July 2026: Airline AI’s Real Battle Moves Below the Interface

Welcome back to the OAG Airline-Tech Innovation Radar, where each month we cut through the noise to spotlight three real-world launches that genuinely move the aviation industry forward.

Last month, in our edition AI Stops Talking and Starts Transacting, we watched conversational agents (finally) carry a traveller from inspiration to a paid ticket without leaving the chat. The interface, it seems, has grown up.

But the past few weeks told a quieter, and arguably more consequential, story. The most important airline-tech developments did not happen on the booking screen at all. They happened deeper in the stack, specifically, in the system that keeps aircraft flowing through the world's busiest airspace, in the fare-distribution layer that determines what an AI agent is even allowed to see, and inside the core technology estate of a major carrier.

As a company built on flight data and intelligence, we will admit a bias: we find the digital foundation is often more interesting than the façade. And following our recent critique of the industry’s “partial in, partial out” approach, one belief sits at the heart of our company thesis: there is no agentic future without a high-quality data backbone.

Innovation #1: Air Space Intelligence rebuilds how US air traffic flows

The US air traffic management system has long had a coordination problem disguised as a delay problem.

• Airlines, operators, and the FAA are constantly trying to balance demand against capacity, but that balance depends on how quickly everyone can see the same operational picture around scheduled demand, filed flight plans, aircraft positions, weather constraints, airport capacity, and airspace bottlenecks.
• When those signals are fragmented or arrive too late (for example, in the case of flight irregularities), the system becomes reactive. Constraints are managed only after they have already begun to affect the network, making the available interventions more disruptive than necessary.

That is the layer the FAA is now trying to modernise.

Two weeks ago, the FAA awarded Air Space Intelligence (ASI), the Boston-based aviation AI startup, a 12-year, $875 million contract to deploy two systems for the US National Airspace System: Flow Management Data and Services, or FMDS, and Strategic Management of Airspace, Routes, and Trajectories, better known as SMART.

The decision matters not just because of the contract size, but because ASI beat much larger names, including Palantir and Thales, for one of the most consequential air traffic modernisation programmes in the US.

Here’s how it works:

• FMDS will replace the FAA’s legacy Traffic Flow Management System and become the new technological backbone of the FAA Air Traffic Control System Command Center. It brings together flight plans, airline schedules, real-time position updates, and capacity constraints to estimate current and anticipated traffic flows across the National Airspace System.
  
• SMART sits inside FMDS as the predictive layer. It uses data on airline schedules, weather, airport capacity, airspace conditions, and operational constraints to identify likely congestion before it turns into delay, helping the FAA, airlines, and other operators coordinate schedules, routes, and trajectories before aircraft depart.

The system is also designed to support more localised rerouting around severe weather and improve the live data exchange between the FAA and airspace users. ASI says the technology builds on its Flyways AI platform, which already supports live operations with major US airlines and the US Air Force.

Why does this innovation stand out?

Two things make this collaboration more special than a typical government IT contract.

First, the scope. This is not a dashboard bolted onto the side of air traffic control. FMDS is intended to become the data backbone of the FAA Command Center, so the place where the national picture of demand, capacity, weather, schedules, and constraints is assembled and acted upon. In an industry where disruption often spreads because different actors are looking at different versions of operational reality, that shared picture is the product.

Second, the choice of vendor. A defining piece of public aviation infrastructure has gone to a software-first, startup-like aviation AI company whose systems were already operating in commercial and defense environments. That is a meaningful signal. Aviation modernisation is no longer only about replacing hardware, screens, and telecoms. Increasingly, it is about whether proven commercial software can move into the institutional core of the air transport system.

We have seen AI applied to air traffic control before, most notably when we covered Heathrow’s AIMEE system in December 2024. But the layer is different this time. AIMEE helps controllers interpret what is happening on and around one airport surface. FMDS and SMART aim to help coordinate traffic flow across an entire national airspace system.

The caveat is important. This is not AI taking over air traffic control. The FAA’s own material makes clear that controllers remain responsible for keeping aircraft safely separated. The harder questions are about validation, governance, and decision rights.

• How will SMART’s recommendations be tested?
• How will airlines know why a flight has been moved, slowed, or rerouted?
• And when demand exceeds capacity, who decides which flights absorb the pain?

The FAA is targeting initial SMART operations in fall 2026. Whether that date holds, and how transparent the safety and operational validation work becomes, will tell us more than the contract value alone.

There is also a neat piece of connective tissue worth noting. ASI is the same company Joby has chosen to help prepare US airspace for scaled eVTOL operations. That puts ASI in an unusual position where it builds predictive airspace intelligence for both today’s airline network and tomorrow’s air-taxi market.

Innovation #2: Amadeus tries to keep AI agents from melting airline pricing infrastructure

Last month, we showed agents are learning to buy. The logical next question is what happens when those agents learn to shop.

In early June, a user pointed Claude Code at Etihad Airways’ website and asked it to find a flight. The result was not one itinerary, or ten, or even a few hundred. The agent returned 881,076 fare options for a single trip, crawling dates, stopovers, and route combinations at a scale no human traveller would ever attempt manually.

That is the uncomfortable economics behind agentic travel. Human search has always had a natural brake where people get tired, bored, distracted, or satisfied with “good enough.” AI agents do not. They can run comparison shopping around the clock, across dates, routes, fare families, points of sale, connection options, and ancillary combinations, while the systems answering those requests absorb the cost.

This is exactly the pressure point our CCDO, Jeremy Burke, mapped in our recent Look-to-Book primer. Flight search has moved from a human clickstream to machine demand.

• In the early online era, the industry settled around roughly 100 to 200 searches per ticket sold.
• Today, we estimate the figure is closer to 10,000 to 20,000.
• In the agentic era, Amadeus and Sabre have both pointed to a possible future of up to 200,000 searches per ticket sold

Decius Valmorbida, president of travel at Amadeus, calls this the “infinite search” problem. His answer is not another chatbot, another booking widget, or another commercial agreement telling agents to behave better. The proposed fix is architectural: stop answering every agent query live. Amadeus says its systems already handle around 150,000 transactions per second, but even at that scale, the agentic search pattern forces a different design logic.

Here’s how it works:

Instead of sending every AI-generated shopping request on a live round trip to an airline pricing engine, Amadeus wants to precompute a large share of likely answers in advance.

• Common routes, date ranges, passenger combinations, and repeatable shopping patterns can be computed once, indexed, refreshed, and served quickly when an agent asks.
• This is the same basic “compute once, serve many times” logic that made the web scalable, but adapted to the far messier world of airline fares, where price, availability, rules, bundles, and personalisation all move constantly.
• Amadeus’ existing Instant Search product already uses precomputed booking results, cache refresh mechanisms, live availability checks, and sub-second responses.

The agentic shift is that this kind of infrastructure may now become central rather than optional. Amadeus has also been attacking the problem from the other side. Its Advanced Airline Profile filtering layer uses machine learning to block irrelevant or unactionable NDC search traffic before it reaches airline systems. Amadeus says early results reduced unproductive traffic by more than 70% in some cases, and Air France-KLM has publicly reported that the solution blocked the majority of irrelevant traffic to its system.