Aviation leaders gathered at the AIAA AVIATION Forum in San Diego to detail how advanced automation and AI are entering live airline, airspace, and aircraft operations right now - not in prototypes, but in daily decisions affecting hundreds of thousands of passengers. The panel, titled AI, Autonomy, and Assurance: Operational Implications, drew executives from United Airlines, Reliable Robotics, Collins Aerospace, NASA, and the standards community to map where these tools are already in use and what it will take to certify them at scale.
United Airlines will hit 5,359 daily departures this year and recently flew 630,500 passengers in a single day, a company record. For Roberta Zimmerman, the airline's director of Air Traffic Strategy, Data Analytics, and Strategic Vision, AI is already embedded where it can reduce friction without touching safety-critical decisions.
"We are using AI to communicate with our customers on a flight-by-flight basis," Zimmerman said. "If you have a connection and your first leg is delayed, we give you alternatives. You can rebook, you can… [see the] predicted time it takes you to walk from one gate to the next gate at your connecting airport." Her team built those plain-language delay explanations at the direction of CEO Scott Kirby. They are also applying AI to crew scheduling, translating dense pilot and flight attendant contract rules into logic that supports more effective planning.
Zimmerman pointed to systems integration as the real barrier, not algorithm maturity. Changing one airport identifier - Palm Beach International switching from PBI to DJT - became a massive internal lift. "We have such a vast display of capabilities in the NAS, and we're not going to change the infrastructure overnight," she said. "We are a system of systems, and something very small, the integration of that is a huge impact to make sure that there's no loss of continuity."
Keeping AI out of the flight-critical stack
Reliable Robotics takes a different approach: embedding automation directly into the aircraft while deliberately keeping AI out of the flight-critical software. Brandon Suarez, the company's vice president of UAS Integration, described a program to automate the Cessna Caravan, an 8,000-pound single-engine turboprop. In a 2023 demonstration, a remote pilot 50 miles away controlled the aircraft through taxi, takeoff, and landing using GPS, INS, and radar-altimeters.
"AI as a tool that needs to go through a certification process is basically a non-starter for a startup company, because there are no rules and procedures and standards to follow," Suarez said. "So we're working to do everything that we're doing on the aircraft automation side with classic software coding languages and classic algorithms." Certification, he added, is about explainability: "Just trying to convince a disinterested smart person that what I did was correct, and I can't do that if I can't even explain what's happening in the software tool that I'm using."
Reliable also operates a Part 135 cargo carrier, Reliable Airlines, which will be the first operator of the automated system. Suarez said bringing pilots, maintainers, operators, and dispatchers into the design process early has been "hugely fruitful." He believes long-term autonomy must prove operational benefit, not just safety metrics: "We all want to increase safety, but meanwhile, there's billions of dollars of an ecosystem going, so we have to show the rest of the industry that there's actual operational benefit."
Increasing automation, not autonomy
Travis Klopfenstein, innovation program manager at Collins Aerospace, brought the avionics supplier view across cockpits, data services, and cabin systems. Collins is testing ATC speech-to-text conversion on an experimental flight deck and experimenting with lower-criticality applications like "Galley AI," which uses optical sensing to track inventory and passenger needs in the cabin. The company clusters early AI deployments in low-criticality zones while working through certification.
"We started to use phrases less about autonomy and maybe increasing automation," Klopfenstein said. "We're comfortable with increasing automation… really trying to optimize the human decision-making." He urged investment in high-fidelity modeling and simulation - potentially using commercial game engines - with an eye toward certifying those components. The larger threat, he suggested, is cultural: "One of the biggest threats is this kind of closed-mindedness… 'This is kind of how we've always done things, and this is how it has to go.' Now it's a limiting factor."
Defining "good enough" for machines
Chester Dolph, an engineer at NASA Langley Research Center, described a future where urban air mobility, multi-rotor UAS, supersonic demonstrators, traditional jets, and space vehicles share crowded skies. NASA's three pillars for this environment are strategic ATM to handle high-density, diverse traffic; safe routine operations that integrate onboard and ground-based sensors; and safety assurance that defines failure scenarios so AI and machine-learning approaches can be validated. Any AI-driven system, Dolph argued, must be generalizable, reproducible, and explainable. "Can you explain when it works and why it works, and when it fails, and why it fails?" he asked.
Anna Dietrich, a consultant and former COO of Terrafugia, said one of the hardest open questions is deciding how good is good enough when machines take on functions normally handled by humans. "We don't have quantitative consensus around reliability that we want out of our humans that are in these systems," she said. "We give people a lot of grace to screw up. We're not giving the systems that same grace… setting the bar is proving to be the hardest part." She sees promise in fail-functional architectures where a mature, simple core can bring an aircraft home if higher-level autonomy fails, and floated the idea of applying pilot-training-style, experience-based certification to autonomous flight managers.
Why this matters for operations teams
The panel sketched a picture where AI and automation are being judged not by novelty but by how well they fit into and improve existing workflows. For operations professionals, the takeaway is clear: the hardest problems are not the algorithms. They are systems integration, certification explainability, and the cultural willingness to move beyond "how we've always done things." Whether you're managing a fleet, an airspace, or a crew schedule, the tools arriving now demand that you define what "good enough" means for a machine - and that you bring your maintainers, dispatchers, and frontline operators into the design process from day one. For structured guidance on applying these principles to your own workflows, see the AI Learning Path for Operations Managers or explore resources on AI for Operations.
Your membership also unlocks:
