Think about the best person on your team, and where they actually learned what they know. It almost certainly wasn't a training program. It was some stretch of tedious, repetitive work (the reports nobody read, the calls nobody wanted) that you'd automate tomorrow if the right vendor showed you the right demo.

That matters, because the costliest automation mistake happening right now doesn't look like a mistake. Stanford's Digital Economy Lab tracks payroll data for roughly one in six American workers. It finds that employment for 22-to-25-year-olds in the most AI-exposed occupations is falling more than 4% a year. Workers over 30 in the same jobs are holding steady. The experienced people are fine. It's the pipeline that's disappearing.

In 2017, Elon Musk set out to build the most automated car factory on Earth and promised 5,000 Model 3s a week by year's end. In the entire third quarter, Tesla built 260 cars. Not 260,000. Two hundred and sixty.

It's tempting to call that ego. It wasn't. Musk is one of the most rigorous manufacturing minds of his generation, which means the trap isn't about intelligence. It's structural.

Lisanne Bainbridge explained it in 1983, in a five-page paper called "The Ironies of Automation." When you automate, you start with the predictable parts. Whatever's left for the human is everything the system couldn't handle. Automation doesn't shrink the human role. It distills it down to the hardest part. The second irony is worse. People used to build the skill for those hard moments by doing the routine work you just automated, and monitoring isn't practice. Matt Beane at UC Santa Barbara has watched this happen in operating rooms. The surgical robot works, but the resident who watched instead of operating walks out no better a surgeon. Nothing breaks. You just stop producing the people who could do the work without the machine.

Tesla dug out by tearing out automation and putting people back. By mid-2018 it was building 5,031 cars a week, partly on a line assembled in a tent in the parking lot. But it could reverse course that fast because the skill was still in the building. More recently, Ford ran into a similar problem not by design...but by neglect.

Overtime, Ford's most experienced engineers retired or left for suppliers, which is ordinary attrition. Everyone quietly assumed AI quality tools would cover the gap. They didn't. Quality problems cost billions, and when Ford went to improve the AI, it found the tools needed its most experienced engineers to train them. The people the AI needed were the ones who'd left.

So Ford bought them back: 350 veteran engineers over three years, many of them former employees. They now run design reviews, retrain the AI, and mentor younger engineers. In the 2026 J.D. Power Initial Quality Study, Ford was the top mainstream brand, up from tenth the year before. It isn't a tidy story, since Ford's quality issues predate the AI push. But nobody decided to trade expertise for software. It just drained away.

You don't need Musk's factory or Ford's headcount to act on this. You need two questions:

• Ask what a task teaches before you automate it. Ask "Who does this today, and what do they learn by doing it?" If the answer is nothing, automate it this afternoon. If it's where someone's judgment quietly gets built, you're removing a classroom, so decide on purpose where the new one will be.
• Test what you've already automated. Ask "If this tool went down tomorrow, who could do it by hand, and how long since they last did?" If the name that comes to mind is within five years of retirement, you don't have a staffing plan. You have a countdown.

You can un-automate a process. You can't un-automate a pipeline. The real risk isn't that AI takes the job. It's that AI takes the training ground. I walk through the full Tesla and Ford stories in this week's video, worth a watch if there's a task in your sights right now.

or listen to the podcast​

Here's some more resources I recommend:

Small talk isn't the problem: Most of us were never taught how conversation actually works, so we default to rehearsed lines or overanalyze every exchange afterward. In Conversation, behavioral researcher and my good friend Vanessa Van Edwards offers a science-backed playbook for the unspoken rules of connection: how to break the ice, tune in to the person in front of you, and end a conversation without damaging the relationship. It never asks you to sound like someone you're not. If you lead people, relationships are built one interaction at a time. This book helps you make each one count. Check it out here.​

What Not To Delegate to AI: Speaking of AI and what NOT to do, last week I published a video focused on managers. AI can summarize meetings, organize information, draft emails, and analyze data. That's a good thing, because it takes a lot of administrative work off your plate. But efficiency isn't the same thing as leadership. As AI becomes more embedded in the workplace, the parts of management that stay human matter more, not less. In this video, I break down the four parts of management I would never hand to AI, and why. If you're figuring out where AI fits on your team, start here. Click here to watch.​

Getting Teams Aligned: Most dysfunction in organizations comes down to one thing: a gap between the big-picture strategy and what individual teams are actually incentivized to do. In my LinkedIn Learning course, I show you how to close that gap—using your organization's purpose to set team priorities that genuinely connect to the broader mission. You'll learn how to organize and prioritize your team's work, test your goals against real success criteria, and build a communication rhythm through regular huddles that keeps everyone focused, even when conditions change. When every team is aligned, everybody wins—teams, leaders, customers, and stakeholders alike. Click here to get started.​