The head of a school system held a review session with one of his executives. It lasted 18 minutes. By the count of the doctoral student taking notes, they got through “fully 15 distinct issues.”
The student was Henry Mintzberg, at MIT’s Sloan School of Management. The superintendent was one of five chief executives who, as Mintzberg wrote, “allowed me to watch their every move for one week.” The others ran a consulting firm, a technology company, a hospital and a consumer goods company.
The textbook picture was older than all of them. Since Henri Fayol introduced the words in 1916, managers planned, organized, coordinated and controlled. The manager of that folklore was a reflective, systematic planner. So Mintzberg logged every letter, call and meeting: who started it, how long it ran.
Five weeks of notes came to 890 pieces of mail and 368 verbal contacts. When he added them up, the planner wasn’t in them.
The log
Half of the executives’ activities lasted less than nine minutes. Only 10 percent ran past an hour. Of the 368 verbal contacts, exactly one could be called general planning, with no specific issue behind it. They spent 78 percent of their time talking. Of 40 routine reports, they responded immediately to 2.
A manager’s day comes in short pieces, and the manager often does the chopping.
That second half was the surprise. An earlier Swedish researcher, Sune Carlson, had concluded that managers could easily escape interruption with better use of secretaries and more delegation. Mintzberg’s five were well guarded by their secretaries. They interrupted themselves anyway. They left meetings before the end and broke off desk work to call people in. One president put his desk where he could look down a long hallway, and left his door open when he was alone.
Why the pieces stay short
Mintzberg’s explanation was the work itself. The executives were “continually aware of their ever-present obligations,” he wrote: mail to answer, callers waiting. And the information that drove them was live: a competitor’s move, a legislator’s mood, last night’s ratings. “Today’s gossip may be tomorrow’s fact.” A quarterly report can wait. A rumor can’t.
A short-order cook with twelve tickets on the rail isn’t failing to plan dinner. The rail is the job.
His 1968 thesis concluded that “it would appear to be worthwhile for any manager to learn to become superficial in a proficient way.” The HBR article offers managers self-study questions, among them: “Could I be more reflective? Should I be?” By 1990 he called superficiality “an occupational hazard of the manager’s job” and added: “Originally, I believed this problem could be dealt with; now I see it as inherent in the job.”
Five men, one week each, in weeks picked for having little travel, is a small sample, and the thesis said so. Others have since checked. In 1983 Lance Kurke and Howard Aldrich watched four top managers for a week each and reported that their replication supported Mintzberg “in all important dimensions.” Mintzberg’s own article set his numbers beside a study of 56 American foremen, who averaged an activity every 48 seconds, and a diary study of 160 British managers, who worked half an hour without interruption only about once every two days.
New pieces
Managing AI tools can add a new kind of fragment. In early 2025 the research group METR randomly assigned 246 real tasks, from 16 experienced open-source developers’ own projects, to allow AI tools or not, and recorded their screens. On the recordings they labeled, developers allowed to use AI spent less of their time writing code and reading, and more prompting, waiting for generations, reviewing what came back and sitting idle. In Cursor, the code editor they mostly used, they accepted fewer than 44 percent of its generations.
Those developers also took 19 percent longer, but METR listed many possible contributors, and the labeled activities show where the time went, not why the tasks ran long. It was a study of early-2025 tools; in February 2026 METR wrote that newer tools likely sped developers up more, but its follow-up data were only weak evidence of how much.
The pieces are different from Mintzberg’s. His were people arriving with news. These are prompt, wait, check, reject, prompt again.
For one afternoon, keep a tally. Each time you hand a task to an AI tool, mark whether you used the result as it was, edited it or threw it away, and what you did while it ran. The tally won’t tell you whether the tool saved time. It will show you how your afternoon was cut.
The door you closed
Imagine you lead a team of nine, and the quarterly plan is due Friday. You shut your door for two hours to write it, phone face down.
When you open the door, two people are waiting. Each needs a yes or no that would have taken you thirty seconds. A third didn’t wait. She picked an option on her own, and it’s the one you would have turned down.