The FAA has committed $875 million to an AI-assisted air traffic management tool, with a pilot rollout starting at three major Washington, D.C.-area airports — and the CEO of the company building it is working to draw a hard line between optimization and automation.

Transportation Secretary Sean Duffy unveiled the software, called Strategic Management of Airspace, Routes and Trajectories (SMART), last week. Boston-based Air Space Intelligence (ASI) is developing the system as part of the federal agency's broader overhaul of U.S. air traffic control infrastructure.

"I want to be very clear because this is important. This is not about automating air traffic control," said Phillip Buckendorf, ASI's co-founder and CEO, in a wide-ranging interview. "What we're focused on is how do we optimize the airspace in its entirety."

Who makes the decisions under SMART?

Buckendorf stressed that air traffic controllers will keep responsibility for separating aircraft. SMART targets traffic managers — the operators who manage airspace as a whole — by generating recommendations on flight routing and scheduling.

"The AI does the math, the AI does the computation," Buckendorf said. But the human operator decides whether to accept the recommendation. "They make the decision, they are the decider, they are the judge."

He compared the tool's function to chess, letting managers see four or five moves ahead. The system draws on more than 200 data sources, including:

• weather and turbulence
• congestion
• winds
• runway closures

The goal, he said, is computing the most optimal flight path for every flight in the country — reducing delays and cancellations, and improving on-time performance.

SMART builds on ASI's Flyways AI, a dispatch tool already in use at Alaska Airlines and other carriers. The FAA deployment raises the stakes considerably: it places the technology inside federal airspace management rather than airline operations centers.

What is the pushback?

The D.C.-area pilot has already drawn political objections. Rep. Don Beyer (D-Va.), whose Northern Virginia district includes Ronald Reagan Washington National Airport, attacked the rollout in a statement.

"The FAA should not use my constituents as guinea pigs for an unproven AI air traffic control system," Beyer said. "Dangerously rushing a system that holds Americans' lives in its hands is not a good use case for artificial intelligence."

Buckendorf acknowledged public anxiety about AI in air traffic control but argued it stems from unfamiliarity with how the software works.

"Come and see it. Look at the technology. I think a lot of the skepticism comes from actually not knowing," he said. "The machine never makes a decision. It's a human operator, the air traffic manager. And we want them to have the very best technology. We don't want them to run on 1980s software. Who can disagree with that?"

Can software fix a staffing problem?

Skeptics question how much difference SMART can make in a system already stretched to capacity at major hubs. Dave Riley, a former FAA air traffic controller, called the system "a step in the right direction" — but not a "cure all" for structural problems, including a longstanding controller staffing shortage that forces many controllers into mandatory overtime and grueling schedules.

Riley also accused the FAA of overselling the tool's impact on the traveling public.

"Throw the buzzword 'AI' on it and get everybody's attention," Riley said. "That's what I see."

Buckendorf's counterargument rests on the gap between consumer technology and the tools traffic managers currently use. He noted that mapping apps on their phones often carry more sophisticated predictive capabilities than the software they run at work.

"You want to make sure that those specialists, those very, very well-trained operators that make really critical operations when they're on shift, that they have world-class software," he said.

The rollout's outcome at the three D.C.-area airports will offer the first hard evidence of whether SMART delivers measurable capacity and delay reductions — or becomes another optimization layer on top of an understaffed system.