about proving who the agent represents, what terms apply, and how preferences and permissions travel with the customer across the trip.
3. Security, resilience, and governance are becoming first-class requirements
   • The uncomfortable truth is that agentic capability is arriving faster than safe deployment patterns. This is not theoretical. WIRED reports that major firms have restricted OpenClaw over security fears, and Kaspersky describe a week of disclosed vulnerabilities, malicious skills, and secrets leaked from Moltbook, which they call “Reddit for bots”.
   • You can go even deeper on the “what to do about it” side. Experienced CISO Jason Rebholz at Evoke Security has a pragmatic four-step approach that starts with inventory and ends with control, which is exactly the sequence most enterprises skip when they are in a hurry.
   • ​A final anecdote to keep you honest about fraud and trust: a Malaysian couple travelled hours to visit a tourist attraction after watching an AI-generated video, only to discover it was fictional. It is a clean illustration of what happens when synthetic content, confidence, and human behaviour collide.

This is the key ITB point: agentic travel is moving bottom-up, but enterprise adoption will move at the speed of trust. The vendors who can explain their control plane clearly, with evidence, will win serious accounts.

Trusted Agents ITB Executive Briefing

ITB is loud, but the underlying decisions are practical, and we help you separate signal from noise and leave Berlin with a defensible “what we do next” plan.

At Trusted Agents we offer a 2-hour executive briefing on Agentic Commerce covering how to build an agentic mindset inside your organization, the technologies and protocols, and the customer experience considerations. Email us at Shift@thetrustedagents.ai.