What is it even supposed to mean?
Airplanes have had autopilot (the genuine kind, not Musk's snake oil) for ages now. Commercial airlines have been using autoland on well-equipped airports for decades. Garmin's fully autonomous emergency autoland has already saved a few Cessna owners' lives. With the ongoing adoption of CPDLC the ATC-to-pilot link is also actively being automated and standardized.
There are no big technical hurdles left to solve! The main thing preventing fully-automated flight from taking off is the industry and regulators (rightfully) being incredibly conservative, and preferring paying pilots over the horrible PR fallout of an incident aboard an automated flight killing hundreds of people. Artificial intelligence isn't going to be of any help here!
You seem to be confused about the reasons. Fully automated flight is technically feasible (with the right equipment) under normal conditions. However, it will never be a replacement for a pair of human pilots for Part 121 commercial passenger flights — at least not without some fundamental advancement in AGI. The problem is not with routine flight operations but with emergency procedures. By their very nature, emergencies can't be anticipated and it's impossible to write code in advance to handle them. Whereas humans can improvise solutions on the fly based on first principles. A prime example is US Airways Flight 1549 "Miracle on the Hudson" where the pilots intentionally disregarded parts of the emergency checklist to safely ditch in the water.
The Garmin Autonomí autoland system is an amazing technical achievement but it's intended as a last-ditch way to save the passengers when a Part 91 single pilot is incapacitated. It would never be approved for routine non-emergency use and can't even take VHF radio instructions from controllers.
I already covered that part.
Pilots are supposed to rigidly follow their checklists - to the point of having to execute some of them from memory. A computer could easily do the same for the vast majority of emergencies.
The "Miracle on the Hudson" is called a miracle for a reason: the expected outcome is for everyone to die. Same with UA232. Nobody would've blamed the pilots if there had been zero survivors! Even with human pilots it is the exception, not the rule.
The fact that you are willing to blame an autopilot for not improvising a 1-in-1000 Hail Mary attempt is exactly the conservatism I mentioned: we prefer routine human failure and the occasional miracle over predictable automated performance.
The technical hurdle to solve they feel is getting the barrier to 'flying' dumbed down even more. They want this to be something that Joe off the street can get in with minimal flight training and go zip around in a high performance jet once their vesting clears and they can cash out a few mill.
So... basically, an even more digital cockpit with more touchscreens and less verbatim information presentation on the screens. Why give you multiple engine gauges for N1, N2, temps, etc, when we can just give you one dumb "Thrust" gauge? Why make programming the autopilot a fifteen day course on the ground when you can just have a LLM figure out what your flight plan should be and punch it all in automatically?
It's like how Cirrus positions themselves to be the family SUV of the skies with their products and falls back on "just pull the chute / push the Autoland button, bro".
How's that going to solve the FAA-mandated training requirements?