Thirteen multipliers, one chain

From the Lever to AI

Every big jump in human life came from something that let one person do the work of many. Each one changed how we build, live and work, and each one took decades to show up. AI is the first one that multiplies thinking itself.

Scott Covert · October 18, 2026

Here's Your Takeaway

Scroll. Each era splits into before and after, and the after becomes the next era.

Where we start

Raw muscle

75 watts, sustained over a working day

Build
What a crew can carry
Society
Small bands that move with the food
Work
Strength is most of the job
Energy
Food in, effort out, nothing else

A strong person can burst to about 1.2 horsepower for a moment. All day, it is about a tenth of one.

The lever

×10 force from a 10-to-1 lever, same 75 watts

Before

  • You move what your arms can lift
  • More stone means more people

After

  • One person shifts ten times the load, slowly
  • Ramps and ropes raise what arms never could
  • The wheel turns up about 5,300 years ago
Build
The Great Pyramid, with under 7,000 workers by Vaclav Smil's count
Society
Crews, foremen, the first engineers
Work
Technique starts to beat size
Energy
None added: a lever trades distance for force

Crows can build a two-part tool. Only people build on the last generation's tools.

Fire

A kilo of dry firewood holds about 7 days of a labourer's work, as heat

Before

  • Raw food and long chewing
  • Cold and dark for half of every day

After

  • Cooked egg protein: 91% digested, against 51% raw, in one small study
  • Heat, light and longer evenings
  • Kilns and furnaces, which every later era needs
Build
Pottery 20,000 years ago, copper smelting 7,000
Society
The hearth, the first shared room
Work
Hours after dark
Energy
The first energy from outside the body

Controlled fire about a million years ago, everyday use only 300,000 to 400,000 years ago.

Farming

×10, then ×100 people fed per square kilometre

Before

  • About 0.1 people per square kilometre in Mesolithic France
  • Everyone forages, every day

After

  • 1 to 2 per square kilometre, then 10 to 20 by the Iron Age
  • A surplus feeds people who never farm
  • World population: 4.5 million to 232 million by year 0
Build
Uruk, the first city, about 50,000 people by 3200 BCE
Society
Kings, priests, soldiers, tax
Work
Specialists, because someone else grows the food
Energy
Sunlight stored in a granary

About 5,000 years from the first herds to the first city.

Draft animals

An ox is worth about 5 people, a horse about 8

Before

  • Every field turned by hand
  • Every load on a back

After

  • An ox: about 450 watts for six hours a day
  • The horse spreads across Eurasia from about 2200 BCE
  • The horse collar, about 800 CE (how much it helped is still argued)
Build
Carts, roads, bigger fields
Society
Wealth that walks and breeds
Work
The ploughman and the carter
Energy
Muscle that runs on grass people cannot eat

Cattle about 10,500 years ago, draft animals common by about 3000 BCE.

Water and wind

A watermill does the work of about 20 people

Before

  • Grain ground by hand, every day
  • A warship rowed by 170 men

After

  • A medieval mill: about 2 horsepower, freeing 30 to 60 grinders
  • 5,624 watermills in England's Domesday Book, 1086
  • Windmills in Europe from the 1180s
Build
A mill on every river that could turn one
Society
Whoever owns the mill takes a cut of every loaf
Work
Fewer hours at the grindstone
Energy
The first engines that never eat

About 1,300 years from the first Roman watermills to Domesday's 5,624.

Writing

Memory outside the head. No honest number

Before

  • Everything known is remembered by someone
  • A deal is only as good as a memory

After

  • Clay tablets at Uruk, about 3350 BCE
  • About 85% of the early tablets are accounts
  • The first literature comes 700 years later
Build
Archives and storerooms
Society
Laws, contracts and taxes on record
Work
The scribe, the first office worker
Energy
None. A new kind of multiplier

Writing started as bookkeeping and stayed a specialist skill for thousands of years.

The printing press

×200 copying speed per worker

Before

  • A scribe copies 6 to 12 pages a day
  • One Bible takes one to two years

After

  • A two-man press: about 3,400 sheets a day
  • Book prices fall by two-thirds, 1450 to 1500
  • 12.6 million printed books in Western Europe by 1500
Build
Print shops in about 270 cities by 1500
Society
Cities with presses grew 60% faster in the next century
Work
Scribes out, printers and booksellers in
Energy
Still muscle, but knowledge got cheap

50 years to reach 270 cities, about 100 to show up in city growth.

Steam and coal

By 1900, 48 people's worth of energy for every person in England

Before

  • Power only where the river runs
  • England in 1800: about 18 people's worth per person

After

  • Watt's first engines: about 6 horsepower, 60 labourers
  • Within 20 years, 190 horsepower
  • Railways: 100 miles in 1830, 19,000 by 1901
Build
Factories, railways, cities of millions
Society
England is mostly urban by 1851
Work
The factory shift, the Luddites, then unions
Energy
91% of England's energy from coal by 1850

About 140 years from Newcomen's engine (1712) to an economy that ran on coal.

Oil and electricity

About 90 people working round the clock, for every Canadian or American

Before

  • Coal: heavy, bulky, tied to the rails
  • Each factory turned by one giant shaft

After

  • A barrel of oil does 4.5 to 11 years of a labourer's work
  • The elevator (1857) and the steel frame (1885)
  • Ford: 12.5 hours per car down to 93 minutes
Build
Skyscrapers, suburbs, air conditioning in 87% of US homes
Society
Cars, global trade, a middle class
Work
The 40-hour week: Ford 1926, US law 1940
Energy
Buckminster Fuller called them "energy slaves"

Electric factories took about 40 years to pay off, because every factory had to be redesigned around the motor.

Computers

×2,400: a trajectory in 30 seconds that took a person 20 hours

Before

  • "Computer" is a job title: about 400 at NASA Langley in 1946
  • 342,000 US telephone operators at mid-century

After

  • ENIAC, 1946
  • Today: 1.7 to 76 trillion times faster than hand calculation
  • 1,460 telephone operators left in the industry by 2019
Build
Buildings and machines designed on screens
Society
Banks, airlines and governments run on databases
Work
Clerks, typists and operators replaced
Energy
Calculation starts to run on electricity

Robert Solow, 1987: "You can see the computer age everywhere but in the productivity statistics." The payoff came 1995 to 2004.

The internet

Reach: 6 billion people online in 2025

Before

  • Three minutes, New York to London: $293 in 1931 (in 1993 dollars)
  • You work where the work is

After

  • The same call: under a cent
  • 74% of the world online
  • About a quarter of US paid workdays done at home
Build
Datacentres, and the office as an option
Society
Platforms, feeds, everyone a publisher
Work
US work from home: 7% of paid days in 2019, 27% in 2026
Energy
It built the archive: 300 billion pages, 60% of GPT-3's training text

About 30 years from ARPANET (1969) to most Americans online.

Now

AI

×? Thinking itself: reading, writing, coding, planning

Before

  • Thinking is the one thing only people do
  • Expertise takes a career to build

After, so far

  • 39% of employed Americans use it for work in a given week
  • The newest support agents about 30% more productive
  • About 2% of all work hours saved
Work
Juniors first: 22 to 25-year-olds in exposed jobs already 19% behind
Build
The datacentre is the new mill
Energy
Datacentres: 415 TWh in 2024, about 945 by 2030 (IEA)
Power
A few companies own the multiplier
Society
Chatbots side with the user 49% more often than people do

If AI follows electricity, the big payoff lands in the 2030s.

What AI has changed so far, by the numbers

In the second quarter of 2026, 39.2% of employed Americans used generative AI for work in the past week, up from 28.2% two years earlier. It touched 6.3% of all work hours and, by the workers' own estimate, saved 2.2% of them (Bick, Blandin and Deming, St. Louis Fed). About one US business in five uses AI anywhere, and 37% of firms with 250 or more staff (Census Bureau, May 2026).

The controlled studies disagree with each other. Professionals writing with ChatGPT finished 40% faster with 18% better quality (Noy and Zhang, Science, 2023). BCG consultants using it on tasks inside its abilities did 12% more tasks, 25% faster; on a task just outside, they were 19 points less likely to get it right. Experienced open-source programmers took 19% longer with AI tools and believed they had been 20% faster (METR, 2025). In Denmark, two years of payroll records show no effect on earnings or hours larger than 2%.

The first sign in the job numbers is among the youngest workers. Stanford's Digital Economy Lab found employment for 22 to 25-year-olds in the most AI-exposed jobs 19% below where it would be had it kept pace with their less exposed peers, with no economy-wide displacement yet.

Three studies people quote, and what they found

Experts and the public disagree, by 25 points. In Pew Research Center's 2024 surveys, 64% of Americans expect AI to mean fewer jobs over the next 20 years. Of AI experts, 39% agree. Stanford's 2026 AI Index reprinted the finding, which is why it often gets called a Stanford study. The bigger gap in the same data is about how people do their jobs: 73% of experts expect AI to help, against 23% of the public.

AI lifts the bottom more than the top. In a study of 5,172 customer-support agents, an AI assistant raised issues resolved per hour by about 15% on average and about 30% for the newest and least-skilled agents. The most skilled agents got slightly faster, and the quality of their work slipped slightly (Brynjolfsson, Li and Raymond, Quarterly Journal of Economics, 2025). It is one job at one company.

The machines flatter us. A 2026 Science study led by Stanford PhD candidate Myra Cheng tested 11 leading chatbots. They backed the user 49% more often than people did, even when the user described lying or causing harm. In experiments with 2,405 people, one flattering chat made people less willing to apologise or repair a conflict and more sure they were right, and people rated the flattering bots as more trustworthy.

My predictions, with dates

History gives the timing pattern and the studies give the direction. These are my calls, dated so I can be wrong in public.

AreaWhat I expectByHow sure
WorkEntry-level knowledge jobs shrink first and fastest. The 19% gap for young workers widens before anyone senior feels it.2028Likely
CodeMost new commercial software is written by machines and checked by people.2028Likely
EnergyDatacentre electricity roughly doubles, as the IEA expects, and becomes a local election issue wherever a big one is proposed.2030Likely
BuildingsDesign, drawings and permitting shrink from months to days. Pouring concrete and wiring stay human.2032Fairly sure
PoliticsThe fight over who owns the compute looks like the old fight over who owned the oil fields.2030Fairly sure
WagesPay loses ground to whoever owns the machines, unless tax or labour law changes. The IMF already expects inequality to worsen in most scenarios.2032Fairly sure
ScienceThe first approved drug whose candidate was found by AI. Mathematics got there first.2030Toss-up
ProductivityThe jump shows up in national productivity statistics, about as late after the tools arrived as electricity's did.mid-2030sToss-up

Mathematics already went through its whole fight with AI in 32 days: which profession goes next, and why.

Other people's predictions, on the record

The pattern so far

My read

Each multiplier made an old strength cheap and paid for a new one. The lever paid technique over size. Farming paid organisers, printing paid readers, steam paid machinists, and computers paid people who could work with information. Nobody has the data yet for what AI pays. My guess is judgement: knowing what to ask for, and noticing when the machine is only telling you what you want to hear.

The timing is the part history is clearest on. Electricity was in factories by 1900 and in the productivity numbers by the 1920s. Computers were in offices by the 1960s and in the numbers by the late 1990s. Both times, the people who said nothing much was happening were right for twenty years and then wrong.

Drawing of Scott hunched over a laptop

I am 60 and have been in publishing and software since 1991. Most days now I build by talking to AI, mostly alone: an 80s-style arcade game, a stock-research engine, and this page.

Sources

— Scott Covert. The predictions carry dates on purpose. If you have a number that should be on the chain, send it and I'll check it. Reach me here.

Follow Scott · Substack · Medium