The Easy Part

Nobody in that room thinks they just watched the hardest thing in the building. Earlier that afternoon, an AI drafted a lease amendment for one of the diners in a few seconds, and that felt like a miracle. The server felt like nothing at all. She felt like a person doing her job.

That reaction, miracle for the contract and nothing for the tray, is backwards. It has been backwards for as long as there have been computers, and the people who build robots have known it for almost forty years.

The paradox

In 1988 Hans Moravec, a roboticist at Carnegie Mellon, wrote down what his field kept running into. It was comparatively easy to get a computer to perform like an adult on an intelligence test or at checkers. It was hard, often impossible, to give it the perception and mobility of a one-year-old. The machines could do the things we found difficult and could not do the things we found trivial.

His explanation was evolution. Seeing, balancing, grasping, catching, reading a face: these have been refined by natural selection for hundreds of millions of years. They are so well built that they run below awareness, and so they feel effortless. Abstract reasoning is new. Moravec called it the thinnest veneer of human thought, a trick perhaps a hundred thousand years old that we have not had time to get good at. Marvin Minsky said the same thing from the inside: we are least aware of what our minds do best. A few years later Steven Pinker compressed it into one line that now travels on its own: the hard problems are easy and the easy problems are hard.

Effortless is the signature of depth. Hard is the signature of newness.

Here is the principle underneath, stated plainly. The difficulty of building a skill into a machine tracks how old the skill is, not how impressive it looks. Calculus feels hard because we are bad at it. Walking across a crowded room feels easy because we are astonishingly good at it, and we are astonishingly good at it because a very long line of ancestors died when they were not.

The scale of it

Take the last 600 million years or so of animal life and stand it up as a column 100 meters tall, the height of a thirty-story building. Eyes and balance fill almost all of it.

• Language the top two centimeters
• Writing and arithmetic the top millimeter
• Calculus, chess, contract law about the thickness of a hair

The order of arrival

AI has come in exactly the order Moravec predicted: from the top of the column down. It took the hair first. The bar exam, the olympiad proof, the contract, the code, the diagnosis read off a page. In the summer of 2025, AI systems reached gold-medal level at the International Mathematical Olympiad. As I write this in November 2026, the people who track home robots most carefully still cannot point to an owner whose delivered humanoid does the chores on its own. The demos are real. The dishwasher in a stranger’s kitchen is still mostly unsolved.

The obvious reading is that physical jobs are safer than desk jobs. That is true, and it is the least interesting thing the paradox says.

The interesting thing is this. We cannot see our own deepest skill, because the deepest skill is the one that feels like nothing. We price work by two measures: how hard it feels to us, and how long it took to learn in school. Both measures point straight at the hair-thin layer. So AI is taking exactly the work we were proud of and leaving the work we never thought to be proud of. And for that remaining work we have almost no vocabulary. No line on the résumé. No credential. No wage scale. We have shrugs: she’s just good with people. He has good hands. She knows when something is off.

The philosopher Michael Polanyi put the reason in one sentence in 1966: we can know more than we can tell. The deep layer was never written down, which is why machines that learn from text learned it last, and why it never appeared in a job description either. The same silence that protects it from the machine hides it from us.

In The Attender I argued that the most capable machine in you is the one you never notice running: the subconscious prediction machine that keeps you breathing without a dashboard. What it has absorbed sits beneath attention, and what sits beneath attention never gets credit. Moravec’s paradox is the economic face of the same fact. The work we can see gets the title. The work that has sunk below seeing carries the tray.

Five predictions

1. The easy skills get names.

   Hiring will start testing what it used to assume: composure under load, steady hands, reading a room, noticing the person who needs something before they ask. Job posts will list these first, because everything listed after them can be done by the candidate’s AI.

2. Prestige turns over before pay does.

   The first change is respect. The phrase “just a server” will start to sound as dated as “just a typist” sounds now, except in the opposite direction. Pay follows more slowly. I made the wage argument in The Expensive Hour; this is the reason underneath it.

3. The robots arrive later than the demos promise.

   Every year the home humanoid slips is evidence for Moravec, not against progress. The bottom of the column will fall, but it will fall slowly, chore by chore, and each chore will fall on a mountain of recordings of people doing it well.

4. People get paid to do ordinary things well, on the record.

   Because the deep layer cannot be described, it has to be demonstrated. A real job category will grow around people who fold, plate, pour, carry, comfort and tidy while a machine watches and learns, and around the people who take over a robot remotely when it gets stuck.

5. People discover their real skill late, and from the outside.

   The most common career moment of the next decade will be someone whose official job was taken by AI realizing that the thing they were praised for in passing, the thing they thought was just personality, was the job all along.

Where I am standing

At Charleston AI we deliver finished AI work to restaurants, inns and local owners. When we built an AI receptionist for a hibachi steakhouse here in town, the phone was the part we could take. Nobody, me included, proposed taking the grill: the chef with the knife and the flame and a table of strangers who came to watch him. That is the paradox in one building. The front desk was the top of the column. The grill was the base. I have written about what happens to the people promoted to managing their AI and about the work you can finally stop carrying. This is the other half. What you keep is mostly what you never thought to count.

The ledger

The easy skills become the hiring test

Already true Written work no longer separates candidates, because anyone can produce polished writing on demand.

What has to happen Employers in service, care and the trades have to find cheap, fair ways to observe tacit skill in person, and start trusting those observations over the paper.

Where I am probably wrong Employers may simply keep hiring on paper out of habit and convenience. If, five years from now, a typical service job is still filled from a résumé and a short phone screen, I was wrong about this.

The robots lag the demos

Already true Home humanoids exist as demos, betas and remotely assisted products; independent reports of one doing a stranger’s chores unaided are still missing.

What has to happen Nothing new. The claim holds if the base of the column keeps falling chore by chore rather than all at once.

Where I am probably wrong Robot models trained on enormous amounts of video may close the gap far faster than the history suggests. Moravec himself expected the base to fall eventually; the paradox describes the order of conquest, not the outcome. If a delivered home robot is reliably doing laundry and dishes unassisted in ordinary homes by the end of 2028, the window this piece describes is short.

We find our real skill late

Already true Most people describe what they are best at in terms of what they studied, and describe what others praise in them as personality.

What has to happen Enough top-of-column work has to be absorbed that people are forced to look at what is left.

Where I am probably wrong People may never revalue what feels easy, because feeling easy is exactly what makes it invisible. If the shrug survives, the skill stays underpaid even after it becomes the only skill left.

It was the hardest thing in the room. It always was. It only looked easy.

Background

• Hans Moravec, Mind Children: The Future of Robot and Human Intelligence (Harvard University Press, 1988), the original statement.
• Marvin Minsky, The Society of Mind (Simon and Schuster, 1986).
• Steven Pinker, The Language Instinct (1994).
• Michael Polanyi, The Tacit Dimension (1966).
• Moravec’s paradox, Wikipedia.
• What can a home humanoid robot do today? A reality check, Humanoids for Home, November 2026.
• From this blog: The Expensive Hour, The Attender, The Last Line Is You, Masterpiece.