Why two people still sit up front
Autopilot has flown the cruise portion of airline flights for decades. That is the part of the job most people picture when they ask whether pilots will be replaced by AI. It is also the part that was easiest to automate, because cruise flight is stable, predictable, and well described by physics.
The rest of the job is less tidy. Pilots run pre-flight inspections and check the aircraft and its paperwork before pushback. They brief the crew, talk to air traffic control, and decide whether to accept a runway, a hold, or a diversion when weather closes in. When a system fails, the crew works the checklist, judges what the failure means for the rest of the flight, and owns the outcome. Software can suggest. Under current rules, a licensed human signs for the decision.
So the honest picture is task erosion, not a cockpit without people. Flight management systems, autoland, and alerting displays such as EICAS already absorbed chunks of the monitoring and navigation work. What is left leans toward judgment under pressure, coordination with other humans, and legal responsibility. That is why the Still needs a human figure for this job sits at 76 out of 100 (higher is safer).
What software does, what it assists, and what crews keep
AI and conventional automation already handle part of the work end to end. Route navigation in cruise, altitude and speed holds, and continuous engine and systems monitoring run largely without hands on the controls. Across the full task list, that fully handled share comes to 0% of task time.
A second group is assisted work, where the tool drafts and the pilot approves. Fuel and route optimization, weather and turbulence forecasting, and performance calculations are produced by software, then checked against what the crew can see out the window and on the radar. Assisted tasks account for 43% of task time. The skill shifts from calculating to catching a bad answer.
The remaining 57% stays with people: non-normal and emergency handling, go or no-go calls, taxiing and ground maneuvering in congested airports, negotiating clearances with controllers, and commanding a crew and cabin when something goes wrong. The task list above shows which group each duty falls into. Our overall coverage score, which measures how much task time AI can handle today, reads 21; the coverage method page explains how that is built.
What the evidence actually tests
Plenty of systems have been tested in flight: autoland, flight management computers, military autonomy programs, and large uncrewed aircraft. What has not happened is a published, head-to-head test of an AI system against a typical qualified airline crew across the whole job, including degraded aircraft and unexpected weather.
That is why the quality-parity evidence grade here is D, and why we publish no parity number for this occupation. A D grade means not measured, not tied. Read how that grade works on the quality-parity method page.
What would settle it: certified line operations where a reduced crew or an autonomous system flies revenue flights with published safety data; simulator studies comparing crew and automated handling of the same failure sets; and a regulator signing off on reduced-crew operations for passenger service. Until that exists, claims in either direction are projections, not results.
Good to know: airline cockpit changes arrive through certification, which means years of documented testing before anything flies with passengers.
When single-pilot operations could change the job
Most likely after 2044 (8 in 10 of our scenarios). We publish a median with its range rather than one date, because the pace depends on rules and hardware as much as on models; the replacement-year method page explains what the window covers.
Two things could pull it earlier. Cargo and regional operations are the likely testbeds for reduced-crew flying, and a credible safety record there would move the regulatory conversation fast. Second, the cost gap is wide: running software on these tasks is cheap next to crew hours, which keeps operators pushing.
Two things hold it back. Certification and liability are slow by design, and no regulator has approved single-pilot passenger operations in the United States. And a meaningful part of the job is physical. Hands on the yoke and thrust levers, overhead switch panels, walkarounds, and manual handling in gusty crosswinds need hardware, not just a model. Our robotics read for this job sits in the dexterous humanoid tier, which is the hardest and least mature class of machine. The guide on humanoid robots and physical jobs covers why that tier lags.
Demand gives some context too. The BLS counts 103,560 airline pilots, copilots, and flight engineers in the United States, projects 8.3% employment growth from 2025 to 2035, and puts median pay at $232,140 (BLS, 2025). Those are not the numbers of a shrinking occupation.
How to stay needed in the right seat
Lean into the work that sits in the human column. First, non-normal and emergency handling: know the failure modes cold, and practice the ones that rarely come up. Second, command and crew coordination, including briefing, delegating, and running a cabin problem. Third, judgment calls on weather, fuel, and diversions, where the data is incomplete and someone has to choose.
Two skills are worth adding. Learn to supervise automation well, which means knowing what each system does, how it fails, and when to turn it off. And get comfortable reading the outputs of optimization and forecasting tools critically, so a bad fuel figure or a stale weather product does not go unchallenged.
Nearby work is worth comparing if you are planning a path. The closest jobs are Commercial Pilots, who fly charter, cargo, and survey operations, and Air Traffic Controllers, who own the other half of every clearance. Aviation Inspectors sit on the compliance side of the same aircraft. You can also browse the air transportation workers family, the wider transportation and warehousing sector, or our list of jobs that mostly need a person.
Want a side-by-side read? Put this job against another on the compare tool, or see how every score is built on the methodology page.