Google's Newest Travel Plays

Today on the AITB — at their I/O event, Google made its travel ambitions much bolder, but at the same time, a bit unclear.

Honestly? The whole event was kind of confusing. Two days in Mountain View. A wall of announcements. New models, new agents, new infrastructure, new consumer products — and the bottom-line, which is not easy to determine, is that Google is done positioning Gemini as an assistant. They're calling this the agentic era.

Here's what we got.

Gemini Omni. a new model that takes any input and creates video output grounded in real-world physics. Upload a video clip, describe what you want changed, the model handles it. Starts with video, but the vision is any output from any input. Cool, but not sure how well it performs against Seedance 2.0 and frankly, I won’t be testing that one.

Then Gemini Spark. A 24/7 personal AI agent that runs on its own cloud virtual machine, has its own Gmail address you can email tasks to, and keeps working in the background even when your laptop is closed. This kind of looks like Google's answer to Open Claw. Maybe. Honestly, we don't know yet - it's only rolling out to Ultra subscribers in beta next week.

And then Gemini 3.5 Flash, the new workhorse model. Google says it outperforms 3.1 Pro on coding and agentic benchmarks while running four times faster than rival frontier models. And it is fast, but I’m hearing it’s not a lot cheaper and the main concern these days is token cost, not speed.

And then there was the new Antigravity, AI mode, a blog post on GEO and SEO, and a bunch of other things, some of which aren’t even remotely related to travel, so I’ll skip them and focus on what matters for us.

Google made two moves at I/O, both very different, both targeting travel bookings. One is new, the other one not so much. And in all honesty, I have no idea how related the two are.

The one everyone's talking about is the Universal Cart. Google's new AI-powered shopping hub — persistent across Search, Gemini, YouTube, and Gmail — where Gemini actively monitors price drops, flags availability, and can complete purchases on your behalf. The payments infrastructure underneath it is called AP2, Agent Payments Protocol. It lets an AI agent pay for something within limits you set in advance, with a signed digital contract logging every step. Quick note here — AP2 is similar to ChatGPT's agent commerce protocol developed with Stripe, but I think Google's version has a much bigger chance of actually being deployed beyond the demo stage.

For travel, the key line from the announcement was that Google is expanding the Universal Cart to hotel bookings. That's not speculation, it's written black on white on Google's website. And who's doing the actual fulfillment inside that travel booking experience? OTAs. Actually, we already knew that. Google's blog post from November 2025 already named Booking Holdings and Expedia as agentic travel partners. No surprise there, as the platforms rank among Google's largest global advertisers.

I firmly believe that Google has no interest in becoming an OTA, but they do want to be deeply embedded in the travel booking journey. They've clarified they won't act as merchant of record. Their play is at the orchestration layer — they capture the intent, the agent executes, the infrastructure settles the payment. Booking and Expedia are suppliers. They bring the inventory and the merchant rails. Google wraps around them. Who pays the bill? Operators of course.

Is that a good long-term deal for the OTAs? Search Engine Journal points out that Universal Cart pulls more of the purchase journey inside Google-owned experiences, while historically Google Search sent users outward to retailer websites. Referral traffic to OTAs is the obvious casualty here, but the booking volume might compensate. We'll see.

Now. The second move. This one is quieter. It didn't come from the shopping blog - it came from the Chrome developer team. It's called WebMCP, and to be clear, it is something that Google announced back in February. They are now reframing it as part of Chrome's "agentic web" updates, and we are getting more information on what it actually can do.

Here's the short version. When an AI agent tries to interact with a website today — say your booking engine — it's essentially guessing. It reads raw HTML, takes screenshots, clicks around, tries to figure out what your search form does. WebMCP fixes that. It gives websites a standard way to tell agents: here are the functions I support, here's what they do, here's what they return, knock yourself out. One clean function call instead of dozens of guesses. Chrome's own documentation calls out travel as an explicit use case. It will, I quote, "Help agents book complex, multi-city and multi-passenger trips with fewer steps."

Let me clarify something. This is not the same as Anthropic's MCP, even though the names are close. Anthropic's MCP is a back-end protocol - it's how OpenAI or Claude connect to your systems server-to-server. WebMCP lives in the browser, with the user right there in the tab.

Google does not explicitly connect WebMCP and the Universal Cart, which raises a bunch of questions. Strategically, they fit together. Universal Cart is the consumer-facing commerce layer — the cart, checkout, and payment experience. WebMCP is one possible infrastructure layer that helps AI agents interact with websites cleanly enough to reach that checkout flow. But if WebMCP is able to do that on any website, OTAs will lose their advantage. Is that the reason WebMCP is so quietly rolled out and nobody is reporting on it? I have no idea, maybe I’m overthinking here, but it's definitely something I'll be following up on.

As for operators, the practical question is: is your booking engine agent-readable? Not just mobile-friendly - that's so 2016. Can an agent call your search, parse your availability, and complete a booking without guessing? Because that's what the infrastructure is being built to expect.

I'd look into implementing WebMCP compatibility, and for the operators out there - I suggest you do too. If it becomes standard, any agent — Claude, ChatGPT, Gemini — could use your website directly. Worth the shot. But don't expect it to automatically shift bookings to your direct channel. Look at who's named in every single Google travel announcement. Booking. Expedia. Every time. The pattern isn't changing — it's deepening. The interface is shifting from a search results page to an AI agent, but the fulfillment still runs through the same partners. Being agent-readable gets you on the field. It doesn't guarantee you get to play. 
That will be all for the AI Travel Brief for today. Thank you for listening or watching! Cheers!