The History of Knowledge Work: From Sumerian Scribes to Drucker and the AI Desk

By Victor Hoang · Updated Oct 23, 2026

Picture a modern office. Nobody can see a decision being made. So managers count what they can see: hours, emails, meetings. That same trap is built into the earliest records we have. They counted grain and stone because grain and stone could be counted.

People have earned a living by recording, counting and advising for about 5,000 years. The first Sumerian administrators pressed wedges into clay around 3200 BCE. But the idea of a "knowledge worker" is only a few decades old. Peter Drucker introduced the phrase in Landmarks of Tomorrow, a book dated 1957 or 1959 depending on the edition. The throughline is that the work came first, the name came late, and the measuring tools never caught up with either.

Timeline at a Glance

The Scribes' World (c. 3200 BCE-1860s)

Clay and papyrus create the first paid record-keepers

Picture a temple storehouse in Uruk around 3200 BCE: clay tablets, a stylus, sheep and grain waiting to be counted. Cuneiform was first developed around 3600/3500 BCE, and in its earliest form it served an exclusively administrative function. By the Early Dynastic period (about 2900 to 2350 BCE), scribal schools called edubba, "House of Tablets," had been set up to teach the craft. These scribes are among the earliest documented workers whose product was records.

Egypt left a closer look at the daily routine. In 2013 a French-Egyptian team found hundreds of papyrus fragments at the Red Sea harbor of Wadi el-Jarf, from the end of Khufu's reign (about 2551-2528 BCE). Ancient Egypt Research Associates describes them as the oldest papyrus archive found in Egypt: accounts of goods delivered to workers, and logbooks of their daily activities. The logbooks follow Inspector Merer, whom the team leaders Pierre Tallet and Gregory Marouard described in 2014 as "in charge of a team of about 200 men", hauling limestone from the Tura quarries to the Great Pyramid. A later reading, reported in the London Review of Books, puts Merer's own team at forty men, so crew sizes vary by account. The entries are laid out as a timetable with two columns per day. That is a work log, kept by someone whose job was to record other people's labor.

The profession had a recruiting pitch. In the Middle Kingdom (2040-1782 BCE), the Satire of the Trades shows a father sailing his son Pepi to school and spending most of the text presenting every other job as misery. Most of the skilled professions required literacy, which meant first becoming a scribe, and that took many years of training.

China built a different system around the same need. Choosing officials by recommendation had favored elite clans. Around 605 CE, in the Sui dynasty, the jinshi examination was institutionalised, and by the Tang it had grown into the full keju system (some scholars therefore date the system to the Tang). From the Ming, it tested mainly the Confucian classics. It lasted until September 1905, when the court ended it, roughly 1,300 years in all.

Medieval England shows the clerk at a desk with a tool. The Dialogue Concerning the Exchequer, written around 1180, describes a table covered with a striped cloth, with counters placed on it by value. The treasurer and the sheriff, who came to render account, sat across from each other and settled accounts while the other officials watched, a scene the text compares to a game of chess. A clerk of the treasurer is listed among the officers present.

These were precursors, not the concept. Nobody in Uruk, Giza, Chang'an or Westminster called this "knowledge work", and most of these records are about counting goods, not judging ideas. What they share is an institution that pays for records and then wants a way to check them.

The next turn this article can document came with a machine, in the 1870s.

What it means for teams today: a log of what happened and a way to check it are the oldest management tools there are. If your team can't show what it decided last month, it is working like a storehouse without tablets.

The Typewriter Office (1870s-1950s)

A machine and a bigger company turn paperwork into a mass job

Imagine a Milwaukee workshop in 1873, where Christopher Latham Sholes and his partners are building a machine with keys that strike paper through an inked ribbon. The Sholes & Glidden "Type Writer" was designed in 1873 by Sholes with Carlos Glidden, Samuel Soulé and Mathias Schwalbach and made by E. Remington and Sons. It was the first commercially successful device to print characters rapidly on paper in any order, and it created a demand for typists that let women enter the business world in unprecedented numbers. The machines that later replaced much of that typing are the subject of our history of office software.

The scale shows up in labor statistics. The Bureau of Labor Statistics reports that professional, managerial, clerical, sales and service workers grew from one-quarter to three-quarters of total employment between 1910 and 2000. Clerical workers alone went from 5.2 percent of employment in 1910 to 19.3 percent in 1980, or from 2,026,000 to 18,758,000 people. The BLS authors tie the growth to larger, more complex business, government and nonprofit organizations with more reports, transactions, records and correspondence to handle.

Management had by then learned to study factory work. Drucker later wrote that Taylor was the first person to work as a manual worker and then study manual work, in the 1999 article discussed below. Our history of scientific management covers how that approach spread, and the story of the time clock shows how attendance became a stand-in for effort.

The new problem was that clerical work could be counted, so it was. Pages typed and forms filed were easy numbers. Judgment, the part that later defined the professional, stayed invisible.

What it means for teams today: if the metric you can easily count is "tickets closed" or "emails sent", ask what the equivalent of "pages typed" is in your team, and what it is hiding.

The Naming Years (1957-1973)

Drucker, Machlup and Bell give the workers a name and a size

The name came first. Peter Drucker's Landmarks of Tomorrow is credited with introducing the term "knowledge worker". Claremont Graduate University says Drucker coined the phrase in that book. The date varies by source: the Drucker Institute's bibliography lists the book as 1957, while the Internet Archive catalogues a 1959 edition and most citations use 1959.

Meanwhile, an economist was putting numbers on the same shift. Fritz Machlup's The Production and Distribution of Knowledge in the United States (Princeton University Press, 1962) counted education, research and other knowledge activity as an industry. Lewis Coser's 1963 review reports Machlup's finding that total annual knowledge expenditure reached $136,436 million in 1958, equal to 29 percent of adjusted gross national product. It also reports that knowledge-producing occupations rose from 10.7 percent of the labor force in 1900 to 31.6 percent in 1959. The review notes that Machlup counts even telling anybody anything as knowledge production, so the share depends on a generous definition.

The sociologist Daniel Bell widened the frame. His 1973 book The Coming of Post-Industrial Society (Basic Books, built on research from the 1950s) identified five trends: a service economy, the pre-eminence of the professional and technical class, the primacy of applied theoretical knowledge, the planning of technological growth, and a new "intellectual technology". One later account calls its central idea a society organized around knowledge and services rather than labor and manufacturing, and notes Bell popularized the term "post-industrial" without coining it.

Three different people, working in management, economics and sociology, reached the same conclusion within about fifteen years. What none of them could yet say was how a manager should raise the output of the people they described.

What it means for teams today: naming a kind of work is the first step to managing it. If a role on your team has no clear definition of its output, you are managing it by feel.

Drucker's Challenge (1980s-2005)

The office peaks, then management asks what the task is

The routine office job reached its high point as personal computers spread. The BLS data show the clerical share peaking at 19.3 percent in 1980 and falling to 17.4 percent by 2000, while computer specialists, a category that did not exist in 1910, grew sharply from 1960. Our history of productivity measurement covers how output gets counted.

In the Winter 1999 issue of California Management Review, Drucker published "Knowledge-Worker Productivity: The Biggest Challenge". He argued that the 20th century's great management achievement was a fifty-fold rise in manual-worker productivity, and that the most valuable asset of a 21st-century institution would be its knowledge workers and their productivity. He listed six factors. The first is a question: "What is the task?" In manual work, he wrote, the task is always given, but in knowledge work someone has to define it. A second factor is that knowledge workers have to manage themselves.

His own example was hospital nurses. They named "chores" such as paperwork, arranging flowers and answering patients' relatives' calls as the things that made them unproductive. When those were handed to a non-nurse floor clerk, Drucker wrote that the nurses' productivity, measured as time at patients' beds, immediately more than doubled. He also admitted how early the field was: work on this productivity, he said, was in 2000 roughly where manual-worker work had been in 1900.

Six years later Thomas Davenport published Thinking for a Living (Harvard Business Press, September 13, 2005). The publisher's description argues that companies were still managing knowledge workers with techniques designed for the Industrial Age.

Drucker is also the usual bridge to the rest of this history; see our profile of Peter Drucker's leadership, and the wider history of productivity for the other eras that led here.

What it means for teams today: before adding a tool or a metric, ask Drucker's question. Then find the chores that someone with a more expensive skill is doing, as the nurses were.

The Distributed Age (2013-present)

Remote work, collaboration overload and AI reach the desk

In 2013 a randomized experiment gave office work its first hard remote-work numbers. Bloom, Liang, Roberts and Ying had volunteer call-center staff at Ctrip work from home or the office for nine months. Home workers performed 13 percent better, about nine points from longer shifts and about four from more calls per minute. In 2024 the same lead author published a larger trial at a Chinese technology company: a six-month randomized trial with 1,612 employees found that hybrid working cut quit rates by one-third and had no effect on performance grades over the next two years.

The cost of working together was getting attention too. In a 2016 Harvard Business Review article, Rob Cross, Reb Rebele and Adam Grant wrote that, according to data collected over two decades, time spent in collaborative activities had ballooned by 50 percent or more. Our guide to remote work productivity covers the working practices that grew out of this.

Generative AI is the first technology aimed directly at the tasks the earliest scribes did: drafting, summarizing, classifying and answering. Eloundou, Manning, Mishkin and Rock estimated in a 2023 paper that around 80 percent of the US workforce could have at least 10 percent of its tasks affected by large language models, and about 19 percent at least half. That is a measure of exposure, and the authors make no prediction about timing. A field study of 5,179 customer support agents found that an AI assistant raised issues resolved per hour by about 14 percent on average and 34 percent for novices, with minimal impact on experienced workers. The wider evidence is in how AI is changing work.

The 5,000-year-old problem is still here in new form. AI can produce a draft in seconds, but someone still has to decide what is worth drafting and whether the draft is right, the same job as the clerk checking the scribe.

What it means for teams today: judge remote, hybrid and AI changes by outcomes you can name, such as quit rates, resolution times and error rates, rather than by hours online.

Key People

Key Facts

Key Facts: Knowledge Work

• Clerical workers grew from 5.2 percent of US employment in 1910 to 19.3 percent in 1980, then eased to 17.4 percent in 2000.
• Machlup's study put knowledge expenditure at 29 percent of adjusted GNP in 1958, under a generous definition.
• Drucker's 1999 article said raising knowledge-worker productivity starts with the question "What is the task?"
• Hybrid working cut quit rates by one-third in a 1,612-person trial, with no effect on performance grades.
• About 80 percent of US workers could have at least 10 percent of tasks affected by large language models, in one exposure estimate.

The Pattern Behind the History

We call this the Record, Count, Name, Question Cycle. It is our own synthesis of this timeline, not a published model.

1. The record comes first. Tablets, logbooks and exchequer rolls existed before anyone studied the work behind them.
2. The countable gets counted. Grain, stone, pages typed: managers measure what is visible.
3. The name arrives late. "Knowledge worker" appeared only in the late 1950s, long after the workers did.
4. The real question is the task. Drucker's 1999 point is that someone must define the task before anyone can improve it.
5. Each new tool restarts the cycle. Typewriters, computers, remote work and AI all create new records and new things to count.

What It Teaches Teams Today

• Define the task first. Write down what each role is supposed to produce, in outcomes, before measuring anything.
• Beware the easy number. Counts of emails, tickets and hours are modern pages typed.
• Remove the chores. Hand routine admin to cheaper roles or tools, as the nurses' floor clerk did.
• Treat collaboration as a cost. Meetings and messages use up the same attention as everything else.
• Test AI where you can measure it. Pick one workflow, record results before and after, and keep the human check.

For the related stories, see the history of the 8-hour workday, the productivity timeline, and the broader history of management. If you want the plain definition first, start with what productivity is.

Knowledge work is old, and the name is young. The teams that handle it well keep Drucker's habit: before measuring or automating anything, ask what the task is.