CORTEXFLOW · FREE FOREVER

News

The FAA Put an AI Agent on Every US Flight Route — Three Days Later, Every Departure Stopped

The FAA Put an AI Agent on Every US Flight Route — Three Days Later, Every Departure Stopped
FIG. 01 — The FAA Put an AI Agent on Every US Flight Route — Three Days Later, Every Departure StoppedSCALE 16:9

Searches for “faa ai agent” spiked up to +3,150% in the US this week — the fastest-rising AI-agent query on Google Trends (US, past 7 days). The reason is a real deployment with real stakes: on November 10, 2026, the FAA switched on an AI route advisor called SMART at all 22 of America’s en route air traffic control centers. Three days later, on Friday, November 13, the FAA ordered a nationwide ground stop and every departure in the country stopped.

This post covers what SMART actually is, what happened on those three days, whether the AI tool caused the stop, and the design pattern every n8n builder should steal from it — because SMART is the clearest real-world example yet of how to give an agent power without giving it authority.

What SMART is

SMART stands for Strategic Management of Airspace, Routes and Trajectories. It is the AI layer on top of FMDS (Flow Management Data and Services), the cloud platform replacing the FAA’s legacy traffic-flow system. The FAA describes it as a platform that gives controllers and operators a shared picture of schedules, weather, airport capacity, and airspace conditions before flights depart.

It works like this: SMART pulls in about 200 data streams — weather, filed flight plans, airline schedules, airport capacity, traffic flow, and controller staffing — and looks for congestion hours before it happens. Then it suggests a reroute or a new departure time while the airplane is still at the gate.

Three facts define what it is — and what it is not:

  • It recommends, it never decides. A traffic management specialist at the SMART center in Washington analyzes and validates each suggestion, then sends it to the Air Traffic Control System Command Center in Warrenton, Virginia. Local facility managers decide whether to use it. Air Space Intelligence chief executive Phillip Buckendorf has put it simply: the machine never makes a decision.
  • It works in hours, not seconds. SMART flags trouble hours ahead. It does not talk to pilots, issue clearances, or move airplanes. Controllers still separate aircraft, on the same radar and automation systems as before.
  • It cannot take over a flight. Transportation Secretary Sean Duffy has said there is “always going to be a human that manages the airspace in America.” The FAA says the tool cannot take over control of an aircraft or replace controllers.

The vendor behind it is Air Space Intelligence (ASI), a Boston company founded in 2018 as Uni Robotics. On June 22, 2026, Duffy and FAA Administrator Bryan Bedford announced ASI had won the FMDS and SMART work — a roughly $875 million, 12-year award — ahead of much larger bidders including Palantir and Thales. On October 20, 2026, ASI announced a $325 million Series C led by Spark Capital.

The three days: Nov 10 to Nov 13

DateWhat happened
Sep 21, 2026SMART goes live in a pilot covering the crowded airspace around Reagan National, Dulles, and Baltimore/Washington.
Nov 10, 2026SMART expands to all 22 US en route centers — nationwide.
Nov 13, 3:54 p.m. ETFAA orders a nationwide ground stop (ATCSCC Advisory 111) over a temporary telecommunications outage at the en route centers.
Nov 13, 8:15 p.m. ETGround stop lifted. Departures resume.

In the pilot, the FAA says SMART helped reroute 180 flights, saving an average of about 10 minutes and nearly 80 gallons of fuel per flight. Those are the FAA’s numbers, reported by AVweb and Gizmodo — I could not independently verify the savings math, and the FAA has published no raw data to replay it against.

The ground stop itself was rare and total: filed departures across the United States fell off a cliff on Friday afternoon. The FAA cited a telecommunications outage at the centers that handle high-altitude traffic. As of this writing, the FAA has not published a final finding on what failed.

Did SMART cause the ground stop?

No evidence says so — and the timing is the whole story. The nationwide rollout finished on Tuesday; the stop came Friday. Within hours, Air Space Intelligence said the stop was “not based on or a result of the SMART system, any product of ASI or our integration with FAA and ATC.” WIRED noted the rollout and added there was no evidence the ground stop related to the tool.

Homeland Security Secretary Markwayne Mullin was less categorical. Asked whether the AI tool was involved, he said: “We’re looking at that,” adding that the department’s cyber agency, CISA, was taking a look, and that there was no indication of a threat to the country.

That is the honest state of it: the vendor denies a link, the government is checking, and no public evidence connects the two events. The FAA’s cited cause — a telecom outage — is a different system from the AI advisor. Treat any claim that SMART caused the stop as unproven, and treat the vendor’s denial the same way: a statement, not a finding.

The oversight picture is worth noting too. The controllers’ union, NATCA, says it was not involved in the design, testing, or implementation of SMART. On October 13, 23 House Democrats led by Reps. Rick Larsen and André Carson asked the Government Accountability Office to audit FMDS and SMART, citing questions about testing and training that the FAA’s rollout announcement did not address.

Why this matters to n8n builders

SMART is the pattern your agents should copy — and the warning your agents should heed. An AI system now touches the routing of every commercial flight in the US, and the reason nobody panicked about the AI itself is that the design never lets it act alone. Here is the pattern, translated to n8n:

Fig 1 — The SMART advisory console: a data-intake drum pulls in ~200 streams, the AI press stamps out route suggestions, and the lit orange approval lever — the human — sits between the press and the route board. The machine never touches a flight directly. Fig 1 — The SMART advisory console. The data drum takes in ~200 streams (weather, flight plans, schedules, staffing); the AI press stamps out route suggestions; the lit approval lever is the human gate between the suggestion and the route board. The readout is the real launch data: 200 streams, 180 reroutes in the pilot, 22 centers, 0 decisions by the machine.

1. Recommendation-only: the agent proposes, a human disposes

SMART suggests a reroute. A specialist validates it. A facility manager accepts or declines it. At no point does the model touch the thing that matters.

In n8n, that is the AI Agent node feeding a draft action into an approval step — a Slack/Telegram message, a form, or a Human-in-the-loop node — before any node that spends money, sends a message, or changes a system. The agent’s output is a proposal until a human says otherwise. Build the workflow so the execute step is unreachable without the approval step: wire the approval as a gate, not a notification.

2. Scope the tools to the safe layer

SMART works hours ahead, on planning — not seconds ahead, on separation. The dangerous job (keeping aircraft apart) was never in the agent’s tool list at all.

Scope your agent the same way. Give it read tools and draft tools; keep the irreversible tools (payments, deletes, publishes, sends) behind the approval gate or out of the agent’s reach entirely. If a tool can hurt you, it should not be callable in the same turn as the reasoning that chose it.

3. Name the human who is accountable

Every SMART suggestion passes through a named role — traffic management specialist, then facility leadership — before it becomes real. There is no step where “the system decided.”

Log the same chain in your workflows: who approved, what was approved, when. n8n’s execution log plus an approval message that records the approver’s name and the exact payload is enough. When something goes wrong at 3 a.m., “the agent did it” is not an answer; “Lalit approved this payload at 14:02” is.

4. Fail safe, not silent

When the FAA’s communications failed on Friday, the system stopped departures rather than guessing. A ground stop is expensive and embarrassing — and it is the correct failure mode for a safety system.

Your agents need the same default. If a tool call fails, a credential expires, or a status check is inconclusive, the workflow should stop and alert — not retry blindly, not proceed with stale data, not silently skip the step. Check the status page before debugging your workflow, queue the work, and replay it idempotently when the service returns. (The same pattern applied when Claude went down twice on Friday the 13th — see Claude Went Down Twice on Friday the 13th.)

5. A vendor denial is not a root cause

ASI’s denial came within hours; the government’s review is still open. Both can be true at once, and neither closes the question.

When your agent sits next to an outage — a failed webhook, a vendor incident, a bad deploy — do not accept the first “not us” at face value, and do not blame your agent either. Check the vendor status page, check your own execution logs, and keep the receipts. The finding comes later; the audit trail is what you build now.

What SMART gets right — and what it doesn’t

What it gets right is the architecture: recommendation-only, human validation, scoped tools, and a fail-safe default. That is why an AI system can be live in national airspace without the public treating it as an autopilot for jets.

What is less reassuring is the governance around it. The controllers’ union was not involved in design, testing, or implementation. Congress asked for an audit a month before the nationwide rollout. And when the ground stop hit three days after go-live, the public had to take the vendor’s word that its system was uninvolved while the government said it was still looking. A recommendation-only agent with a weak audit trail is still a trust problem — it just fails more politely.

For builders, the lesson is the same one this blog keeps returning to: identity, scope, approval, and audit are not features you add later. They are the design. Wells Fargo’s new agent-liability terms, China’s agent ID rules, and now the FAA’s SMART rollout all point the same way — agents that act for people need a provable chain from suggestion to human decision. (Wells Fargo Just Made You Liable for Your AI Agent and China Just Gave Every AI Agent an ID Card cover the other two.)

A commercial airliner flying against a clear blue sky. A commercial airliner in flight — the part of the system SMART never touches.

FAQ

What is the FAA’s SMART tool?

SMART (Strategic Management of Airspace, Routes and Trajectories) is an AI route advisor built by Air Space Intelligence. It merges about 200 data streams — weather, flight plans, schedules, airport capacity, staffing — and suggests reroutes and departure times hours ahead. It went nationwide to all 22 US en route centers on November 10, 2026.

Did the FAA’s AI tool cause the November 13 ground stop?

No public evidence connects them. The FAA cited a temporary telecommunications outage; the vendor, Air Space Intelligence, denied any link; the Department of Homeland Security said CISA was reviewing it. The FAA has not published a final finding.

Does SMART control airplanes or talk to pilots?

No. It suggests routes and timing to human specialists, who validate the suggestions and pass them to facility managers. Controllers still separate aircraft using the existing radar and automation systems. It does not issue clearances or move airplanes.

How much did SMART cost?

About $875 million over 12 years, awarded to Air Space Intelligence on June 22, 2026, ahead of bidders including Palantir and Thales.

What should n8n builders copy from SMART?

The recommendation-only pattern: the agent proposes, a named human approves, and the execute step is unreachable without that approval. Scope the agent’s tools to the safe layer, log who approved what, and fail safe — stop and alert — instead of guessing when something breaks.

Key Takeaways

  1. SMART is a recommender, not a decider — 200 data streams in, route suggestions out, and a human validates every one before it reaches a facility.
  2. The Nov 13 ground stop has no proven link to SMART — the FAA cited a telecom outage, the vendor denied involvement, and CISA is reviewing; no final finding is published.
  3. Copy the architecture, not the hype — recommendation-only, scoped tools, named human accountability, and a fail-safe default are the pattern for any agent that touches something expensive.
  4. Governance is part of the design — NATCA was not involved in testing, and Congress asked for a GAO audit before rollout; a weak audit trail is a trust problem even when the agent behaves.
  5. When your agent sits next to an outage, keep the receipts — check the status page, check your logs, and don’t accept the first denial or the first blame.

Next: wire the approval gate from this post into your own highest-risk workflow — the one that spends money or sends messages — and test what happens when the approval step is skipped. If anything executes anyway, the gate was a notification, not a gate.

Photo: Pexels (free license) — commercial airliner against a clear blue sky. Sources: nacionale.com · aeropattern.com · Economic Times · explainx.ai · aiweekly.co. Trend signal: “faa ai agent” +1,950% (up to +3,150% on first load), #1–2 rising related query for “AI agent”, Google Trends Explore, US, past 7 days, Nov 15, 2026.

DOC. CF-RSS-001 · FORMAT: XML

Want more like this?

Practical tutorials on AI agents, automation, and chatbots — straight to your reader.

Subscribe via RSS

Doc. CF-NL-001 · Dispatch

Get the next build in your inbox

One practical email per new post. No spam, unsubscribe anytime.

One email per new post. No spam. Unsubscribe anytime.