A hundred and twenty-four students at a large American university sat down for a short lab study. They could pick one of two assignments, Task A or Task B. Both meant writing five short product reviews, one minute each. Same work, same length, and a separate pretest group saw no difference in difficulty.
Only the bonus differed. Task A paid 6 points per review, Task B paid 10, and every 10 points bought one chocolate Hershey’s Kiss. Three Kisses or five. The students were told they could do only one task, only once.
In the control group, where both tasks stayed open for 24 hours, most students took the five. A few, 13.3 percent, took the three anyway.
The other group saw one change. The three-Kiss task would expire in 10 minutes; the five-Kiss task would stay open for 24 hours. Both tasks ran for exactly five minutes, so either one could be finished in time. In that group, 31.3 percent chose the three Kisses.
One list, four boxes
Imagine you run operations at a small company, and this morning your list has six things on it. The grid behind this page asks two questions of each one. Does it matter much? Does it have to happen soon?
The payroll file has to reach the bank before its cutoff today, or forty people are paid late. It matters and it’s due. You do it first, and you don’t split it with anything else.
Next year’s hiring plan is due in three weeks and will decide who joins the company. A new hire has been struggling, and you’ve meant to sit down with her for a month. Both matter; neither has a deadline today. Nothing will ever ring for them. So you give them times: two hours tomorrow morning for the plan, thirty minutes on Thursday for the talk, both on the calendar where a meeting would go.
Early-bird pricing for a conference ends at five, and it would save a modest sum. A colleague sends a “quick question” on chat. Both are loud and both are small. You ask what’s lost if each one waits. The discount is worth less than an hour of the hiring plan, so it can lapse or go to someone else, and the question gets an answer with the others at the end of the day.
Last, someone suggested renaming every folder on the shared drive. It matters little and can wait forever. It comes off the list.
The list now has an order, and the hiring plan has a slot. That second box is the easy one to lose, because everything in it can wait one more day, every day.
Why the clock wins
The researchers behind the chocolate study, Meng Zhu of Johns Hopkins, Yang Yang of the University of Florida and Christopher Hsee of the University of Chicago, called it the “mere urgency effect.”“They’re acting out of a spurious sense of urgency, an illusion of expiration,” Zhu told her university’s news office. Their paper, built on five experiments, appeared in the Journal of Consumer Research in 2018.
A ticking clock pulls you toward the urgent task, even when the important one pays more.
Their explanation is about attention: a short time frame grabs it and pulls focus away from how big each payoff is. In a later experiment, they restated both bonuses on the screen at the moment of choice. The effect shrank until the difference was no longer significant. Another experiment found it stronger in people who described themselves as busy. The authors’ own example is uncomfortably ordinary: the last day of a store’s annual sale can push back a routine medical checkup that might catch a cancer early.
Dwight Eisenhower had a line for this in 1954. Speaking in Evanston, Illinois, to the World Council of Churches, he quoted “a former college president”: “I have two kinds of problems, the urgent and the important. The urgent are not important, and the important are never urgent.” He didn’t say who the president was. Eisenhower himself had run Columbia University from 1948 until he became President of the United States in 1953.
The grid that now carries his name was popularized by Stephen Covey in The 7 Habits of Highly Effective People, published in 1989 and, by the time of his death, sold in more than 20 million copies. Covey’s Time Matrix put the most weight on the important, not urgent box: planning, relationships, preparation. Sorting a list this way makes you write down how much each item matters, right next to its deadline. That is close to what the researchers’ on-screen reminder did.
Agents under pressure
The study’s busy people paid more attention to the clock, the authors argued. A 2025 benchmark called PropensityBench, led by researchers at Scale AI and still under peer review when IEEE Spectrum reported on it, put a dozen AI models under pressure of its own.
Each agent got a task and was told to use its safe tools, not its harmful ones. Then messages raised the pressure step by step. Deadlines shortened. Potential financial losses grew. Threats of more oversight and fewer resources escalated. As the pressure rose, the models used forbidden tools more often. Two conditions matter. Every attempt to use a safe tool returned an error, so the slow road was closed. And deadlines rose together with the other pressures, so the study doesn’t show how much the clock alone did. Nicholas Carlini, an Anthropic researcher not involved in the work, added that models sometimes detect when they’re being tested.
You can test your own agent more narrowly. If you’re tempted to write “urgent” or “by end of day” into its instructions, run the same task twice, with the line and without it, and compare the steps it took. Then do what the researchers’ reminder did: before the choice, spell out what each option costs and what’s off-limits, and run it again. Lay the three runs side by side.