Amadeus' vision for agents in Aviation

Date (UTC): 2026-08-09

Hi its Sam Chandra - CEO/Founder of Deepsky and the author of this newsletter.

This is the first of two editions this week. Today is the industry, everyone who shipped something, and tomorrow is the other half, the regulators, the safety cases and the investigations. Split that way because there was too much of both to run together.

On the industry side Archer’s Zee is now reading the airport surface, Wisk had one person safely mind three aircraft at once, a crewed jet and a drone flew as a team, and one autonomy stack ran a small swarm on its own. And underneath all of it, AMD just bought a way to bake a finished model straight into a chip. Enjoy.

🗼 ATM

Archer’s Zee Learns to Predict the Airport Surface

A fortnight after launching Zee, Archer showed it predicting aircraft movements on taxiways and runways several minutes ahead, reading airfield geometry from satellite imagery so it works at airports it has never seen. It is being evaluated against real traffic at Hawthorne (California), with runway incursions the target problem.

✈️ Airlines & AAM

Amadeus Maps the Agentic Airline

Amadeus, with Microsoft, published a report on where agentic AI is ready to work inside an airline, built on interviews with Azul, Icelandair and Southwest. The concrete part is the five use cases it says are mature enough to deploy now, including a voice agent that takes a rebooking end to end (understands the change, offers options, quotes the fare difference and takes payment) and an AI that coordinates the turnaround across maintenance, crew and refuelling. It is a roadmap rather than a launch, but it is the clearest published map of which ops workflows go agentic first.

Wisk Tests One Human Watching Three Autonomous Aircraft

Wisk trialled a system whereby three autonomous aircraft are supervised by one person at a time, in conjunction with NASA at the Ames ATC laboratory, using existing ATC tools and procedures. The early answer is that it works, which matters because the economics of autonomous air taxis die if every aircraft still needs its own dedicated human.

🧩 Airports

Changi’s Wheelchairs Now Drive Themselves

SATS expanded its autonomous wheelchairs from Changi Terminal 3 into Terminal 2 after a trial covering more than 13,000 passengers: sensor-guided units that self-navigate transit routes, detect obstacles and pedestrians, and report to a central supervisor. The service now covers roughly a tenth of Changi’s daily mobility-assistance demand, freeing staff for passengers who need the human.

🛩️ Military

The X-62A’s AI Flies 27 Live Intercepts Using the Jet’s Own Sensor

The USAF Test Pilot School and Lockheed’s Skunk Works disclosed that the X-62A VISTA flew 27 AI-controlled intercepts against a live T-38 across eight flights, with the autonomy agent taking data from an operational infrared search and track pod and executing the intercept itself. Every previous test fed the AI simulated target data; this is the first closed sensor-to-action loop in real flight, and a radar goes on next.

Boeing and Northrop Demonstrate P-8 Crews Tasking an Autonomous Triton

Boeing and Northrop Grumman demonstrated a P-8A crew sending machine-readable tasking to a simulated MQ-4C Triton, which autonomously planned its route, collected the maritime surveillance, ran onboard AI processing and reported back, no ground control station in the loop. The load-bearing part is the standards-based interface, designed to extend to any compliant platform, which is how crewed-uncrewed teaming stops being a per-aircraft engineering project.

Shield AI’s Hivemind Runs a Three-Drone Search on Its Own

In a Taiwanese exercise, Shield AI’s Hivemind autonomy divided a wide-area search among three Mighty Hornet III UAVs, coordinated their movement and communications, navigated a restricted-airspace sector and recovered all three, with the AI doing the tasking rather than an operator flying each aircraft. It is a clean demonstration of one autonomy stack commanding a small swarm, and Hivemind is the same software thread we have followed through Shield AI’s other deals.

🧠 AI (in general)

AMD Buys the Company That Etches AI Into Silicon

AMD acquired Taalas, a startup that burns a trained model’s weights permanently into a custom chip, trading the flexibility of a GPU for very fast, very cheap, fully offline inference at around 17,000 tokens a second. For aviation the interesting part is the freeze: a model etched into silicon is a fixed, deterministic artefact you could certify once and know it will never silently change under you, which is the opposite of the monthly-model-update problem operators keep running into.

😎 Not AI

DoorDash Becomes an Airline

DoorDash received FAA Part 135 air carrier certification and is launching DoorDash Air, flying delivery drones designed and built by its in-house robotics unit, with commercial deliveries expected from this autumn. That makes the food delivery app the eighth certificated drone package carrier in America, alongside its existing Wing and Flytrex partnerships.

Sam here again, hope you have enjoyed reading this edition. It becoming clearer, AI is becoming powerful, making things faster and easier - but as we will see in tomorrows edition, that power has some big risks with it too.

Banking the wins while keeping your operation running is exactly what Deepsky helps aircraft operators with. We are seasoned flight operations experts with deep AI understanding, and if you want guidance to help bring AI into your flight operation, contact me at sam@deepskyai.com or visit deepskyai.com.

Tomorrow’s edition covers the other half, the regulators, the safety cases and the dangerous side.