The FAA Is About to Hand Air Traffic Congestion to an $875 Million AI

Hero photo: The control tower at Ronald Reagan Washington National Airport — Carol M. Highsmith, Wikimedia Commons, public domain (Library of Congress collection).

01 What is actually launching

On September 21, the Federal Aviation Administration begins live operation of SMART — Strategic Management of Airspace, Routes and Trajectories — an AI system that predicts air-traffic flows and flags potential conflicts for the three major airports serving Washington, D.C.: Reagan National, Dulles, and Baltimore-Washington International.

SMART is the first product of an $875 million, 12-year contract awarded in June to Boston-based Air Space Intelligence. It ingests airline schedules, weather forecasts, runway capacity, and airspace conditions, and produces routing recommendations and congestion forecasts before disruptions cascade. Officials told the Wall Street Journal the D.C. debut is deliberate: start small, in the hardest, most-watched airspace in the country, then expand.

02 Why Washington first — and why now

The choice of D.C. is not random. The January 2025 midair collision over the Potomac — the deadliest U.S. aviation disaster in a generation, 67 killed — happened in exactly this airspace, and it converted modernization from a budget line into a political mandate. A fatal runway crash at LaGuardia in March kept the pressure on. FAA Administrator Bryan Bedford and Transportation Secretary Sean Duffy have made AI-assisted modernization the agency's signature push.

Industry reaction has cycled from panic to acceptance. Airline officials initially worried about leaning on a brand-new tool during the fall schedule; multiple people familiar with the rollout describe that fear as largely resolved, with the caveat that the 90-day test will decide whether controllers and carriers can trust its predictions in live conditions.

03 What SMART does — and what it deliberately does not do

The FAA's framing is precise: SMART is a decision-support tool. It forecasts congestion hours ahead, recommends alternate routing, and identifies where schedules and capacity are on a collision course. What it does not do, in any announced phase, is separate aircraft. That authority stays with human controllers — the FAA has repeated this in every description of the program, because the distinction is the whole policy story.

Future iterations are more ambitious: managing arrival intervals and mapping routes days in advance. A separate system under the same contract, Flow Management Data and Services, will eventually replace the machinery at the FAA's Air Traffic Control System Command Center in Virginia. SMART, in other words, is the first visible piece of a plan to make prediction — not reaction — the operating principle of the national airspace.

04 The trust problem: 90 days that decide a decade

The official plan is a roughly 90-day live test in Washington, after which the FAA decides on wider expansion, with nationwide rollout projected for 2028. That test carries enormous weight. Aviation safety runs on a simple rule: nothing enters the system until it has been proven safer than what it replaces. AI advisory tools invert some of that logic — they ask controllers to treat a probabilistic forecast as an input to judgment, not a clearance to act.

The failure modes are asymmetric. If SMART's advice is too conservative, it will be ignored, and the system becomes expensive wallpaper. If it is overconfident, one bad recommendation in a disruption could poison trust for years — the aviation-safety equivalent of a recall. The 90-day window is where the FAA learns which of those failure modes it actually has.

There is also a workforce dimension. The FAA faces a controller shortage that predates SMART by a decade. The honest case for the system is that it multiplies scarce human attention; the risk is that it becomes a substitute for hiring, a tension the controllers' union has already flagged in other modernization fights.

05 The bigger question: AI in safety-critical infrastructure

What makes SMART nationally significant is not aviation — it is precedent. This is the first time a U.S. safety-critical infrastructure operator has put a frontier-scale AI system into daily live operation as an official layer of the system itself. The questions the FAA is answering in Washington are the same ones the power grid, rail, and healthcare will face within the decade: How much decision authority do you give a model? What is the audit trail? Who is liable when the advice was wrong?

The FAA's answer — advise, never act; human authority always; test small and expand only on evidence — is the conservative template. If the 90-day test succeeds, expect other infrastructure operators to copy it wholesale. If it fails publicly, it sets the AI-in-critical-infrastructure conversation back years. Very few government technology deployments carry that kind of either-way significance.

06 The verdict

The verified facts: SMART debuts around September 21 for DCA, IAD, and BWI under an $875 million, 12-year contract with Air Space Intelligence. It predicts flows and conflicts from schedules, weather, and capacity data. Controllers keep full separation authority. A roughly 90-day test precedes wider expansion; nationwide rollout is projected for 2028.

The stakes: this is the first live test of AI as official infrastructure in U.S. aviation — run in the airspace where the case for modernization was made in the worst way. The test is not whether AI can help manage congestion. It is whether a bureaucracy built on deterministic rules can build trust around probabilistic advice.