August 18, 2026 | Read online

The autopilot did its job. That's what killed them.

What a 2009 plane crash can teach late-career professionals about handing judgment to AI.

Welcome back to the Launch Key 🚀

With a weekend road trip that took me from Eastern North Carolina tobacco fields to the rhododendron of the high country, I asked a friend to guest author Launch Key.

George Pillari and I met because of our late-career newsletters. But that’s a minor detail in his career portfolio. George has raised several billions of capital in his career and has more than 30 years of experience in scaling, selling, merging and acquiring companies. He’s written a couple of books and shares real-world strategies and historical stories in his newsletter to help challenge conventional wisdom.

His post is a wake up call for people trusting their AI.

Let's get into it.

Read online

Gmail users may wish to read online since some parts may be clipped.
Take this week’s poll or comment below and let me know if we’re on the right track.

Pull to Eject
Old School Wisdom
Modern Tools
Free Knowledge
Recommendations
Visual Crapshoot

Pull to Eject

George Pillari shares stories and wisdom in his newsletter - the Cautionary. More than 30,000 subscribers tune in for insight like this:

June 1, 2009. Air France 447 is over the Atlantic at 35,000 feet, cruising through the night on a routine flight from Rio de Janeiro to Paris. The captain is two rows back, asleep. Two co-pilots have the cockpit: Pierre-Cédric Bonin, 32, and David Robert, 37.

A few ice crystals, smaller than a grain of rice, block the plane's speed sensors for less than a minute. The autopilot can't get a reliable airspeed reading, so it does exactly what it's built to do. It disconnects and hands control back to the humans.

An alarm sounds. It will sound 75 times before the flight ends.

Bonin pulls back on the side stick. That's the wrong move. Pulling back raises the nose and bleeds off speed, and the plane is already losing speed. Within a minute it stalls: the wings stop generating lift, and the plane starts falling, nose high, engines running, at nearly 10,000 feet a minute.

Neither pilot recognizes the stall. Robert says, at one point, "The plane has suffered no mechanical malfunction. Aside from the loss of airspeed indication, everything is working fine." Nothing was broken. The plane was doing exactly what a stalled plane does. Nobody in that cockpit had ever hand-flown one at cruise altitude, because nobody ever does. The autopilot always has it.

Four minutes and 23 seconds after the autopilot let go, AF447 hit the ocean. All 228 people on board died.

The investigation traced the crash to a training gap. Neither co-pilot had ever practiced flying with unreliable airspeed at altitude, because the scenario was considered too rare to spend simulator time on. The plane’s AI system had become so good at flying the plane that the people paid to fly it had stopped practicing the part where they actually had to do something.

Air France 447 - Metron

Air France Flight 447 disappeared during a flight from Rio to Paris. Metron produced the probability map that guided searchers to the wreckage.

Metron

Aviation had been warned. Back in the 1990s, an American Airlines training captain named Warren VanderBurgh started cautioning younger pilots about what he called "children of the magenta line," a reference to the pink flight path drawn on the cockpit display. His point: a generation of pilots was learning to follow the line on the screen instead of learning to fly. The name stuck because the problem kept showing up.

Four years after AF447, it showed up again. Asiana 214 came in low and slow on approach to San Francisco in July 2013. The pilots believed the autothrottle was holding their airspeed. It wasn't. Nobody caught it in time. The plane hit the seawall short of the runway. Three people died.

That same year, the FAA convened a working group to study what automation was doing to pilots. Its finding, in plain language: pilots who spend most of their flight time supervising automation lose the manual skill, and the instinct, to take over when it fails. Autopilot has made flying far safer over the last 50 years, and nobody wants to go back to hand-flying every leg of every trip. The panel's recommendation was narrower than that. Keep the automation and make sure pilots still practice the skills.

This is the state of AI in one of our most advanced industries.

The terms we need to get used to hearing in AI discussions are: human-in-the-loop, human-on-the-loop, and human-out-of-the-loop. An airplane’s autopilot system is a human-on-the-loop design: the AI does everything, and the human monitors and intervenes when necessary.

Sounds good on paper, but the problem is the conundrum of skills erosion: the better the machines get at doing things for us, the worse we get at doing them for ourselves. Pilots 50 years ago did not have worry about AI flying the plane, since they were “real” pilots who could fly technically, but also developed a feel for the plane. This does not happen anymore.

You're not flying a plane. But if AI is writing your first drafts, building your financial models, running your due diligence, or drafting the market research behind a decision, you're running a version of the same experiment.

Most of the time it works. AI catches things you'd have missed, moves faster than you would have, and frees up hours for the judgment calls that need you.

The failure doesn't show up in the 95% of cases where it works. It shows up the one time it's confidently wrong, after you've trusted the output long enough that you've stopped checking for the difference. A financial model with one bad assumption baked in reads exactly like a sound one until the numbers come due. A diligence summary that missed the one clause that mattered looks just as clean as one that didn't.

Bonin had a stick in his hand and 75 alarms telling him something was wrong. He still couldn't diagnose a stall because he'd never had to. That's what skill erosion looks like from the inside: not a gap you notice, but a gap you can't see, because the tools you would use to notice it are the things that atrophied.

If you're building something new later in your career, a company, a fund, an advisory practice, your judgment is the whole product. It's the reason a client hires you instead of a chatbot. Let AI take the busywork. Just keep doing the hard parts yourself often enough that you'd still recognize wrong if the tool got there first.

Everyone serious is already using AI.

Here's the harder question: could you still fly the plane if it handed you the stick tomorrow?

Now go launch something 🚀

❝

The future is already here; it's just not evenly distributed.

William Gibson

CrowdHealth - Healthcare Crowdfunding

Tired of health insurance policies that are confusing and expensive? Join CrowdHealth and finally break free from the insurance system. CrowdHealth is not health insurance. It's an innovative model to use and pay for health care without insurance middlemen and government influence.

Old School Wisdom

We share an audience of experienced professionals with complimentary messages. Where I say that our experience and judgement is our differentiator, George positions experience-without-AI-fluency as a specific vulnerability. And he shows how to improve that in his new book.

No Adult Left Behind

The split between AI-fluent professionals and AI-observers is widening faster than any previous technology adoption curve, and the experienced operators most at risk are precisely the ones whose existing competence makes the urgency feel optional. The fix is not technical mastery; it is building a habit.

Modern Tools

ChatGPT — you've used it. The question is whether you're briefing it or interrogating it.

Most people type a question and take the first answer, which is roughly like hiring a consultant and then not telling them anything about your business. The upgrade costs nothing: give it the context first — who you are, what you've done, who the client is, what you're actually trying to decide — and then ask. Same tool, different order, noticeably better output.

The other habit worth building is asking it to argue against you. "Here's my plan, make the strongest case that it fails." It's more useful as a sparring partner than an oracle, and it's the one job where you don't need to trust it to get value from it.

Free tier handles all of this.

ChatGPT

ChatGPT helps you get answers, find inspiration, and be more productive.

ChatGPT

Free Knowledge

Pillari argues that a quiet professional split is happening between people who reach for AI as a daily instrument and people who plan to learn it "when things settle down," and that the gap is widening faster than email or the smartphone wave — fast enough that experienced mid- and late-career professionals are at particular risk of being stranded by it.

He has created a Learning AI series in his newsletter, The Cautionary.

The Cautionary

Learn to challenge conventional wisdom.

The Cautionary • George Pillari

Recommendations

📕 MGMT Playbook : Practical management insights straight to your inbox every Wednesday.

🏢 Retirepreneur : Your guide to turning decades of experience into profitable consulting, coaching, or course businesses, without risking your retirement security.

🗃️ Dealroom Business Success Uncovered : Learn directly from billionaire entrepreneurs on how to grow a business. Join a community of 2,000+ innovators.

🧾 Redefining Retirement : Too Young to Retire, Too Old To Take Orders? Yeah, Me Too.

Visual Crapshoot

You've Spent 30 Years Building Expertise. It's Time to Turn It Into Income That Works for You.

The complete framework for professionals 50+ who want to build multiple revenue streams without quitting their job or betting everything on a startup.

Help me help you: What's the #1 challenge holding back your late career side hustle?

I'm creating resources specifically for late-career entrepreneurs. Which topic would help you most?

Share Launch Key | The Newsletter for Late-Career Entrepreneurs

Thanks for referring the Launch Key newsletter! 🚀

You currently have 0 referrals, only 1 away from receiving Free tools to launch your new idea or digital product. Instant PDF download with 1 referral.