The FAA has received roughly 200,000 applications for air traffic controller positions over the past several years. Fewer than 2% of applicants ultimately become certified. That number is not a story about AI replacing a profession. It is a story about a profession the country cannot staff fast enough, in a job where AI's role is explicitly limited by law, physics, and the consequences of getting it wrong.
The displacement risk is classified as Moderate, with a horizon toward 2038. That reflects real automation already underway, sequencing tools, conflict detection, surface-management systems, layered into towers and control centers to reduce the manual workload on a controller during a busy push. It does not reflect a path toward removing the controller from the loop. Certifying a fully autonomous separation-of-aircraft system to the reliability standard the job requires is not currently possible under any existing regulatory framework, in the United States or internationally.
The more urgent story is not automation. Between 2015 and 2025, the FAA's certified controller workforce shrank by roughly 6% while U.S. flight volume grew about 10%. As of 2025, an estimated 91% of the country's air traffic control facilities were operating below the FAA's own staffing targets. AI is being deployed into this gap as relief for an overworked workforce, not as a replacement for it.
Key Points
- Air traffic controllers are rated Moderate risk with a displacement horizon toward 2038, constrained less by AI capability than by certification standards that make fully autonomous separation control legally and technically unworkable today.
- The FAA's controller workforce fell about 6% between 2015 and 2025 while U.S. flight volume rose roughly 10%, and the National Academies estimates the system needs 4,049 more certified controllers to meet its own staffing standards.
- Real AI and automation tools, including the FAA's Terminal Flight Data Manager, Time-Based Flow Management, and NASA's ATD-2 surface-management system, are already reducing manual coordination workload at dozens of major airports.
- Regulatory bodies including Eurocontrol and ICAO explicitly reject full automation, citing an accountability gap: aviation law assumes a human controller holds legal responsibility, and no framework yet assigns that responsibility to software.
- Chronic understaffing has been directly linked to real safety incidents, including a spike in near-misses in 2023 and the January 2025 midair collision at Reagan National Airport, underscoring that the near-term risk to the profession is burnout and short-staffing, not job loss.
What an Air Traffic Controller Actually Does
Controllers direct the movement of aircraft on the ground and through controlled airspace, sequencing takeoffs and landings, maintaining legally mandated separation between aircraft, and adapting continuously to weather, equipment failures, and traffic surges that no schedule anticipates. The job is executed in real time, under strict regulatory standards, with consequences for error that are immediate and irreversible in a way few other professions share.
It is also, by design, a job built around sustained attention during unpredictable peaks. A controller working a busy sector during a weather reroute is managing dozens of moving variables simultaneously, aircraft speed, altitude, spacing, pilot communication, and contingency planning, all while holding legal responsibility for the outcome.
What AI Is Already Doing
The FAA's Terminal Flight Data Manager is the clearest example of automation already embedded in the job. TFDM consolidates legacy systems, fuses data from multiple sources, and gives controllers dynamic tools like the Airport Resource Management Tool to allocate runway demand and manage delays. Deployment began in 2022 and is scheduled to reach 89 sites by 2029. Paired with Time-Based Flow Management, which schedules aircraft to reach constraint points at specified times rather than requiring manual vectoring, these systems are explicitly designed, in the FAA's own language, to improve capacity "without decreasing safety or increasing controller workload."
NASA's Airspace Technology Demonstration 2 program took a similar approach to surface operations, demonstrated at Charlotte Douglas International Airport between 2017 and 2020. Instead of aircraft idling in physical queues on the taxiway, ATD-2 created virtual, reservation-style queuing, aircraft wait at the gate with engines off until their slot opens. The demonstration saved more than a million gallons of jet fuel and prevented an estimated 933 hours of passenger delay.
On the forecasting side, machine learning is entering weather-based rerouting. Google's WeatherNext model processes atmospheric data far faster than traditional physics-based forecasting and is now used commercially, including in a 2026 partnership with Ryanair covering fleet operations and weather-informed scheduling. Eurocontrol, meanwhile, has more than 30 AI-based applications in development across traffic prediction, trajectory optimization, and conflict detection under its SESAR modernization programme, each one framed explicitly as workload reduction for controllers rather than as a path to removing them.
THE GAP
The National Academies estimates the FAA needs 4,049 additional certified controllers just to meet its own staffing standards. In 2024 alone, controller overtime cost the system more than $200 million. No automation program currently deployed or scheduled closes a gap that size. It reduces how hard each remaining controller has to work. It does not reduce how many the system needs.
Why Full Automation Isn't Coming
The regulatory ceiling here is explicit, not incidental. Certifying an autonomous system to perform aircraft separation, the core safety function of the job, requires demonstrating a reliability standard that current automated decision systems cannot be independently verified against. The European Union Aviation Safety Agency only began working on a certification framework for automated air traffic systems in 2023, and no comparable framework exists yet for fully autonomous separation authority anywhere in the world.
Eurocontrol has been direct about where it draws the line: controllers bring "indispensable skills such as judgment, flexibility, and the ability to handle unexpected situations," and its AI strategy is built around explainable systems that forecast and suggest while leaving controllers with final decision authority. Underneath the engineering question sits a legal one that international aviation law has not resolved: accountability. Aviation law assumes a human holds operational responsibility for a separation decision. No framework currently assigns that responsibility to an algorithm, and until one does, full automation has nowhere to go even if the technology were ready.
When Understaffing Becomes a Safety Problem
The consequences of the staffing shortfall are not abstract. In August 2023, reporting on FAA data found 46 close calls involving commercial airliners in a single month, part of a broader spike investigators linked substantially to controller staffing shortages and overwork. An NTSB investigation into a February 2023 near-miss at Burbank's Bob Hope Airport, where a landing aircraft came within 1,680 feet of one taking off, found that the controller had been distracted while managing a third aircraft, a pattern consistent with sustained understaffing rather than an isolated lapse. The FAA logged 1,757 runway incursions in 2024 alone.
On January 29, 2025, a midair collision over the Potomac River near Reagan Washington National Airport killed 67 people, the deadliest U.S. commercial aviation accident in more than two decades. The FAA confirmed that controller staffing at the facility was not at normal levels at the time of the accident. The investigation into the full chain of causes was ongoing at time of writing, and staffing was one factor among several under review, not an established sole cause. What the incident did make unambiguous is the stakes attached to a staffing shortfall that had already been flagged for years by the GAO and the National Academies before it happened.
How to Use AI as an Air Traffic Controller Now
The controllers and facilities getting the most out of these systems are treating automation as a way to protect attention, not replace judgment.
For sequencing and surface management: tools like TFDM and TBFM absorb the repetitive coordination work, freeing attention for the situations that actually require judgment, weather deviations, emergencies, unscheduled traffic. Treat the automated sequence as a strong default, not a fixed outcome, when conditions change faster than the system updates.
For smaller and lower-traffic facilities: remote digital tower systems, such as Saab's r-TWR platform, already operational at sites including a NATO air base in Germany since 2022, use AI-enhanced camera and sensor processing to let a single controller support multiple smaller airports from a centralized location. This is a targeted answer to a specific staffing problem at low-traffic sites, not a general-purpose substitute for towered control at major hubs.
What I Think
The 2038 horizon reflects the technology curve reasonably well. What it does not fully capture is that this profession's most urgent risk in 2026 has almost nothing to do with AI. The system is short thousands of controllers against its own staffing model, spending hundreds of millions of dollars a year on overtime to cover the gap, and the safety record shows what happens when that gap runs for too long without correction. It is the same shape of problem AI Doomsday found in public transit driving: a transportation workforce shrinking faster than automation can replace it.
AI's role here is legitimately useful and legitimately limited at the same time. Sequencing tools and surface-management automation measurably reduce workload per controller, and that matters. But workload reduction per person is not a substitute for having enough people, and no automation roadmap currently on the table, in the U.S. or in Europe, proposes to close the headcount gap rather than make each remaining controller's job more survivable. Hiring and training more controllers, faster, is the actual bottleneck. AI is not positioned to solve it, and no one seriously building these systems is claiming otherwise.
"No certification standard exists for a machine to hold what a controller holds: legal responsibility for keeping two aircraft apart. Until one does, the job isn't going anywhere. What has to change is how many people are doing it."