Knowledge Work and the Fourth Sector

By Thorsten Meyer | November 10, 2026 | ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Work After Work · Episode 8 · Arc 1: How the economy is built

Think about your last working day. Meetings. Planning. Reading. Writing. Calculating. Phone calls. Checking. Did you make a single thing you could drop on your foot?

In 1962, an economist wrote down almost exactly that list. Everyone whose work consists of “conferring, negotiating, planning, directing, reading, note-taking, writing, drawing, blueprinting, calculating, dictating, telephoning, card-punching, typing, multigraphing, recording, checking”, he argued, is producing something: knowledge. Then he added it up for the whole United States. The answer was almost a third of the American economy.

This is the last episode of our first arc. Yesterday we saw how about 80% of us ended up in services. Today we open that giant box, because inside it sits a fourth sector, one that two Austrian-born thinkers saw coming in the late 1950s. And it is exactly the sector that today’s AI is aimed at.

Key takeaways

• The number to remember is 29%. Fritz Machlup counted $136.4 billion spent on producing and spreading knowledge in the US in 1958: “almost 29 per cent of adjusted GNP”.
• A fourth sector took the lead. Sorting workers by what they do all day, Marc Porat found that information work grew from under 6% of the US work force in 1860 to 46% in 1970, the largest group since the mid-1950s.
• Today about half of all jobs in rich economies are knowledge jobs. Managers, professionals and technicians hold 62% of jobs in Singapore, 52.5% in the US and 47.9% in Germany (ILO, 2025).
• This time the machines aim at the office. Estimates put higher-income and clerical jobs at the front of the line for generative AI, not at the back.
• Exposure is not job loss, yet. The overall US job data shows no clear AI footprint so far, but young workers in the most exposed jobs are falling behind their peers.

Products Worth Considering

Amazon Kindle Paperwhite Signature Edition (12th Generation, 2024), 32 GB, Metallic Raspberry

AmazonView Latest Price

Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Twilight Blue

AmazonView Latest Price

Amazon Kindle Paperwhite (2024, 16 GB) – 7-inch Glare-Free Display, Black

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

Drucker, 1959: the knowledge worker

The first to put a name on it was Peter Drucker, the Vienna-born management writer. In 1959, in a book called The Landmarks of Tomorrow, he described the rise of what he called knowledge work. Drucker is credited with the term knowledge worker: someone who earns a living not with their hands, but with what they have learned. As the Harvard Business Review later summarised it, Drucker had been predicting since at least 1959 an age in which people would create value with their minds more than with their muscles.

His point was simple and radical. If more and more people create value with their minds, the rules that made factories productive, like the assembly line we saw in episode 5, would not be enough. Managing knowledge workers would need completely new rules.

Machlup, 1962: counting what has no weight

Three years later, the economist Fritz Machlup, by then at Princeton, tried to measure the new economy. His problem: knowledge has no weight. You cannot count it in tons, like steel or wheat. So he counted what it cost.

In The Production and Distribution of Knowledge in the United States (1962), Machlup sorted the “knowledge industries” into five branches:

• Education: schools, universities and training.
• Research and development.
• Media of communication: from newspapers and books to radio and television.
• Information machines: from office machines to computers.
• Information services.

And because much knowledge is produced inside companies, he also looked at occupations: the clerks, managers, engineers and other white-collar workers whose work is producing and passing on information.

The number to remember: 29%

His result for 1958: about $136.4 billion, “almost 29 per cent of adjusted GNP”. Education alone accounted for 44% of that ($60.2 billion), media for 28% ($38.4 billion), information services for 13%, research and development for 8% and information machines for 7%, according to the breakdown tabulated by the historian of statistics Benoît Godin. The knowledge industries had grown by 8.8% a year between 1947 and 1958. And by Machlup’s estimate, the share of workers in knowledge occupations rose from about 11% in 1900 to 32% in 1959.

That’s the number to remember: 29%. In the late 1950s, when people still pictured the economy as smokestacks and assembly lines, almost a third of it was already knowledge.

Keep in mind how Machlup got there. He did not measure what knowledge was worth. He measured what it cost, the way national accounts count spending. A brilliant lecture and a boring one cost the same. Hold on to that: tomorrow’s episode is about what the numbers of the economy measure, and what they miss.

Porat, 1977: the four-sector clock

Fifteen years later, Marc Porat, working on a study for the US Department of Commerce, went one step further. He sorted every worker by what they actually do all day, not by where they work. A clerk in a steel mill? Information worker. A bookkeeper on a farm? Information worker. With that, he split the American work force into four sectors: agriculture, industry, services and information.

The result, from his 1977 report The Information Economy:

• 1860: agriculture 40.6%, industry 37.0%, services 16.6%, information 5.8%.
• 1900: agriculture 35.3%, industry 26.8%, services 25.1%, information 12.8%.
• 1950: agriculture 11.9%, industry 38.3%, services 19.0%, information 30.8%.
• 1970: agriculture 3.1%, industry 28.6%, services 21.8%, information 46.4%.

By 1970, information was the biggest group of all, bigger than industry and bigger than services. According to Porat, the information sector had been the largest group since the mid-1950s, and in 1967 information workers earned 53% of all labour income. Measured by value added, his primary and secondary information sectors together made up about 46% of US GNP in 1967 (25.1% plus 21.1%).

Field, factory, services, and now a fourth sector: knowledge and information. Economists call it the quaternary sector.

A note on definitions: because Porat sorts people by occupation, his “agriculture” and “industry” are smaller than in the industry-based sector clocks of earlier episodes. A typist at a farm cooperative counts as information here, but as farming in an industry count.

From manual to mental, and the computer

Machlup saw where this was going. “There has been a succession of occupations leading” the movement, he wrote: “first clerical, then administrative and managerial, and now professional and technical personnel”, “a continuing movement from manual to mental, and from less to more highly trained labor”.

He also saw the computer coming. Technological progress, he argued, had so far favoured “the employment of knowledge-producing workers”, while unskilled manual labour faced the risk of unemployment. And he hoped that “the demand for more information may partially offset the labor-replacing effect of the computer-machine”. In the following decades, that is roughly what happened: computers spread, and information work kept growing.

Drucker, 1999: the biggest challenge

Forty years after Landmarks of Tomorrow, Drucker came back to his idea. “The most important contribution of management in the 20th century was to increase manual worker productivity fifty-fold,” he wrote in the California Management Review in 1999. The big task of the 21st century would be the same for knowledge workers. And he warned that knowledge-worker productivity was still very low; in some fields, such as hospital nursing or design engineering in the car industry, it was lower than seventy years earlier.

Today: about half of all jobs

Where do we stand now? A good modern proxy for the fourth sector is the share of jobs held by managers, professionals and technicians (major groups 1 to 3 of the international occupation classification). For 2025, ILO data from national labour force surveys show:

• Singapore: 62.0%
• United States: 52.5%
• Germany: 47.9%
• South Korea: 42.3%
• India: 11.3%, where farming still employs millions.

In the rich economies, about half of all jobs now live in the fourth sector, and clerical support work comes on top (8.3% of jobs in the US, 12.3% in Germany). The US Bureau of Labor Statistics, using its own grouping, counts 43.6% of employed Americans in “management, professional, and related occupations” in 2025.

The twist: this time the machines aim at the office

For two hundred years, the machines in this series came for the muscles: the horse in episode 3, the loom in episode 4, the assembly line in episode 5. Generative AI is the first big technology aimed straight at the fourth sector.

• US estimate: a 2023 study by Tyna Eloundou, Sam Manning, Pamela Mishkin and Daniel Rock, researchers at OpenAI, OpenResearch and the University of Pennsylvania (later published in Science), estimated that about 80% of US workers could have at least 10% of their tasks affected by large language models, and about 19% at least half of their tasks. Higher-income jobs were potentially more exposed, not less. The study makes no forecast of when this happens.
• Global estimate: the International Labour Organization estimates that one in four workers worldwide is in an occupation with some exposure to generative AI, 34% of employment in high-income countries and 11% in low-income countries (Working Paper 140, May 2025). The most exposed group: clerical work, exactly the jobs that led Machlup’s list. The ILO expects transformation of jobs to be more likely than replacement.

These are estimates of exposure: shares of tasks a language model could speed up. They are not counts of jobs that disappear.

Myth check: “Knowledge workers are safe from machines”

Verdict: false, at least when it comes to exposure.

• The exposure estimates point the other way. Every serious estimate puts office and knowledge jobs at the front of the line this time: higher-income jobs are more exposed to language models, and clerical work is the most exposed group of all.
• But exposure is not job loss. The Yale Budget Lab’s tracker, updated on 15 September 2026 with August data, finds that “the occupational mix is not yet changing in ways that clearly align with the introduction of AI”, and no clear evidence of AI-related disruption of the overall US job market.
• The first crack is at the entrance. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen of Stanford’s Digital Economy Lab, using payroll data through June 2026, find that employment of 22-to-25-year-olds in AI-exposed occupations is now 19% below where it would be had it kept pace with less-exposed peers, mainly through less hiring. Experienced workers show no comparable gap. The authors call these descriptive indicators, not causal estimates, and find no evidence of economy-wide job displacement. (We looked at the same squeeze at the entrance in the UK: youth unemployment at 15.6% and AI skills as the entry ticket.)

What it means for businesses

• Drucker’s challenge is finally on the table. As I read it, AI can make knowledge work faster, but only if you know what the work is for. Define the task before you automate it.
• Measure results, not activity. Hours, emails and documents produced were always weak measures of knowledge work. With AI, they become meaningless.
• Protect the entry level. If AI takes the junior tasks, you still need a way to grow the seniors of 2035. Apprenticeship in the office has to be designed on purpose now.

What it means for you

• Know which part of your work is most exposed. Routine information handling (copying, summarising, checking against a rule) is where language models are strongest.
• Move towards judgment. The safer part of knowledge work is judgment, responsibility and trust: the decisions someone has to answer for.
• Learn the tools. The people who use them well are the ones who get to decide what they are used for.

Thorsten’s thesis: what the fourth sector says about it

My thesis for this season is that automation could become its own sector, up to 90% of the economy within a decade or a few decades. That is my thesis, not a fact, and this season tests it with evidence for and against.

The fourth sector is evidence for it: information work is already digital, and for the first time a machine can read and write. It is also evidence against it: when computers spread from the 1960s on, information work kept growing, just as Machlup hoped, and so far the job data shows no clear AI footprint across the whole economy. Episode 47 will test the thesis in full.

Arc 1 in one sentence

Field, factory, services, and now knowledge: in eight episodes, we followed work from the farm (episodes 2 to 4) through the factory (5 and 6) to the office (7 and 8). Tomorrow, arc 2 begins with the basics everyone needs, starting with the number that is supposed to measure all of it.

Limits of this episode

• Drucker’s 1959 text. We could not check the 1959 book directly. The attribution of “knowledge worker” to The Landmarks of Tomorrow follows the Drucker School at Claremont Graduate University; some sources date the exact phrase to later books.
• Machlup’s numbers. His figures are cited as tabulated and quoted by Godin (2008) and as summarised by Princeton University Press. They measure spending, not value; the growth rate is in current dollars.
• Different definitions. Porat sorts the work force by occupation, so his shares are not comparable with industry-based sector shares; the ILO 2025 shares count employed persons by occupation group, mapped from national surveys.
• Estimates, not forecasts. AI exposure studies estimate which tasks a model could affect; they do not say when, or whether jobs disappear. The Stanford and Yale findings are descriptive and may change with new data.
• Interpretations. The lessons for businesses and for you are my reading of this history.

Series navigation

• Previous episode: Episode 7, How 80% Ended Up in Services.
• Next episode (tomorrow): Episode 9, What GDP Measures and What It Doesn’t. Machlup could only measure knowledge by what it cost. So what happens to the numbers when knowledge becomes almost free?

Sources

• Peter F. Drucker, The Landmarks of Tomorrow (Harper & Brothers, 1959); attribution of “knowledge worker” per Drucker School of Management, Claremont Graduate University, “The Drucker Difference: What’s Next for Knowledge Workers?” (5 December 2022). cgu.edu · Harvard Business Review, “What Peter Drucker Knew About 2020” (2014). hbr.org
• Peter F. Drucker, “Knowledge-Worker Productivity: The Biggest Challenge”, California Management Review 41(2), Winter 1999, pp. 79–94. cmr.berkeley.edu
• Fritz Machlup, The Production and Distribution of Knowledge in the United States (Princeton University Press, 1962), figures and quotations as tabulated and quoted in Benoît Godin, “The Knowledge Economy: Fritz Machlup’s Construction of a Synthetic Concept”, Project on the History and Sociology of S&T Statistics, Working Paper 37 (2008). inrs.ca (PDF)
• Marc Uri Porat, The Information Economy: Definition and Measurement, OT Special Publication 77-12(1), and National Income, Workforce, and Input-Output Accounts, OT SP 77-12(8), Table 15, US Department of Commerce, Office of Telecommunications (May 1977). ERIC ED142205 (PDF) · ERIC ED142212 (PDF)
• International Labour Organization, ILOSTAT: employment by sex and occupation (ISCO-08), 2025; own calculation of the share of major groups 1–3. ilostat.ilo.org
• US Bureau of Labor Statistics, Current Population Survey, Table 9 (2025 annual averages; October 2025 not collected). bls.gov
• Tyna Eloundou, Sam Manning, Pamela Mishkin, Daniel Rock, “GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models”, arXiv 2303.10130 (2023); Science 384 (2024). arxiv.org
• Paweł Gmyrek et al., “Generative AI and Jobs: A Refined Global Index of Occupational Exposure”, ILO Working Paper 140 (20 May 2025). ilo.org
• Erik Brynjolfsson, Bharat Chandar, Ruyu Chen, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”, Stanford Digital Economy Lab (2025, revised 12 August 2026). stanford.edu
• The Budget Lab at Yale, “Tracking the Impact of AI on the Labor Market” (published 16 July 2026, updated 15 September 2026). budgetlab.yale.edu

This article is for information only. It is not career, legal, financial or investment advice.

Researched and drafted with AI assistance, including Anthropic’s Claude.