FAA's $12B SMART program is advisory AI for flight scheduling weeks ahead, not autonomous control — and Palantir is at the table

Transportation Secretary Sean Duffy announced this week that the FAA is contracting with three vendors — Palantir, Thales SA, and Air Space Intelligence — for an AI flight-scheduling system called SMART, which stands for Strategic Management of Airspace Routing Trajectories. The system is being pitched at $12 billion in total spend, and Duffy went out of his way to clarify the scope on CBS: the software is advisory, not autonomous. It looks ahead weeks at flight schedules, identifies likely conflicts, and suggests small timing adjustments — Duffy's example was nudging some flights five, seven, or ten minutes earlier or later, 45 days in advance. He explicitly said the software would stop short of "replacing humans in how we manage the airspace."

That framing is more conservative than the news cycle is treating it. The actual cognitive load on a working air traffic controller during a live shift — separating aircraft within seconds, handling go-arounds, reading weather — is not what SMART is meant to help with. Strategic scheduling weeks out is exactly the kind of slow, structured, large-state-space optimization problem that ML systems do well: you have most of the relevant information, the consequences of bad suggestions are reviewable rather than catastrophic, and a human always signs off. If you were going to introduce AI into the FAA's pipeline, the part that runs 45 days ahead with full human review is the right entry point, and SMART as described is a sane product design rather than a leap into autonomous safety-critical control.

The substantive story underneath the announcement is which three vendors are competing for the contract. Palantir's presence is the part that matters politically: the company has spent the last two years cementing itself as the default federal AI infrastructure provider across DoD, ICE, IRS, and now aviation, and the SMART procurement extends that pattern. Thales SA is the European defense and avionics incumbent and the one with the most legacy ATC integration depth. Air Space Intelligence is the smaller specialist play, founded in 2018 with a focus on aviation-specific ML. The choice of vendor will determine whether SMART becomes a bespoke aviation system, a Palantir Foundry deployment with an aviation skin, or a hybrid. Each path has different long-term implications for who controls the data plane and who can audit the model behavior.

The honest concerns about SMART are not the alarmist ones. The real questions for builders watching this deployment are whether the suggestion-acceptance interface gives controllers time to understand the model's reasoning, whether the training data covers the long tail of unusual operational conditions (severe weather diversions, runway closures, geopolitical airspace shifts), and how the system handles disagreement when its suggestion conflicts with controller intuition. None of those are addressed in the announcement. The FAA has a long track record of deploying decision-support tools cautiously and rolling back when they introduce more risk than they remove; the SMART rollout is worth watching as a real-world test of whether the federal procurement and oversight stack can validate ML behavior at the level needed for safety-of-life systems. Probably yes for the strategic-planning use case Duffy described. Definitely not yet for anything closer to the live tower, regardless of what the next administration tries to claim.