The last flight you were on almost certainly flew itself for part of the trip. Cruise, usually. Sometimes the approach. Nobody makes an announcement about it and most passengers never think about it, because whatever the automation was handling, there were two trained pilots sitting up front the whole time.
That arrangement is older than most people guess. In June 1914, a crowd at a French aviation safety competition watched a 21-year-old American fly past the reviewing stand with both hands in the air. His mechanic climbed out onto the wing. The aircraft held its line anyway.
That was Lawrence Sperry showing off the gyroscopic autopilot he had built two years earlier, nine years after the Wright brothers got off the ground at Kitty Hawk. Automation has been flying aircraft for about as long as there have been aircraft worth flying, and you have been trusting it for years without giving it a thought.
Here is the part worth sitting with. In the century since, autopilot got good enough to cross an ocean and land in weather no person can see through, and the industry never once decided the pilots were optional. Every commercial flight you have taken carried two of them, by law.
A hundred years of trusted automation, never left alone
In 1933, Wiley Post flew around the world by himself in seven days and change, with a Sperry autopilot doing most of the actual flying. Today the automation can handle takeoff, cruise, and in the right conditions the landing too. On paper, the case for taking the humans out has never been stronger.
The industry went the other way, and it is worth being precise about why. It is not that the technology is weak. It is that the technology is good enough to be worth supervising. The most safety-obsessed industry on earth looked at a century of data and landed on: keep the automation, keep the expert, and spend the money training the expert to work with it.
Then the industry found the failure mode
In 1997 an American Airlines training captain named Warren Vanderburgh gave a lecture that still gets passed around cockpits almost thirty years later. He called it Children of the Magenta Line.
The magenta line is the route the flight computer draws on the cockpit display. Vanderburgh’s argument was that crews had become excellent at managing that line, and quietly worse at flying the aircraft. They were programming and monitoring, which is a different skill from flying. When the automation did something they did not expect, the hands came back slowly, and sometimes not at all.
Twelve years later, Air France 447 went into the Atlantic after the autopilot disconnected following a sensor failure and the crew did not correctly recognize what the aircraft was doing. 228 people died. The investigation is long and the causes are layered, and this is not the place to flatten it into a marketing point. What matters here is the response: nobody proposed removing the automation. The industry doubled down on training people to stay ahead of it.
The same thing is happening in offices right now
More than 80 percent of organizations using generative AI report no tangible impact on earnings from it (McKinsey, State of AI 2025). That gap is rarely about the model being bad. It is about nobody owning what the model is doing.
Picture a team that starts using AI to draft customer quotes. Month one, it is a marvel. Hours come back. Month four, a quote goes out with a margin assumption nobody checked, because the person who was reviewing every line has settled into skimming. That is the magenta line, in a quoting workflow, in a building with no runway anywhere near it.
The tell is always the same, and you can hear it in a meeting. Somebody asks why the system did what it did, and the honest answer in the room is a shrug.
What aviation did instead of switching it off
Three habits came out of all this, and all three translate to a business without much squinting.
- Know what the system is doing right now. In a cockpit that means calling out the mode out loud. In a company it means somebody can say, in one sentence, which decisions the tool makes on its own and which ones it only drafts.
- Know where it gets things wrong. Every system has a shape to its mistakes. Write yours down as you find them. A list of six real failures beats any vendor documentation you will ever be sent.
- Be able to take over. Not in principle. Somebody on the team did this job by hand recently enough that they could do it again this afternoon.
What this looks like in a 40-person company
None of it requires a platform, a governance committee, or a budget line. For most of the organizations we work with in Halifax, Moncton, and St. John’s, it comes down to four unglamorous things.
- Run the new tool on real work for a few weeks, beside the person who does that work today. Vendor demo data has never once predicted how a tool behaves on your messiest file.
- Decide up front what a person signs. Anything with money, safety, or somebody’s reputation attached goes through a human before it leaves the building.
- Keep a one-page runbook: what it does, where it fails, how to turn it off. One page. If it runs to twelve, nobody will read it in the moment they need it.
- Spend thirty minutes a month looking at what it got wrong. This is the whole discipline, and it is the first thing that gets skipped.
If you want a sense of whether your organization is ready for any of this, our piece on transformation readiness covers the groundwork that has to be in place first.
Five questions before you switch something on
- Which decisions is this making on its own, and which is it only drafting for someone?
- Who signs the output when real money is attached to it?
- What does wrong look like, and would we catch it in week one or in month four?
- If the vendor tripled the price or shut down on Friday, what do we do Monday?
- Who here could still do this by hand?
If the answers come easily, you are in good shape. If the room goes quiet on the last two, that is useful information and it cost you nothing to find out.
The point
Aviation has the longest relationship with automation of any industry on earth, and the best safety record. Those two facts are held together by one thing: somebody in the seat who is still paying attention, on the flights where nothing happens as much as the ones where something does.
Your AI does not have to be smarter than that. It just needs somebody flying it.