What about robots
Written September 19, 2026. Everything else on this site is practice: what runs in my shop, what I’m building, what I’ve planned. This chapter is the one exception. It’s where I think this is going, and what I think we should get ready for. The facts in it are dated and checked. The belief is mine, it runs ahead of the facts, and I say so where it does.
Where this chapter came from
Two nights ago I said out loud that robots will replace every hands-on job, and I wanted a chapter that said so. Instead I wrote a research brief, sent it to one AI, and had a second AI open every number in the answer at the company’s own release, the government’s own page, or the paper itself. Two passes, about a hundred claims opened, nine small corrections, nothing invented. What survived is below. What didn’t survive is any date. Nobody I could find, in any company or any government office, has one they’d defend. So I’m going to give you mine, and I’m going to tell you plainly that I can’t defend it either.
What’s true this week
- Agility Robotics announced Digit 5 on September 15. Its own release: lifts 50 pounds, reaches 7.2 feet, 90-minute battery that charges in 9 minutes, “more than 20 hours of productive work in a 24-hour day.” Built to work without safety cages by avoiding a person, stopping, or sitting down when someone comes close. Early access first half of 2027, general availability end of 2027. The models before it had to work behind barriers. One generation, and the cage came off.
- Toyota told investors it could need about 400,000 of its own robots across its plants and suppliers from 2028, at up to a trillion yen a year, about $6.4 billion. An estimate, not a signed order. Its robot, ELEY, weighs 110 pounds, rolls on wheels, and has two fingers per arm. Workers teach it by wearing finger-shaped jigs and repeating the job. It folded T-shirts after 1,500 tries in two weeks. The plan is to teach a skill once and copy it to every plant.
- China made “embodied-intelligence robot application technician” an official occupation on September 9, with two trades inside it: robot data collector and robot trainer. A training center near the Beijing trade fair runs nearly 270 robots on chores, caregiving drills, and factory tasks, and produces nearly 20 million training-data entries a year. A government wrote the job title before most of us had heard of the job.
- People already said where they’ll accept one. A nine-country survey of 9,000 adults, published June 2026: comfortable with robots in factories and warehouses, 63 percent. Hospitals and clinics, 45. Classrooms, 39. Would choose a robot to care for the sick or elderly, 12.
What’s true about our trades
This is the part I didn’t have two nights ago. Each line was opened at the source.
- Nobody sells or leases a robot to a residential trade. Not one vendor publishes a price for a machine that does a plumber’s, an electrician’s, or a repairman’s job. The closest things you can actually buy: a $13,500 research humanoid with a two-hour battery and a four-pound arm; a $499-a-month home robot whose terms say personal use only, no commercial work; and a warehouse robot whose maker’s own filing prices it at $8,500 a month and calls its useful life “largely unproven at commercial scale.”
- The money is already the same. The job isn’t. By the government’s own benefits ratio for repair occupations, a technician at $25 to $35 an hour costs a shop about $76,000 to $106,000 a year loaded, before the truck, the tools, or the fuel. That warehouse robot’s illustrative lease is $102,000 a year. Same money. It moves totes.
- Every license attaches to a person. Plumbing, electrical, the HVAC contractor’s license, the EPA refrigerant card, the building permit, the restaurant’s person-in-charge. All of them name a human. No licensing board in the country has published a word about machines. A robot can’t pull a permit, and no inspector has said what they’d do with one.
- The customer can say no at the door. Nothing gives a machine a right to enter a house. “No robot” ends the call. That’s a business barrier, not a technical one, and it’s the one nobody talks about at the keynotes.
- The government projects every trade on the list growing through 2035, except mine. HVAC up 11 percent, electricians 9, plumbers 7, restaurant cooks 12, all faster than average. Home appliance repair: 40,500 jobs, up 2 percent, about 800 new jobs in ten years, slower than average. I’m writing that down because it’s true, not because I like it.
- The famous study got the desks right and the hands wrong. In 2013 two Oxford researchers went through 700 jobs and gave each a score from 0 to 1: their estimate of how likely a computer could do that job within a decade or two, so 0.96 means 96 percent likely. They scored secretaries at 96 percent, bookkeepers at 98, landscapers at 95, home appliance repair at 72, and electricians at 15. That’s where the “47 percent of jobs at high risk” headline came from. Thirteen years later the desk jobs on that list are being eaten by software, on schedule. The hands-on jobs with the same high scores aren’t; nobody’s landscaper got replaced by a machine. Two later studies looked at what each job actually does, task by task, and got 9 percent and 14 percent, not 47.
- The last time desk work automated, the people didn’t come to us. From 1980 to 2005, computers took the routine office work. The displaced went into food service, cleaning, and care, which grew by half and stayed low-paid. Construction and craft employment fell. Nobody has published a study showing office workers flowing into licensed trades. I believe they will this time. The record says they didn’t last time.
- The people saying five years have said it before. In January 2016 the man now promising humanoids for 2027 said full self-driving was “approximately two years” away. Ford promised a car with no steering wheel by 2021. Waymo went driverless in one city in October 2020 and is still fenced to a few cities six years later. The one roboticist with a dated public scorecard wrote in 2018: dexterous hands not before 2030, an affordable robot that can get around a house not before 2035. His 2025 update: “not for decades.”
What I take from it, today
The factory version of hands-on work is starting to work. The house-call version isn’t, and nothing in the research says when it will.
A factory buys the robot and then spends money making the world around it boring: same parts, same shelves, same light, same floor. A house is the opposite on purpose. Every machine is a different year, brand, and install. The sticker is faded. The dryer is stacked on the washer in a closet with no light. The floor slopes. The customer is in the doorway asking whether to fix it or replace it, and you answer with your eyes, your ears, a wet finger, and thirty years of “I’ve heard that sound before.” No machine announced this year is close to that, and the companies building them aren’t claiming otherwise.
A line from the research I keep coming back to: a robot doesn’t have to beat a good tech. It has to beat a vacancy. That’s already happening on warehouse night shifts. It isn’t happening in laundry closets. Yet.
What I believe is coming
Here is the belief, labeled as belief. I think physical labor, all of it, gets replaced. Appliance repair, plumbing, electrical, painting, home building, cleaning, landscaping, the whole list. My range is ten to twenty years, and if the machines keep learning at the rate the T-shirt story suggests, it could be five. I can’t defend any number in that range, and neither can anyone selling you one.
The evidence above runs against the five-year end of my range and says nothing that supports the ten. I know that. I’m keeping the belief anyway, and I’m putting the evidence next to it so you can weigh them yourself instead of taking my word.
The reason I believe it is a chain, and I’ll show you the links so you can see which ones aren’t there yet. A machine that can learn any skill from watching, then a machine that can design and build other machines, then machines that can do anything a person can do with their hands. The first link has a foot in the door: Toyota’s jigs and China’s training center are that link at the factory. The second and third are steps nobody has taken. I believe they get taken. That’s a belief about the direction, not a measurement of the date, and I’ll come back and change this paragraph when I’m wrong about either.
What to get ready for, as a shop
Whether it’s five years or twenty, the shop that’s ready looks the same. These are in order.
- Take the office half now. Phones, booking, tickets, the manual you photographed last year. That’s software, and it’s already here. That’s chapters 1 through 5. Every month you run it is a month of records the next step needs.
- Own the shop’s memory. Your tech sheets, your ledger of what failed and what fixed it, your pricing rules, your customers. If a machine ever learns your trade, it learns it from records like these. Keep them yours, in your files, not in somebody’s app.
- Keep the hands sharp, and hire while the hands still matter. They’re the moat for as long as the moat lasts. A green tech with the system is worth more than either alone. The government says my trade adds 800 jobs in ten years; it says HVAC adds 48,000. Read your own trade’s number before you plan a hire.
- Learn to teach a machine. China put “robot trainer” and “robot data collector” on the books this month. Toyota’s workers teach ELEY by doing the job with jigs on their fingers. The person who knows the work and can show it to a machine is a job that exists now. Thirty years of hands may be worth more as a teacher than as a worker. I had the idea of recording my own hands through camera glasses to train a machine. The research found the market for that: a public hub with 791 robot datasets from 235 people, and nobody paying for any of them yet. Whether a tradesman’s recordings are worth something to anyone is still unknown.
- Be the shop that leases the robot, not the shop it replaces. No such lease exists for a residential trade today; the research looked. When one does, somebody local is going to be the one who rents it, sends it, and answers the phone when the customer says it scratched the cabinet. Be that shop. That’s the same business you have now with a different truck.
- Keep the doorway. The customer who can say “no robot” is the same customer who says “send Marc.” The conversation where you tell them the truth about repair versus replace is the last thing to go, and it’s the thing they come back for. Everything above serves it.
What would change my mind
Sooner: a paid service call in a stranger’s house finished by a machine with nobody from the shop in the room, with an invoice and the customer’s signature. A licensing board saying a machine may do the regulated work without a licensed person present. An insurer publishing a policy, with a price, for a robot working inside a customer’s home. A robot-employee lease offered to a one-truck trade for less than a tech’s loaded cost.
Never, or much later: five more years of factory robots getting better at factories and no better at houses. The government’s ten-year numbers for the licensed trades still pointing up in two more releases. The scorecard’s dates slipping further out instead of pulling in.
What I still don’t know
Whether a board would count a licensed person watching a machine over a camera as “the licensed person did the work.” What it would cost to insure one in a customer’s house. Whether a health department would ever allow a kitchen with no person in charge. Whether my hands on video are worth a dollar to anyone. Those are the questions the research came back with “nobody knows” on, and each one names the document that would answer it. When one of those documents exists, this chapter gets a new date.
What I’d do
Keep the hands. Use the software now, because the shop that still answers with voicemail is your competitor this year, and the robot in the warehouse isn’t. Watch factories first: Digit 5 early access in 2027, Toyota’s rebuild after 2028. If a humanoid starts doing useful work in a stranger’s kitchen, you’ll hear about it from a customer before you hear it from a keynote. And when you do, be the one holding the lease.
Back to the front page. That is the last chapter.