It was the Spring of 1977. I was 10 years old, making my way through the social minefield that is 4th grade. At that time of year, social standing for boys centered on Little League baseball.
I was not a particularly athletic child. My hand-eye coordination was probably below average. I was a physically awkward aspiring nerd who spent most of his spare time alone with my nose in a book.
But I enjoyed baseball and aspired to be a good baseball player. I was not. Placed in right field where the fewest balls got hit. Perennially stuck in the 8th or 9th batting slot reserved for kids whose most likely path on base was getting hit by a pitch.
I attended every practice and even came early or stayed late for extra batting practice. My coach was patient and thoughtful. He taught me how to have a proper stance, how to shift my weight, and how to hold the bat. The problem was my swing. It was comically awkward. When I did make contact, the usual outcome was a miserable pop fly that never left the infield.
My coach was brilliant in a way I only appreciated much later. After a Saturday practice, he suggested talking with my Dad. I remember the gist of what he said. "Swinging the bat well cannot be taught. It must be trained."
The problem was simple. With 9 kids on the team, each got only 10 or 20 pitches per practice. And those pitches, whether from the coach or a teammate, came with natural variance. I was not getting enough training cycles. I was dealing with too many variables at once.
He suggested my Dad get me a Johnny Bench Batter-Up training rig. We went directly to the store (I think it was Modell's, for those who remember the chain), bought the rig, and mounted it in our backyard.
The device was genius in its simplicity. A vertical pole. A semi-flexible horizontal arm that rotated around it. A baseball mounted at the end. Thick rubber bands connected pole to arm so that when you wound it up, the ball swung around the pole.
Hit the ball dead level and the arm whizzed around several times before reversing course, coming back on a nearly perfect, predictable trajectory. Hit off center, up or down, and most of the energy went into making the arm wobble, with limited travel around the pole.
The tension in the bands was just right. After a good hit, you had just enough time to recover to a ready stance before the ball completed its last return at maximum speed. The better you hit, the more rotations, the less wobble. I quickly learned that a square, level swing mattered far more than bat speed.
I stayed out that first day until the sun went down, probably close to 8pm. I spent most of the next day going at it. Roughly a hundred swings an hour. Immediate feedback on every one.
Whiff!..reset…Whiff!…reset…Biff!…the arm clattered up and down with almost no rotation.reset…Biff!….wobblereset.
Thwack!! Two or three rotations — and on the return….Whiff! Damn.
Then a miracle. Thwack!!!…3 rotations….reset….here it comes….Thwack!!!….3 rotations…..reset.Thwack!!!….2 rotations and a little wobble….Biff!
OK. That was rewarding. What I did not know then is that my body was learning tempo and muscle memory.
I set a goal. Ten good hits in a row. No whiffs or biffs. I got to 7 on the second day. Within two weeks I could stand at that rig and grind out Thwack after Thwack until I got tired or lost concentration. Not every shot was perfect. But I rarely biffed badly enough to get fewer than two full rotations. Sometimes I got three and a half. Then four.
Our next game was a Tuesday evening. I stood confidently in the batter's box ready to use my new skill. Three pitches. Three swings. No contact.
My coach pulled me aside. "Your swing is a lot better! But you have to wait for a good pitch! You were chasing junk!"
Next at bat, I waited for "my pitch." Foul ball. Solid contact. Then a grounder to second, which was easily fielded. Out.
Third and last at bat. I let a perfect strike pass. On the second pitch…THWACK!
I would like to tell you the ball sailed over the home run wall. There was no wall, and the hit was not impressive enough even if there had been. But it sailed over the second baseman's head and dropped between right and center for a textbook base hit.
For the rest of the season I kept practicing with the Johnny Bench rig. I became a progressively better batter. Good enough to move to 6th in the lineup. A fully contributing member of the team.
How might this story help us understand how to train people to use AI effectively and safely?
The first agentic AI use case that gained wide adoption was software coding. Two years in, a pattern is clear.
Almost every software engineer now uses AI somehow, every day. Effective leverage varies widely. A small top sliver has demonstrated 3x to 10x gains in meaningful output. Not tokens used. Not lines of code. Not story points. Positive business impact.
Below that group, leverage drops quickly. Engineers "feel" at least 2x more productive, but those gains often fail to show up as measurable business value the way the top practitioners' do. Some of that is organizational, not AI skill. Still, there is ample anecdotal evidence that most engineers struggle to move past incremental leverage.
Based on conversations with over 20 CTOs managing more than 2,000 engineers collectively, I am unaware of any factors that accurately predict, a priori, who will become an AI super-user.
There are predictors for who will not:
- Engineers who have not yet mastered software architecture and design patterns.
- Engineers who do not understand what good looks like for unit and regression testing.
- Engineers who want faster automation of their existing coding, using AI as a better auto-complete.
Even among people who clear those bars, only a fraction rapidly become super-users. Days are early. Different learning curves may close much of the gap.
But how much time? And how much of the gap?
The path to AI super-user status depends on more than education. Having spent time with several AI super-users, I can say with high confidence that they think differently and act differently than most engineers. The differential seems associated with behavioral thought patterns.
If that is right, the path requires not just education, but training.
Nearly everything we call "training" in the professional world is actually education. Education transfers knowledge and skills. Training changes behavioral thought patterns and develops natural abilities.
Education can happen in a concentrated period, separate from regular work.
Training requires continuous feedback during extended application of those abilities and thought patterns.
Education is my baseball coach showing me how to stand and hold the bat.
Training was several thousand swings on the Johnny Bench Batter-Up, with clear, consistent feedback after every swing.
So what is the Batter-Up for AI?
It is not another course. It is a practice environment where the variables are constrained, the reps are dense, and the feedback is immediate — and honest.
On the rig, the ball came back on a predictable path. I could isolate the swing. In a live game, pitch variance hid my progress. Most AI "training" at work looks like the game: noisy tickets, shifting contexts, production pressure, and feedback that arrives days later as a code review or an incident. You cannot tell whether you improved the swing or got lucky on the pitch.
A useful AI practice loop looks more like the backyard:
- One clear objective per session, not a grab bag of tasks.
- Constrained tools and context, so you are training judgment, not drowning in options.
- Immediate signals for good contact: tests that pass for the right reasons, diffs you can explain, failure modes you can name. Not vibes and token counts.
- Enough cycles in a sitting that tempo and taste start to form in the hands, not only in the notes.
Safety belongs in that same loop. Unsafe AI use is often chasing junk with a better swing: accepting fluent output you cannot verify, letting the model set the goal, or treating the agent as a gun you point at an objective. The coach's correction applies. Wait for your pitch. Know what good contact feels like. Refuse the swing you cannot defend.
Education still matters. You need the stance, the architecture, the sense of what good looks like. Without it, a thousand reps just engrave bad habits.
Education is necessary. It is not enough.
If we want 3x to 10x operators instead of people who feel twice as busy, we have to stop confusing briefings with batting practice. Build the rig. Put people on it. Count the thwacks and the biffs until the swing holds up when the pitch is live.