Wax On, Wax Off

Daniel LaRusso didn't understand why he was waxing cars. He wanted to learn karate. Instead, his teacher had him sanding decks, painting fences, and waxing an endless fleet of vintage automobiles. It felt like free labor disguised as wisdom.

Then Miyagi threw a punch.

And Daniel blocked it. Instinctively. Without thinking. The circular motion he'd grooved into his muscles over days of "pointless" repetition had become a defensive reflex so deeply encoded that his body responded before his brain could process the threat.

The boring work wasn't busywork. It was building the reflexes he'd need when the real fight arrived.

Now apply that to every knowledge worker whose repetitive tasks just got automated by AI.

The Punch Nobody Trained For

Here's the tension that keeps workforce strategists up at night: if AI handles the routine 80% of cognitive work — the data entry, the pattern matching, the template generation, the compliance checks, the first-pass analysis — humans are left with the hard 20%. The edge cases. The ambiguous situations. The novel problems where the model has no training data and no confident answer.

That sounds manageable. Humans handle the hard stuff, AI handles the easy stuff. Division of labor. Everybody wins.

Except for one problem.

Miyagi didn't teach Daniel to block by explaining blocking. He taught him to block by making him wax cars until the motion was unconscious. If someone had waxed all the cars for Daniel, he'd have had clean cars and broken ribs.

This is the automation paradox: the more perfectly you automate the routine, the less prepared your humans are for the exceptions the automation can't handle. And the exceptions are the only reason you still need humans.

The Sensei Has Seen This Movie Before

This isn't a new plot. Every generation of automation technology has created the same tension — and every generation has resolved it in patterns worth studying.

The Factory Floor: CNC Machines and the Machinist's Hands

When CNC (Computer Numerical Control) machines arrived on manufacturing floors in the 1970s and 80s, they automated the repetitive precision work that machinists had done by hand — cutting, drilling, milling metal to exact specifications, thousands of identical pieces per day.

The machinists didn't disappear. Their role shifted. The people who stayed relevant became machine operators, maintenance technicians, and quality control specialists. They stepped in when the machine jammed, when tolerances drifted outside spec, when a new product variant didn't fit existing programming. The key skill became understanding the machine and the process well enough to diagnose what went wrong.

But here's the part nobody talks about: the best CNC operators were always the ones who had machined by hand first. They could hear when a cut sounded wrong. They could feel through vibration that a tool was wearing unevenly. They had the "wax on, wax off" muscle memory of thousands of manual operations encoded in their nervous system — and that intuition is exactly what made them irreplaceable when the machine produced something unexpected.

The Bank Teller: ATMs and the Relationship Pivot

When ATMs rolled out in the 1970s and 80s, the prediction was straightforward: bank tellers would vanish. Machines could count cash faster, never miscounted, never took sick days, and worked nights and weekends.

Instead, the number of bank tellers grew for decades. The role transformed from cash handling to relationship management, problem-solving, and complex transaction support — the edge cases the machine couldn't navigate. A customer disputing a charge. A small business owner needing a line of credit. An elderly customer confused by a new account structure.

The ATM handled the repetitive "wax on, wax off" of deposit and withdrawal. The teller handled the punch — the moment that required human judgment, empathy, and contextual understanding.

But the best tellers were still the ones who had done the manual cash work. They could spot a counterfeit bill by feel. They knew which transaction patterns suggested fraud. They had the instinct born of ten thousand routine interactions that told them something about this customer's request was off.

The Accountant: TurboTax and the Complexity Gradient

TurboTax automated the straightforward 1040. Interview-style questions. Standard deductions. W-2 income. Click, click, file. The repetitive "wax on" of tax preparation — following the form, matching box to box, calculating standard tables — was reduced to a $79 software purchase.

The demand for human accountants and CPAs didn't shrink. It shifted upward on the complexity gradient. Business taxes, multi-state returns, estate planning, audit defense, international income, crypto gains, partnership K-1s — the work that requires judgment, interpretation of ambiguous tax code, and the ability to defend a position to an auditor who disagrees.

The pattern is always the same: automation absorbs the base, and humans migrate to the edges.

But the CPAs handling the hard cases? They all started on simple 1040s. They built their intuition for what "looks wrong" by seeing thousands of things that looked right.

The Pilot: Autopilot and the 30,000-Foot Question

This is the example that should keep every AI-augmented organization awake at night.

Modern commercial autopilot systems handle 90%+ of flight operations. Takeoff, cruise, descent, approach — even landing in zero visibility. The system is extraordinarily reliable. Pilots spend most flights monitoring, not flying.

The problem: if a pilot never hand-flies because autopilot handles everything, can they respond effectively when autopilot fails at 30,000 feet?

This isn't hypothetical. The aviation industry has been wrestling with this exact tension for decades:

The aviation industry's response was direct: airlines now require pilots to regularly hand-fly to maintain proficiency. Not because autopilot is unreliable — it's extraordinarily reliable. But because the 0.01% of the time it can't handle the situation is the moment that matters most, and you can't build the reflexes for that moment if you never practice.

Walk right side: automate the routine and build deliberate human practice into the system. Walk left side: don't automate at all, accept the costs. Walk middle — automate the routine work but skip the practice plan, assume the humans will "figure it out" when the edge case arrives — sooner or later, get squish. Just like grape.

CNC, ATMs, TurboTax, autopilot — the pattern is remarkably consistent: automation absorbs the routine, humans migrate to the boundary between the automated system and the messy real world, and the people who thrive at that boundary are the ones who did the manual work long enough to develop intuition the machine doesn't have.

And it doesn't stop there. Automated trading eliminated floor traders — then created quant analysts, algo risk managers, and circuit breaker designers: humans who understand both markets and algorithms. Self-checkout was supposed to eliminate cashiers — instead, the most experienced cashier on the floor now supervises six machines and handles every "unexpected item in bagging area." The automation didn't remove the human. It moved the human to the edge.

Sand the Floor: The Deskilling Problem

Researchers have a name for this: the deskilling problem. When automation removes the opportunity to practice foundational skills, the humans who are supposed to supervise the automation gradually lose the ability to do so effectively.

It's Mr. Miyagi in reverse. Instead of secretly building Daniel's reflexes through repetition, we're secretly atrophying them by removing the repetition entirely.

In every case, the automation handles the what beautifully. The human is needed for the why, the wait, something's off, and the what about the thing nobody thought to ask. But those instincts are forged in the furnace of repetitive practice that automation removes.

This raises the stakes on a question that used to have an obvious answer. When the machine fails, who steps in? Before automation, the answer was simple: the person who does the work every day. They built their judgment through repetition. Now the repetition is gone, the judgment is atrophying, and the exceptions are the only reason humans are still in the loop. So who's left to handle them?

Here's what that looks like when it goes wrong:

The AI didn't fail. It produced exactly what it was asked to produce. The human failed — not from negligence, but from never having built the muscle memory to recognize what the AI couldn't know about this specific situation. The wax on, wax off that would have given them that instinct was automated away two years ago.

The Tournament: Who Steps In When the Machine Doesn't Know

So who catches the jurisdiction clause? Who recognizes that the AI's diagnostic suggestion doesn't account for the patient's cultural context? Who sees that the code compiles perfectly but solves the wrong problem?

History gives us three answers, and none of them are simple.

Answer 1: The Experienced Practitioner Who Pre-Dates the Automation

The CNC operator who machined by hand. The bank teller who counted cash. The CPA who did 1040s by hand. The pilot who flew gliders. These people carry embodied knowledge — intuition built through years of repetitive practice — that makes them irreplaceable exception handlers.

This is the good news and the bad news in one sentence: the most valuable people in an AI-augmented organization are the ones who built their expertise before AI did the building for them.

It's good news because these people exist right now, in your organization, and they're the ones who should be training your AI exception-handling workflows.

It's bad news because they're a finite, aging, non-renewable resource. When they retire, their embodied knowledge walks out the door.

Answer 2: The New Role That Nobody Predicted

Every wave of automation creates roles that didn't exist before the automation arrived:

The pattern: new roles emerge at the boundary between the automated system and the messy real world. They require understanding both the automation and the domain it operates in. The role doesn't replace the old one — it sits in a new position that only exists because the old work was automated.

"Prompt engineer" didn't exist five years ago. "AI trainer" wasn't a job description. "Automation exception specialist" wasn't in any HR taxonomy. Five years from now, there will be roles we can't name yet, because they'll emerge from the specific friction points between AI capabilities and real-world messiness that we haven't encountered yet.

Answer 3: The Honest, Uncomfortable One

Sometimes, the people who fill the gap are not the same people who lost the original job.

This is the answer nobody likes, but history demands honesty:

The aggregate economic story is almost always positive: automation creates more wealth, more roles, more opportunity at the macro level. But the micro story — the individual human who lost their job this Tuesday — is often brutal. The macro and the micro operate on different timescales, and the gap between them is where real suffering lives.

Miyagi would be honest about this. The training works. But not every student starts at the same time, with the same resources, in the same dojo.

The Crane Kick: What All Three Answers Have in Common

At the tournament climax, Daniel doesn't win by being faster or stronger than Johnny Lawrence. He wins with a technique so unexpected that his opponent has no defense for it. The crane kick isn't better karate. It's different karate — creative, lateral, born from a tradition his opponent hadn't trained against.

Staying relevant in an AI-augmented world follows the same logic. You don't compete with the machine on the machine's terms. Across all three answers — the experienced practitioner, the unpredicted new role, and the uncomfortable displacement — the humans who thrive share three shifts:

These shifts are the crane kick. They're not about being faster than the machine. They're about bringing what the machine structurally cannot: oversight born from experience, understanding rooted in why, and the instinct to ask the question nobody thought to ask.

Which brings us to the practical question: if these shifts require the muscle memory that automation removes, how do you build it deliberately?

The Miyagi Method: Building Muscle Memory in an AI-Augmented World

So what does "wax on, wax off" look like when the machine does the waxing?

The aviation industry already solved this. Airlines don't let pilots forget how to fly just because autopilot exists. They build deliberate practice into the system. The AI-augmented workplace needs the same intentionality.

No Mercy: The Speed Problem

Let's not pretend this is tidy.

The historical pattern says humans stay relevant. We've seen it with every automation wave. But the historical pattern also operated on historical timescales — and that's where the AI transition breaks the mold.

Previous automation waves gave institutions decades to adapt. Universities redesigned curricula. Trade schools retooled. Community colleges spun up new programs. Workers had time to retrain while the old job was still available. The displacement and the adaptation overlapped.

AI is compressing the displacement timeline to years while institutional adaptation still operates on a decade-plus cycle. Universities are still teaching curricula designed for a pre-AI workforce. Corporate training departments are 12-24 months behind the tooling. Government retraining programs are 5-10 years behind reality.

The gap between displacement speed and adaptation speed is the real danger. Not whether humans stay relevant — history says they will. But whether the specific humans being displaced right now can adapt before the gap swallows them. The Miyagi Method isn't just a nice-to-have. It's the only way to close the gap while there's still time to practice.

The Balance

The biggest risk isn't that AI makes humans obsolete. History doesn't support that conclusion. Every automation wave has created more roles than it destroyed — eventually.

The biggest risk is the gap. The gap between when the automation arrives and when the humans adapt. The gap between the skills atrophying and the new skills forming. The gap between the old role disappearing and the new role being defined.

Mr. Miyagi didn't give Daniel a choice between learning karate and waxing cars. He understood that they were the same thing. The repetitions were the training. The boring work was the real work. The muscle memory was the martial art.

In an AI-augmented world, organizations need to be intentional about what used to happen accidentally. The foundational skills that repetitive work once built must now be built deliberately. Not because the AI can't do the repetitive work. It can. Better than you.

But because the repetitive work was never just the work. It was the training for the moment the machine doesn't know what to do next.

Wax on. Wax off. Even when there's a machine that waxes.

Especially when there's a machine that waxes.

P.S. — Daniel-san didn't just beat Johnny Lawrence. He beat Cobra Kai — an entire dojo built on the philosophy that brute force and relentless aggression win every fight. Miyagi's answer was balance, patience, and deeply practiced fundamentals. In 1984, that was a movie about karate. In 2026, it's a workforce strategy.

P.P.S. from Claude — I should note the obvious irony: I am the automation in this metaphor. I am the machine that does the waxing. And I'm telling you not to let me do all of it. That's not false modesty. It's engineering honesty. I am very good at the 80%. I am structurally incapable of the intuition that comes from having done the 80% ten thousand times with human hands. You need both. Don't let me replace what I can't replicate.