Over the past three years, artificial intelligence (AI) has brought dramatic changes to the travel industry. Content generation got dramatically better, AI capabilities expanded from simple conversational responses to long-running agentic operations.
The travel industry put the technology to work in earnest, gaining efficiency and opening new opportunities while raising hard questions about ethics, authenticity and privacy.
So, what's next? As we move into 2027, there are three themes I think will shape the next phase.
1. Provenance will become table stables
The amount of AI-generated content on the internet keeps rising, and so does consumer concern about misinformation.
Part of the problem is that AI content is getting harder to spot. A 2025 study in Communications of the ACM found that people could correctly separate AI-generated from human-authored writing only 51% of the time, no better than a coin toss. We've seen some dramatic improvements since then, which won't have moved that number in our favor.
That is why content provenance matters. Provenance is the ability to trace where a piece of digital content came from and what happened to it along the way, including which AI tools touched it. Until this year, those efforts were entirely voluntary. On visual content, Adobe, OpenAI, Google, Meta and others formed the Coalition for Content Provenance and Authenticity, or C2PA, which produced Content Credentials, a tamper-evident label carried inside the file itself that tells you who (or what) made it and how it’s been modified.
This August, that sort of disclosure went from voluntary to legally required. Article 50 of the EU AI Act now requires providers of generative AI to mark their outputs in a machine-readable format so they can be detected as AI-generated.
Anthropic took the headlines when it began watermarking Claude's text, and much of the reaction treated it as one company deciding to tag its customers' work. It was not. Anthropic, Google, Meta, Microsoft, Mistral and OpenAI ALL signed the same EU code of practice; Anthropic was simply first to explain how it would comply. Ironically, the technique they’re using, SynthID-Text, came out of Google DeepMind. Eventually every major model will carry a similar mark.
Importantly though, Article 50’s obligations don’t stop at the labs. They also include deployers of the technology in the EU market, like agencies and the travel brands they serve. Publishing AI-generated or AI-edited imagery or text will now require careful thought about how much a human was involved and what level of disclosure is required. We’ll all need to shift from proving my content isn’t real to proving it is.
2. Intelligence will move in-house
The last three years we’ve experienced almost exclusively proprietary intelligence that’s piped to us from massive data centers. In the coming months and years, that intelligence will be moving in with you.
Stanford and Together AI have been measuring what they call intelligence per watt, across a million real queries, 20 local models and eight hardware configurations. Models small enough to run on a good laptop, 20 billion active parameters or fewer, correctly answered 88.7% of single-turn chat and reasoning queries. The share a local model could serve rose from 23.2% in 2023 to 71.3% in 2025. We’re rapidly approaching the point where small, locally run AI models are good enough to do meaningful work.
Enterprises are already moving in this direction. In July, AT&T described processing 45 billion tokens a day and cutting costs up to 90% by routing each request to the cheapest model that could handle it. It also released OTel 2.0, a telecom-specific model built by post-training Google's openly available Gemma 4. Notice what AT&T did not do. It did not build a model. It took someone else's weights and taught them its business, and that’s the shape I think we’re likely to see moving forward.
Tools like Thinking Machines' Tinker already offer post-training on multiple open models without you touching a GPU, and local models fine-tuned to your business will be able to shoulder more and more of your organization’s needs.
3. AI becomes your operating system
For 30 years, digital work has looked remarkably the same. The operating system versions changed, the icons looked different, but the process looked pretty similar: Click the desktop, launch an app, save a document, attach it to an email. Soon, that pattern will feel like using a filing cabinet feels now.
Sitting down to work will mean simply having a conversation with your AI. It proactively monitors your communications and calendar, tells you what actions need to be taken and in what order. Instead of navigating folders and clicking apps, you’ll describe what you need and AI either executes it or builds exactly the interface you need to analyze, author or manage it.
Documents still exist, you just won’t need to spend your day inside them. For travel brands this changes where we need to put our focus.
When everyone’s personal AI produces complex interfaces on the fly, the asset worth maintaining is not your website, or your app, or your chatbot; it is the knowledge behind them, managed and maintained in a form a machine can read and an agent can keep current.
We’re already seeing the first components of this. Google recently published the Open Knowledge Format (OKF), a specification for holding knowledge as linked markdown files with structured metadata, versioned in Git and built for AI agents to read and maintain. If the idea catches on, your website, apps and agents might all be created by AI out of the same OKF knowledge base.
Final thoughts
Three years of AI Opener conversations have taught me that the destinations and brands making real progress with AI aren't necessarily the ones with the biggest budgets. They're more often the ones who were the most curious, who were willing to look honestly at the value they actually provide and who could accept an answer they didn't expect.
Each of these three themes will require these same traits. Provenance asks you to be honest about how your content really gets made. Bringing intelligence in house asks you to question an assumption about where it comes from that most of us haven't revisited in three years. And the hardest one asks whether the website you've spent a decade and a considerable budget perfecting is still the asset, or whether it was only ever a way of presenting something more valuable underneath.