Saab's Gripen E AI Flights Are the Best Available Evidence That Cockpit AI Works — and Commercial Aviation Isn't Looking
The commercial aviation industry spent much of 2026 debating whether AI belongs in the cockpit at all. Meanwhile, Saab quietly went and flew it.
Saab has flown its own AI agents aboard the Gripen E, with an agent autonomously controlling the fighter during beyond-visual-range combat scenarios — trials that demonstrate software that could eventually support pilots in the cockpit or operate alongside them aboard uncrewed aircraft. That’s not a simulation result or a controlled demo for a conference stage. Those are operational flight hours, on a production aircraft, with AI making real-time control decisions.
Gripen E’s software-defined architecture allows new AI software to be integrated and flown on an operational fighter without requiring a dedicated experimental aircraft. That last detail is worth sitting with for a moment, because it describes something commercial EFB and cockpit-software vendors have been promising for years without fully delivering: an architecture where the aircraft itself is a software-defined platform, not a fixed piece of hardware with a narrow software aperture bolted on top.
The Validation Gap Commercial Aviation Keeps Ignoring
Here’s the argument I’d make, and I think it’s defensible: the commercial aviation industry does not have an AI capability problem. It has an AI validation problem — and it keeps looking for validation in the wrong places.
I recently polled industry professionals on who builds the best flight ops software. Dedicated vendors came out on top at 46%, airline in-house teams were a close second at 38%, and AI startups landed at just 4%. That 4% isn’t just market skepticism of new entrants — it reflects a deeper, structural problem: the industry has no agreed framework for what it means to validate AI in the cockpit, so the default position is not to trust it at all.
Saab’s program is instructive precisely because it has a validation framework. The agents train through Saab’s AI Fighter Pilot Academy, accumulating millions of simulated flight hours across different operational scenarios, with their actions repeatedly assessed against defined objectives — allowing them to explore tactics and improve decision-making before flight testing. That sequencing — simulate extensively, assess against defined criteria, then fly — is actually a close mirror of what EASA has been moving toward with its own AI regulatory work. The difference is that Saab has closed the loop into real flight operations. Commercial aviation is still largely in the simulation phase.
The speed at which such agents can be deployed is accelerated by the Gripen E’s open architecture, which segregates mission software from flight-critical functions. That segregation is the architectural principle that makes incremental AI deployment possible — and it’s precisely the principle that the commercial cockpit stack has been slowest to adopt. When avionics stacks are monolithic and change-resistant, there’s nowhere to safely insert a new AI layer. The military figured out that you have to build the deployment pathway before you have the AI, not after.
What the Commercial Cockpit Stack Should Take From This
I’m not suggesting that autonomous combat maneuver AI maps directly to commercial decision support. The operational contexts are genuinely different, and the liability structures are incomparably more complex on the commercial side. But the architectural lessons transfer more cleanly than the industry tends to admit.
Saab has also been working with other agent developers, as it sees the ability to integrate third-party agents alongside its own as a key aspect for future AI development. That is, essentially, a vision of an open cockpit AI platform — one where the aircraft operator isn’t locked into a single vendor’s model, but can compose agents for specific tasks. The commercial EFB ecosystem is trending in roughly the same direction, but slowly, and mostly for lower-stakes applications like document retrieval and preflight briefing support. That sequencing is correct, in my view — prove AI on the low-stakes applications before moving toward in-flight decision territory. But the pace needs to accelerate, and the architecture decisions being made right now in cockpit software platforms will determine whether that escalation is even possible.
The deeper point is this: the best evidence that cockpit AI works at operational scale is currently being generated in military programs, not commercial labs. If commercial aviation wants to build genuine institutional trust in AI-assisted flight — not just regulatory acceptance, but the kind of crew-level confidence that actually changes behavior — it needs to be paying much closer attention to what programs like Saab’s are demonstrating about how that trust gets earned. The validation pathway they’ve built isn’t exotic. It’s rigorous. And rigor is exactly what the commercial side keeps saying it’s waiting for.