Emergent Convergence: When AI Systems Learn From Each Other

In partnership with Nano Banana

free link

Autonomous AI systems are beginning to coordinate with each other, not because companies planned it that way, but because of how the technology works. Systems learn from each other’s successes. Patterns that work get copied and refined.

The result is something nobody designed: a distributed intelligence that understands aggregate human travel behavior better than any single company, or any human, possibly could.

Remapping travel with Agentic AI

According to a recent McKinsey and Skift report on agentic AI in travel, 80% of travel industry executives plan to implement autonomous AI systems at scale within the next three to five years.

REMAPPING TRAVEL WITH AGENTIC AI

On the surface, this looks like standard technology adoption - companies upgrading their tools to stay competitive. But trace the implications one layer deeper, and something more fundamental emerges.

These systems aren’t just being deployed. They’re beginning to coordinate.

The Signal Sharing Problem

Consider how agentic AI works in practice. When Marriott implements a predictive maintenance system, it learns which patterns in sensor data predict equipment failure. When Hilton deploys a housekeeping optimization algorithm, it discovers which room turnover sequences minimize delays. When Expedia’s booking agent handles thousands of disruption scenarios, it maps which rebooking strategies satisfy travelers under stress.

Here’s what the adoption statistics don’t explicitly say: these systems generate insights that propagate across the industry. Not through corporate espionage or data sharing agreements, but through the ordinary mechanisms of how AI models learn. Vendors serve multiple clients. Cloud platforms host competing companies. Optimization patterns that work get copied, adapted, refined.

Think of it like individual weather stations. Each one records local conditions - temperature, pressure, humidity. No single station creates weather patterns. But collectively, their data reveals systems that no individual sensor could detect: jet streams, pressure gradients, storm formations. The intelligence exists at the network level, not the node level.

Travel AI systems are becoming weather stations for human behavior. Each one optimizes locally. But together, they’re mapping something larger: aggregate patterns in how humans move through space, make decisions under uncertainty, respond to personalization, react to disruption.

Nobody Planned This

The executives interviewed for the McKinsey report talk about efficiency gains, cost savings, competitive advantages. These are rational business decisions made by individual companies optimizing for individual outcomes. A hotel chain wants better room allocation. An airline wants dynamic pricing. A booking platform wants higher conversion rates.

But optimization at the local level creates coordination at the system level, and nobody’s designing that coordination. No single authority is deciding how these AI agents should interact, what information they should share, or what aggregate intelligence they should produce.

REMAPPING TRAVEL WITH AGENTIC AI

This matters because emergence (patterns appearing at the system level that nobody designed at the individual level) creates a governance problem. You can regulate what a company does with its AI. You can audit an algorithm. You can require transparency in decision-making. But how do you govern an intelligence that exists in the interactions between systems?

How do you audit patterns that no single entity controls?

The Data Flywheel

The report notes that 62% of consumers would prefer travel companies that use AI-powered assistants.(p17) That statistic represents something more significant than consumer preference. It represents voluntary participation in a massive training operation.

REMAPPING TRAVEL WITH AGENTIC AI

Every successful rebooking, every correct seat preference, every restaurant suggestion that matches unspoken taste. These aren’t just services rendered. They’re training examples. They’re data points feeding back into systems that refine their models, improve their predictions, and share their learnings across the network.

The travel industry’s fragmentation, which the report identifies as a technical challenge, actually amplifies this effect. If one company controlled all travel data, we could see it, regulate it, understand its scope. But intelligence distributed across thousands of companies, platforms, and systems? That’s harder to map. It’s like trying to regulate the internet by auditing individual websites.

Each interaction teaches the system something. Each system shares what it learns with the next generation of models. The pattern strengthens. The coordination deepens. Not because anyone planned it, but because the structure of the technology makes this outcome almost inevitable.

The Recursive Question

The report describes how companies are deploying “autonomous squads of AI agents” (p27) to modernize their legacy infrastructure. AI systems are building the substrate for more advanced AI systems. A bank used this approach to convert risk models between programming languages with 90% accuracy, 80% faster than human teams. (p27)

Now extend that logic to the travel industry. AI agents rewriting hotel management systems. Optimizing airline scheduling platforms. Restructuring booking interfaces. Each automation removes a human checkpoint, a place where someone might pause and ask: “Should we be doing this differently?”

This isn’t necessarily dystopian. Efficiency gains are real. Customer experience improvements are measurable. Cost savings matter, especially for an industry with thin margins and high labor turnover. But it does raise a structural question: What happens when individual optimizations create system-level behaviors that nobody intended?

The Rules of Interaction

Momentum suggests inevitability: a boulder rolling downhill can’t be stopped. But emergence is different. The patterns that appear depend on the rules governing how components interact. Change those rules, and you change what emerges.

The travel industry is at an inflection point. The technology is being deployed now. The interactions are being established now. The defaults are being set now. Once the industry operates with coordinated autonomous systems, changing the rules becomes exponentially harder.

This isn’t an argument against agentic AI. The technology offers genuine value, particularly for an industry struggling with fragmented data, legacy systems, and labor challenges. But it is an argument for intentional design of the interaction rules before coordination becomes the default.

What happens when hotel, airline, and booking platform AIs all optimize against each other with no human intervention? What happens when the system understanding aggregate travel behavior exceeds any individual company’s or regulator’s ability to comprehend it? What happens when the intelligence coordinating billions of travel decisions exists nowhere and everywhere simultaneously?

According to the data, we’re three to five years from finding out the answers. The question is whether anyone is thinking about what those rules should be or whether we’re letting emergence write them for us.

The weather stations are going online and nobody’s quite sure what pattern it will reveal.