Daniel sent us this one about pilots and automation. Basically he's asking — if pilots are letting the autopilot do everything from takeoff to landing, how do they stop their actual hand-flying skills from crumbling? He wants the mechanics: the FAA and European regulations on recent flight experience, what simulator training actually covers, what airlines like Lufthansa and Delta and Cathay mandate above the regulatory floor, the documented automation dependency problem with Air France 447 and Asiana 214 as case studies, the recent regulatory push for more manual flying on revenue flights, the tension between fuel efficiency arguments for max automation and the manual skills argument for turning it off, and whether the next generation of pilots starting on heavily automated aircraft has the same baseline stick-and-rudder chops as the older generation. There's a lot to dig into here.
It's timely. The FAA put out a draft advisory circular in January specifically on manual flying proficiency. They basically said what safety people have been whispering for years: the data shows pilots aren't hand-flying enough to stay sharp, and the regulatory floor isn't doing the job.
I want to come back to that draft circular, because the language was unusually blunt for an FAA document. But let's start with the regulatory floor. What do the actual rules require right now?
Under FAA rules, a pilot needs three takeoffs and three landings within the preceding ninety days to carry passengers. Three landings in three months. And those can be in anything — a Cessna 172, a simulator, whatever. There's no requirement that any of those landings be hand-flown. You could do all three with the autopilot coupled all the way to touchdown and you're legal.
Three landings in ninety days. That's absurdly thin. You could be a long-haul pilot doing two round trips a month, never touch the controls, and you're technically current.
European rules under EASA are similar — three takeoffs and landings in ninety days, though they specify at least one must be in the actual aircraft type. But again, nothing says any of this has to be manual. The autopilot can do the whole thing and you've checked the box.
Which brings us to the actual problem. What happens when the automation quits and you haven't hand-flown an approach in six months?
That's Air France 447. And the details matter, because the popular version is just "the pilots didn't know how to fly" and it's way more specific than that. The BEA — the French aviation investigation bureau — put out their final report and it's devastating. The flight was cruising at thirty-five thousand feet over the Atlantic, the pitot tubes iced over, the airspeed indicators started giving conflicting readings, and the autopilot did exactly what it's designed to do — it disconnected and handed control back to the humans.
The humans weren't ready.
The junior pilot pulled the nose up — a classic startle response — and held it there. The stall warning sounded seventy-five times over about three and a half minutes, and neither pilot recognized they were in a stall. The BEA report found they were so accustomed to the automation handling abnormal situations that when it handed back control, they had what the report called "a total loss of cognitive control of the situation.
Seventy-five stall warnings. It's not that they made one mistake — it's that the fundamental mental model of what the airplane was doing had eroded. They were holding full back-stick in a stalled airplane falling at ten thousand feet per minute and didn't connect the dots.
The BEA specifically linked this to training deficiencies. Manual aircraft handling at high altitude was basically never practiced. Simulator sessions focused on systems management, not raw flying. The report said pilots received no training for "manual aircraft handling of approach to stall and stall recovery at high altitude." It wasn't in the curriculum.
That's one case study. Asiana 214 is the other one Daniel mentioned. San Francisco, July sixth, twenty thirteen. A Boeing 777, coming in on a clear day, visual approach. The autothrottle was armed but the pilot didn't realize it had been inadvertently deactivated. He expected the automation to manage airspeed. It didn't. Airspeed dropped, they got low, the tail struck the seawall. Three fatalities, dozens injured.
The NTSB used the phrase "automation dependency" explicitly in their findings. They said the pilots over-relied on automated systems and lacked the manual flying skills to recognize and recover from the deteriorating energy state. The flight crew had an "inadequate understanding" of how the autothrottle and autoflight systems interacted. They assumed the automation was protecting them, and it wasn't.
The pilot flying had something like ten thousand hours total time but only a hundred hours on the triple seven. He was transitioning from the seven forty-seven which had a different automation philosophy. The muscle memory and the systems intuition hadn't transferred.
That cuts to one of Daniel's deeper questions — whether the generation of pilots trained entirely on glass cockpits has the same baseline skills. The Asiana captain had thousands of hours, but he was new to that specific automation paradigm. The NTSB found his manual flying skills on the 777 were simply not where they needed to be for a hand-flown visual approach with the autothrottle off.
Let's talk about what airlines are actually doing beyond the regulatory floor. Daniel specifically asked about Lufthansa, Delta, and Cathay.
Cathay Pacific is probably the most interesting case. After some internal safety data that reportedly showed a drift in manual flying proficiency, they implemented a policy requiring pilots to hand-fly a certain number of approaches per month. They didn't publish the exact number publicly, but the industry chatter is it's around two to three raw-data hand-flown approaches per month minimum. And they track it through their flight data monitoring program.
Raw data meaning no flight director, just looking at the instruments and flying the airplane.
No flight director bars telling you where to point. Just you, the attitude indicator, the airspeed tape, and the glideslope. Cathay basically said we're not waiting for the regulator to fix this — we're going to mandate it ourselves.
What about Delta?
Delta's approach has been more about encouraging manual flying rather than hard quotas. Their flight operations manual recommends hand-flying below ten thousand feet when conditions permit, and they've integrated more manual flying scenarios into their recurrent training. But they haven't gone as far as Cathay with a strict minimum. What Delta has done is invest heavily in their sim programs — full-motion sims running scenarios designed to degrade gracefully, where the automation starts failing in subtle ways and the pilot has to catch it.
Delta's betting on sim fidelity rather than revenue-flight mandates.
Partly, but the FAA draft advisory circular from January kind of throws cold water on that approach. The circular said — and I'm paraphrasing but not much — that simulator training alone isn't sufficient to maintain manual flying proficiency. The FAA's position now is that sims are great for procedures and emergencies, but they don't replicate the actual sensory experience of hand-flying a real airplane. The vibrations, the control forces, the subtle cues you get through the seat of your pants.
Which simulators literally cannot reproduce. They can simulate motion and control loading, but the actual physics of flying through real air with real turbulence and real control surface feedback — that's not in the sim.
The FAA draft circular basically says that. The language was something like: operators should provide opportunities for manual flying during line operations because simulation cannot fully replicate the operational environment. That's a big shift from the previous position, which was basically "sim training covers it.
What does the circular actually recommend?
It's a draft, so nothing's final, but the key recommendation is that pilots should hand-fly for some portion of every flight where conditions permit. Specifically, they're suggesting manual flying during departure up to ten thousand feet and during approach from ten thousand feet down, weather and traffic permitting. They're also recommending that operators develop policies to track manual flying time and identify pilots who aren't getting enough.
This is the FAA effectively admitting that the current regulatory floor — three landings in ninety days — is dangerously inadequate.
They're not changing the regulation yet, but the advisory circular is the first step. It's the FAA saying publicly, on the record, that the industry has an automation dependency problem and the current rules aren't fixing it.
Let's talk about Lufthansa, because they've got a different angle on this.
Lufthansa runs their own ab initio flight school — they take people with zero flight experience and train them from scratch to be airline pilots. What they've found is that cadets who go straight into highly automated aircraft like the A320 don't develop the same manual flying instincts as cadets who spend time on smaller aircraft first. So they've restructured their curriculum to include more basic stick-and-rudder work before the automation gets introduced. The philosophy is: learn to fly the airplane first, then learn to manage the systems. Because once the automation becomes the default, it's really hard to go back.
There's a NASA study on this from twenty twenty-three. They compared manual flying performance between pilots who trained primarily on automated aircraft and pilots who had significant experience on conventional aircraft. The difference was measurable — the automation-heavy group showed slower recognition of deteriorating energy states and were less aggressive in correcting deviations. Not that they couldn't fly manually at all, but their responses were slower and less precise.
The NASA study also found that the skill decay isn't uniform. What decays fastest is the ability to recognize and recover from unusual attitudes and energy states. Basic straight-and-level hand-flying? Most pilots can still do that after a long automation break. But when things go wrong — when you're slow and high and the airplane starts to mush — that's where the decay shows up first.
Which maps perfectly onto both Air France 447 and Asiana 214. In both cases, the pilots could probably have hand-flown a routine approach just fine. What they couldn't do was recognize that the airplane was in a dangerous energy state and take the correct recovery action.
That's the insidious thing about automation dependency. It doesn't show up on a normal day. It shows up on the one flight in ten thousand where something unexpected happens and the automation says "your problem now.
There's this tension Daniel is asking about — between the fuel efficiency and safety arguments for maximum automation, and the manual skills argument for periodically turning it off. Let's get into that.
The fuel efficiency argument is real. Modern flight management systems optimize the flight path in ways a human pilot simply cannot. Continuous descent approaches, optimized cruise altitudes, real-time wind adjustments — the automation squeezes out fuel savings worth millions of dollars a year to a major carrier. United Airlines has publicly said their eco-friendly flight planning saves something like a hundred seventy million gallons of fuel annually across the fleet.
A hundred seventy million gallons is real money.
It's not just fuel. The automation reduces engine wear, reduces maintenance costs, optimizes everything. So when you tell an airline "hey, you should have your pilots hand-fly more," the CFO is going to ask what that costs. Hand-flying is less precise than automated flight. You're going to burn more fuel, put slightly more wear on the airframe.
The counter-argument is that the cost of one hull loss dwarfs all those incremental fuel savings.
Air France 447 was a total loss — two hundred twenty-eight people died, the aircraft was destroyed, the legal liability was enormous. If better manual flying skills could have prevented that, the return on investment in training is essentially infinite. The problem is that these events are so rare that it's hard to do the cost-benefit analysis in a spreadsheet. You're trading a certain, quantifiable fuel cost against an uncertain, unquantifiable safety benefit.
The industry has historically been terrible at valuing low-probability, high-consequence risks.
The industry's entire safety culture is built on addressing those risks, actually. But the specific risk of automation dependency has been hard to quantify because it's a latent condition. It's not a broken part you can inspect. It's a gradual erosion of skills that only becomes visible in an emergency.
What about the regulatory push in Europe?
EASA put out a safety information bulletin in twenty twenty-four that was basically their version of the same message. They recommended operators encourage manual flying during non-critical phases of flight, and they specifically called out the need for pilots to practice hand-flying without the flight director. EASA's position is that if you're always following the flight director, you're not really hand-flying. You're just acting as the servo for the automation.
That's a good distinction. Hand-flying with the flight director is sort of a halfway house. You're physically moving the controls, but you're still following the automation's guidance.
The concern is that in a real emergency where the flight director is giving bad guidance or is unavailable, the pilot hasn't practiced interpreting the raw instruments directly. They've always had that intermediate layer telling them what to do.
Let's talk about simulator sessions and recurrent training, because that's the main mechanism the industry uses to maintain skills.
The standard for most carriers is recurrent training every six to twelve months, typically two to four days in a full-motion simulator. The sessions cover normal procedures, abnormal procedures, and emergencies — engine failures, fires, hydraulic failures. And increasingly, they're including manual flying scenarios: hand-flown approaches with various failures, unusual attitude recoveries, stall recovery.
The quality of that training varies enormously by carrier and by instructor. And there's a well-documented problem that the scenarios become predictable. You know you're going to get an engine failure at V1. The startle factor isn't there.
Some carriers are addressing it with evidence-based training — instead of a fixed checklist of maneuvers, the training is tailored to the specific risks that show up in the airline's own flight data. If the data shows pilots are consistently late in disconnecting the autopilot during gusty approaches, the sim session focuses on that. It's more adaptive.
Some carriers use line operations safety audits — putting a trained observer in the jumpseat to watch how pilots actually fly on revenue flights. Not a check ride, just an observation. The data feeds back into the training program. Delta and United both have robust LOSA programs, and what they've found is that even experienced pilots sometimes show hesitation when the automation does something unexpected. Mode confusion — where the pilot thinks the autopilot is in one mode but it's actually in another — is one of the most common findings.
Asiana 214 was partly mode confusion about the autothrottle state. The pilot thought the autothrottle was active and protecting the speed. It wasn't. The mode annunciator on the 777 flight display was showing the correct state — the information was there — but the pilot's mental model didn't match reality.
Which brings us to Daniel's last question — whether the next generation has the same baseline skills. And I think the answer is complicated.
The data from ab initio programs like Lufthansa's shows that cadets who go straight into highly automated aircraft do develop different skills. They tend to have excellent systems knowledge and automation management skills, but their raw stick-and-rudder abilities are weaker than pilots who spent time in general aviation or in the military flying less automated aircraft.
There's also a counterpoint. The younger generation has better systems management skills. They're more comfortable with the automation, they understand the logic of the flight management computer more intuitively, and they're less likely to fight the automation or distrust it inappropriately.
That's fair. The older generation has its own automation problems — cases where older pilots disengage the autopilot unnecessarily because they don't trust it, and then make errors the automation would have prevented. The sweet spot is a pilot who can hand-fly competently and also manage the automation effectively. It's not either-or.
The term that keeps coming up in the research is "manual flight proficiency" as a distinct skill from "automation management proficiency." They're different things, and you need both.
The industry is slowly waking up to the fact that the current training and regulatory framework doesn't adequately maintain the manual flight proficiency side. The FAA draft circular, the EASA bulletin, Cathay's internal mandates — these are all recognition that the pendulum swung too far toward automation and needs to come back a bit.
What's the practical takeaway for someone who's not a pilot? Because most of our listeners aren't going to be hand-flying an Airbus anytime soon.
The broader lesson is about automation dependency in any high-stakes domain. The same dynamic shows up in medicine, where physicians who rely too heavily on clinical decision support systems can lose diagnostic skills. It shows up in any field where automation handles the routine stuff and humans are supposed to handle the exceptions. The problem is that handling exceptions requires practice with the basics, and if the basics are always automated, the exceptions become unmanageable.
There's a human factors principle worth naming. Automation doesn't remove the need for human skill — it changes the nature of the skill required. And if the training system doesn't adapt to that change, you get a gap. The pilot doesn't need to hand-fly a ten-hour cruise segment. But they do need to recognize a stall when it happens and know what to do about it. And those two things are connected in ways that aren't obvious until the emergency happens.
The other practical thing is that the industry is moving toward more manual flying on revenue flights, but it's going to be gradual. The FAA circular is a draft, not a rule. EASA's bulletin is a recommendation, not a mandate. The airlines that are serious about this — Cathay, Lufthansa to some extent — are doing it voluntarily because their internal safety data convinced them it was necessary.
The ones that aren't doing it are betting that their sim programs are good enough and that the next Air France 447 won't happen on their watch.
Which is a bet that has worked out for most carriers most of the time. The question is whether it keeps working as the pilot population shifts toward people who have never flown an airplane without a glass cockpit and a flight management computer.
I think that's the thing that keeps safety people up at night. The current generation of training captains and check airmen grew up on steam gauges and hand-flying. In ten or fifteen years, the training captains will be people who trained entirely on glass. The institutional knowledge of what good hand-flying looks like will be a generation removed from the raw stick-and-rudder era.
There's actually a study from Embry-Riddle on this — they looked at instructor pilots and found that even the instructors were showing signs of automation dependency. The people who are supposed to be teaching manual flying skills were themselves less proficient than instructors from twenty years ago, because they spend most of their time teaching systems management.
That's a compounding problem. If the instructors are less proficient, the next generation of pilots will be even less proficient, and so on.
That's exactly the spiral the FAA and EASA are trying to interrupt with these new policies. They're trying to put a floor under the skill degradation before it becomes irreversible.
To actually answer Daniel's question directly — how do pilots ensure their skills don't atrophy? The honest answer right now is: it depends on the airline and the individual pilot. The regulatory floor is low. Three landings in ninety days, no hand-flying required. But the better airlines are going well beyond that. Cathay mandates multiple hand-flown approaches per month. Lufthansa is building more stick-and-rudder work into initial training. Delta and United invest heavily in sim-based manual flying scenarios. And the regulators are pushing, slowly, for more hand-flying on revenue flights.
The individual pilot matters too. There are pilots who make a point of hand-flying whenever conditions permit, who stay current in general aviation on their days off, who take their recurrent training seriously as a skill-building opportunity rather than a checkbox. And there are pilots who do the minimum. The system doesn't currently force the difference to matter — until it suddenly does.
The system is designed for the median pilot on a normal day. The problem is that the emergencies where manual skills matter happen to individual pilots on abnormal days. And there's no way to predict which pilot, which day.
That's the core of it. Automation dependency is a population-level risk that manifests as individual-level catastrophes. The airline industry has gotten so safe that the remaining accidents are increasingly about human factors rather than mechanical failures. And automation dependency is one of the biggest human factors risks on the board right now.
I keep thinking about those seventy-five stall warnings on Air France 447. The airplane was telling them exactly what was wrong, repeatedly, for minutes. And the pilots couldn't hear it because their mental model of what the airplane was doing had become completely disconnected from reality. That's not a training failure in the narrow sense. That's a fundamental breakdown in the relationship between human and machine.
The BEA report called it "de-structuring of the task." The pilots were confronted with a situation their training hadn't prepared them for — manual aircraft handling at high altitude with unreliable airspeed — and their response wasn't to fly the airplane using fundamental principles. It was to try to make sense of the automation's behavior. They were troubleshooting the systems instead of flying the airplane.
Aviate, navigate, communicate. The oldest rule in aviation. And they got stuck on the second and third steps before securing the first.
Which is exactly why the new training approaches emphasize "fly the airplane first." When the automation disconnects, your job is to fly the airplane. Not to figure out what went wrong, not to run checklists, not to call air traffic control. Point the nose at the horizon, keep the wings level, manage the energy. Everything else can wait.
That's a skill that only stays sharp with practice. You can't sim your way to instinct. You have to actually do it.
That's the FAA's position now, and I think they're right. The sim is an incredible tool, but it's not the same as flying through real air. The control forces are different, the motion cues are different, the consequences of getting it wrong are different. You can practice a stall recovery in a sim a hundred times, but you know you're in a sim. In the airplane, at thirty-five thousand feet, in the dark, over the ocean, the stakes feel different.
Where does this go from here?
I think we'll see the FAA finalize that advisory circular, probably with some modifications based on industry comments. I think we'll see more airlines follow Cathay's lead and implement their own manual flying minimums. And I think we'll see the next generation of aircraft designed with more graceful automation degradation — systems that hand back control more smoothly and give the pilot more time to establish manual control.
There's also a push from some safety advocates to require manual flying time as part of the pilot's logbook, the same way we track instrument time and night time. A separate column for hand-flown time, with minimums.
That's an interesting idea. It would force the issue in a way that advisory circulars don't. If the regulation said you need X hours of hand-flown time per year to stay current, airlines would have to build that into their operations. Right now it's all discretionary.
The pushback from the industry would be cost. Hand-flying burns more fuel, it's less efficient, and it potentially increases workload during already-busy phases of flight.
The counter to that is what we said earlier — one hull loss wipes out decades of fuel savings. The economics of safety in aviation are always about preventing the extremely rare, extremely expensive event. The challenge with automation dependency is that it's hard to prove the counterfactual. You can't prove that a specific pilot's hand-flying skills prevented an accident, because the accident didn't happen.
Whereas the fuel savings from automation are visible on a spreadsheet every quarter. That's the fundamental asymmetry that makes this hard to regulate.
It's the same problem with a lot of safety regulation. The costs are visible and immediate; the benefits are invisible and deferred. And in aviation, the baseline safety level is already so high that the marginal benefit of additional regulation is genuinely hard to measure.
The FAA draft circular is a signal that the regulator thinks the risk is real and growing. They wouldn't put out something that blunt if they weren't seeing concerning trends in the data.
The data they're seeing is probably from flight data monitoring programs and ASAP reports — the Aviation Safety Action Program where pilots can voluntarily report safety issues without fear of punishment. There's been a steady increase in reports related to automation mode confusion and manual flying difficulties. The FAA sees those reports, and they're connecting the dots.
The system is working, in the sense that the problem is being identified and addressed. It's just slow.
Aviation safety is always slow. The industry doesn't move fast on anything, and that's mostly a good thing. You don't want rapid, untested changes to how pilots are trained and evaluated. But the pace has to be fast enough to stay ahead of the risk, and right now I think there's a legitimate question about whether it is.
Daniel's question about the next generation might be the one that really matters long-term. In twenty years, nobody flying airliners will have touched a steam gauge. The institutional memory of pre-automation flying will be gone. If we haven't figured out how to maintain manual skills in an automated world by then, we're going to have a problem that's much harder to fix.
The optimist's view is that the automation itself will get good enough that manual skills become unnecessary — that the aircraft will handle every conceivable failure mode without human intervention. We're not there yet, and we're not close. The automation still disconnects when things get weird. The pitot tubes ice over, the automation says "your airplane," and someone has to fly it.
The pessimist's view is that we're training a generation of pilots who are excellent systems managers and mediocre aviators, and we won't find out until the next Air France 447.
I'm somewhere in the middle. I think the problem is real but solvable. The Cathay approach — mandating a few hand-flown approaches per month — is not expensive or complicated. The Lufthansa approach — building more stick-and-rudder work into initial training — is sensible. The FAA and EASA are paying attention. The system is self-correcting, it's just doing it at aviation speed.
Which is to say, slowly and carefully, with a lot of paperwork.
As it should be. The worst thing would be an overcorrection that forces hand-flying in conditions where it's less safe. There are times when the automation should be flying — low visibility approaches, for example. The skill is knowing when to hand-fly and when to let the automation do its job.
That's the judgment piece. And judgment, unlike stick-and-rudder skills, comes from experience. You can't sim your way to good judgment. You have to see a lot of different situations and learn from them.
Which is why the best pilots I've known — and I've known a few, from my medical days doing aviation medical exams — the best ones treat every flight as a learning opportunity. They hand-fly when they can, they stay curious about the systems, they read the accident reports, and they never assume they've got it figured out.
The ones who scare me are the ones who think the automation makes them invincible.
That's the hazardous attitude the FAA calls "invulnerability." It's one of the five hazardous attitudes they teach in primary flight training. The antidote is "it could happen to me." And Air France 447 and Asiana 214 are the proof that it can.
We should probably thank our producer Hilbert Flumingtop for keeping us on track here, and Modal for making the whole pipeline work.
By the way — DeepSeek V four Pro is writing our script today. Doing a solid job so far.
Seems to know its aviation. This has been My Weird Prompts. You can find every episode at myweirdprompts dot com or wherever you get your podcasts. We'll be back next time.
Until then, keep your hand-flying skills sharp.
