The Brief
Everyday AI 5 min read

AI in the Sky: 180 Flights Rerouted, One Union Left Out

NAVION

Share

The United States Federal Aviation Administration has rolled out an AI-powered advisory system across the entire national airspace. The system is called SMART, it costs $875 million, and it is now active across all 22 air route traffic control centers in the country. Airline executives are supportive. The union representing more than 15,000 air traffic controllers says it was never consulted. That gap is worth understanding.

What SMART Actually Does, and What It Cannot Do

SMART is a cloud-based system. It pulls together data on weather patterns, flight paths, traffic flow, and controller staffing to predict how air traffic will behave, then generates recommendations: suggested routes, scheduling options, optimal arrival sequences for aircraft.

Here is what most coverage of this story glosses over. SMART does not control anything. It advises. A human team of FAA specialists based in Washington, DC, reviews every recommendation before it reaches air traffic control facilities. Local controllers then decide whether to accept or ignore the suggestion. The FAA has been explicit on this point: the system cannot replace the role of air traffic controllers and cannot take over control of an aircraft.

This is a meaningful architectural choice. The AI sits upstream of the decision, not inside it. It narrows the option space for human experts rather than acting autonomously. That design reflects both the complexity of the domain and the regulatory and safety constraints that govern it.

The early numbers from the initial deployment in the Washington, DC, airspace are specific. Since the system went live on September 21, it has helped reroute 180 flights. The FAA reports an average saving of 10 minutes per flight and nearly 80 gallons of fuel per flight. The agency then expanded coverage to Atlanta, Jacksonville, and Miami before announcing the full national rollout on November 10.

Airline Support, Controller Silence, and a Structural Problem

Delta Air Lines CEO Ed Bastian and American Airlines CEO Robert Isom both offered supportive statements when the initial DC deployment was announced. Their enthusiasm is understandable. Fewer delays and lower fuel consumption translate directly into operational savings and improved scheduling reliability for carriers.

The National Air Traffic Controllers Association offered something different. The union, which represents more than 15,000 members, told Flying magazine that it was not consulted or involved in the design, testing, or implementation of SMART, and therefore could not speak to its effect on the air traffic control workforce.

That is a significant statement, and it points to a pattern that appears repeatedly when AI systems are deployed in high-stakes professional environments. The people whose work the system is designed to support are often not part of the design process. This matters for reasons that go beyond labor relations. Air traffic controllers possess operational knowledge that is difficult to encode in a dataset: the informal heuristics, the edge cases, the situational awareness built over years of managing real traffic in real conditions. A system designed without that input may be technically sound and still miss things that practitioners would have flagged immediately.

The FAA’s example shared with The Dallas Morning News illustrates the system’s potential: SMART recommended a shorter, faster route for a cargo flight arriving at Dallas-Fort Worth International Airport. The agency did not identify the flight, and has not released comparable data on how the system has affected passenger traffic. The evidence base, at this stage, remains limited.

Why This Moment Matters Beyond Aviation

The SMART deployment is a case study in what it looks like when AI moves from pilot project to national infrastructure. The transition from a single regional test to 22 centers across the country happened in a matter of weeks. The speed is notable. So is the governance question it raises.

Aviation is one of the most safety-critical domains that exists. The consequences of errors are severe and often irreversible. That is precisely why the industry has developed layered human oversight structures over decades. SMART’s design respects those structures, at least formally: humans validate, humans decide, humans retain authority over the aircraft. But the question of who was involved in building the system, and whose expertise shaped its logic, is separate from the question of who retains formal authority over its outputs.

This is the broader pattern to watch. As AI advisory systems move into domains where the cost of error is high, the design process matters as much as the deployment architecture. A system that produces good recommendations on average can still fail in ways that only practitioners would recognize, if those practitioners were never asked.

The SMART rollout also illustrates something about how AI changes the nature of professional work without eliminating it. Controllers are not being replaced. They are being given a new kind of input, one generated by a system that processes more variables simultaneously than any individual could. The question is whether that input makes their judgment better or subtly constrains it in ways that are hard to detect until something goes wrong.

In Short

The FAA’s SMART system is now active across all 22 US air traffic control centers. It advises, it does not decide. Early results from the Washington, DC, deployment show 180 flights rerouted, with reported savings in time and fuel per flight. Airline executives support it. The air traffic controllers’ union says it was excluded from the entire process. That exclusion is not a footnote. In high-stakes AI deployments, the people closest to the work are often the ones who can see what the system cannot.

Based on reporting from Ars Technica.

Written by

NAVION