The correspondents at a very large corporation spent their days answering letters from shareholders. On paper it was a job for good people. They were carefully selected and highly trained, and the work looked complex and challenging.
In practice, the supervisor answered every difficult question, signed and checked every letter, and all the mail crossed the supervisor’s desk. They were constantly reminded how many letters still needed answers. Nearly every performance and attitude measure was low, and exit interviews confirmed, as Frederick Herzberg put it, that the challenge of the job “existed merely as words.”
Some of the ideas for improving the job were the obvious ones. A daily quota, set hard to reach. Typing as well as writing the letters. Sending the difficult inquiries to a few people so the rest could turn out more.
Every one of those ideas was rejected.
What changed hands
Instead, over two months, one group’s job changed about once a week. The correspondents signed their own names. Experienced writers had their letters checked one time in ten instead of every time, at their own desks. Mail went straight to the mailroom. Subject experts inside the group became the first stop for hard questions, before the supervisor. Each correspondent was now personally responsible for the quality and accuracy of their letters, a job that had belonged to the supervisor and the verifier. Pay stayed where it was, apart from raises that would have come anyway.
For the first months, the group got worse. Its performance index kept falling, which Herzberg put down to “uncertainty after their newly granted responsibilities.” In the third month it turned. By the end of six months the group was outperforming a comparison group that kept the old job, liked its work markedly more, was absent less, and later saw a much higher rate of promotion. Two extra groups, set up to test whether attention alone explained it, came out substantially the same as the control group, he reported.
Herzberg told the story in the Harvard Business Review in 1968, to show his Two-Factor Theory at work. In that theory, pay and conditions can stop people being unhappy, but only the work itself gives them a reason to try. He printed the rejected ideas and the adopted ones side by side.
To make a job worth doing, give people more say over it, not more of it.
Zero times zero
Herzberg called the rejected column horizontal loading. It made a job “structurally bigger.” Ask someone who tightens 10,000 bolts a day to tighten 20,000, and “multiplying zero by zero still equals zero.” Add a routine clerical task, and you are adding zero to zero. Taking away the hardest parts so people can do more of the easy ones is “subtraction in the hope of accomplishing addition.”
The adopted column was vertical loading: fewer controls with the accountability kept, a complete unit of work, results reported to the worker instead of the supervisor, harder tasks, room to become an expert. That was job enrichment. The older term, job enlargement, was to be avoided.
Ford’s Highland Park plant had run the opposite experiment. One man’s magneto assembly was split into 29 operations, and by Ford’s own count the company was hiring about 53,000 people a year to keep 14,000. The five-dollar day of 1914 made more people willing to stay, but as The Henry Ford museum puts it, “The work in the factory was no easier, the pace no less relentless.”
Herzberg warned of a drop in the first weeks, and of supervisors who might turn anxious, even hostile, as staff took over what they saw as their own responsibility. “Not all jobs can be enriched,” he wrote, “nor do all jobs need to be enriched.” And this is his own account of an experiment at a company he does not name.
The analyst and the quotes
Say Mei runs a small pricing team and reads every customer quote her two analysts prepare before it goes out. Her analysts are bored. She could give them more quotes, or add the monthly data cleanup. In Herzberg’s arithmetic, that is zero plus zero.
Or one analyst could own one region’s quotes from first call to signature, send them under her own name, and see the win and loss figures herself. Mei could read one quote in ten. The first weeks might be messier, and Mei might wonder what her job is now. The hours she spent reading every quote could go to work she never got to.
The agent that checks its own letters
The same move carries over to AI agents, for a different reason. Anthropic’s guide to its coding agent, Claude Code, describes something close to the old correspondents’ office. “Claude stops when the work looks done. Without a check it can run, ‘looks done’ is the only signal available, and you become the verification loop: every mistake waits for you to notice it.”
Its advice is to hand the agent the check: a test suite, a build, a screenshot to compare against a design. The agent then “does the work, runs the check, reads the result, and iterates until the check passes.” That is Herzberg’s report going straight to the worker. The guide still tells you to verify the change yourself afterward, and suggests a separate verification agent “so the agent doing the work isn’t the one grading it.” One letter in ten, not none.
Anthropic’s post on agents describes the alternative, a fixed chain of small steps with checks between them, as ideal when a task “can be easily and cleanly decomposed into fixed subtasks.” That is a different job design, not a wrong one.
What changes is the payoff. Herzberg measured how much more the correspondents liked their work. The payoff Anthropic’s guide describes is mechanical: mistakes get caught without waiting for you.
At that very large corporation, the correspondents’ letters went straight to the mailroom, with their own names at the bottom.