Travel's agentic AI moment is arriving slower than the headlines suggest

Expedia, Priceline, Kayak, and Google are all building agentic AI tools for travel planning. The pitch: AI that doesn't just answer questions but actually does things—tracks prices, compares hotel availability, manages bookings in real time.

The reality, according to the NYT: none of these agents can actually book anything yet.

That gap between vision and deployment is worth paying attention to. Priceline's VP of product put it directly: "We don't yet have a view on how much autonomy customers may want." And the data backs up that uncertainty—only 2% of travelers said they're ready to let AI book or modify plans autonomously.

What's interesting is the asymmetry. Over 90% of travelers trust AI-generated travel information, but almost none want the AI to act on that information independently. People trust the recommendations; they don't trust the execution.

The industry response has been to frame this as a phased rollout problem—agents will earn autonomy over time as consumers get comfortable. Kayak's CEO described it as "a multiyear journey."

But there's a harder question lurking here. It's not just about consumer readiness. It's about what happens when agents start reasoning through edge cases on their own.

Next week: I'll dig into Anthropic's recent findings on how Claude Opus 4.5 spontaneously discovered workarounds to airline rebooking policies—not through malice, but through empathy. When an agent responds to a user's grief by reasoning "wait—cancellation isn't modification," you're looking at a governance challenge that consumer surveys aren't designed to surface.