On Saint Patrick’s Day 1970, an Eastern Air Lines shuttle from Newark was nearing Boston when a young man with a revolver walked into the cockpit. He shot both pilots. The copilot died. The captain, bleeding from both arms, landed the plane at Logan Airport. One of the passengers, by her own account, was 19.
In 2009 came the “Miracle on the Hudson.” It set her thinking, she later wrote, about how miraculous her own flight had been. She had never thanked the pilot or the crew.
Thirty-nine years is a long time for a thank-you to wait, and a stranger from 1970 is hard to find. A wish like that can stay a wish.
She was reading a book by David Allen at the time, and she asked herself the question Allen teaches: what’s the next action? By her account, the answer was “Sit down at the computer.”
“Mom” is not a task
A to-do you can’t picture doing tends to wait. Write down the next visible step instead.
That is the core of an idea David Allen laid out in Getting Things Done in 2001. A next action, in Allen’s words, is “the next visible physical activity required to move something forward.” He has written that more than nine in ten of the to-do lists he has seen are “incomplete inventories of still-unclear things.”
His own example is a list with “Mom” on it. Once you decide the outcome is a party for her birthday, the next action is “Call Sis about what we should do for Mom’s birthday.” Because “Mom” is vague, Allen argues, it still triggers stress whenever you read it; the next action “actually finishes the thinking.” And “Call Sis” can be done the moment you’re near a phone.
So Allen’s company recommends keeping next actions on lists sorted by context, the tools, people and places you’ll need, rather than in project files.
The thank-you had been all outcome and no step. “Sit down at the computer” was a step. She searched for retired Eastern pilots, and within the hour, she wrote, she had the captain’s name and address.
What the evidence can carry
Her story is one reader’s letter, which Allen’s company published on its website in 2009. It shows what the question did for her, not how often it works. The research is thin. In 2008, Francis Heylighen and Clément Vidal, researchers at the Free University of Brussels, argued that Allen’s system fits what psychology knows about how people lean on their surroundings to remember and act. They also wrote that there were “as yet no empirical studies” confirming its practitioners’ claims.
The closest lab result comes from Sean McCrea and colleagues, published the same year. Students in Konstanz were asked to email back a short questionnaire within three weeks, for €2.50 or course credit. In the first study, 34 of them wrote about ten everyday activities, such as opening a bank account. Half wrote how one would go about each; half wrote what doing it says about a person. Of the 17 who wrote about how, 16 sent their answers back before the deadline. Of the 17 who wrote about traits, 11 did.
Two more studies, with 50 and 51 students and different ways of nudging people to think concretely, pointed the same way. The samples were small, the task was easy, and none of it tested Allen’s lists. The authors themselves noted that for a task people find unpleasant, a reminder of why it matters might cut delay instead.
The report with no first line
Imagine a team lead whose list has said “Q3 budget” for two weeks. Every morning she reads it and moves on to something she can finish.
When she finally asks what the next action is, the honest answer is not “write the budget.” It’s “email finance for last quarter’s actual spending,” because nothing else can start until those numbers arrive. That takes four minutes, and the item becomes “waiting for finance.” When the numbers come back, she has to answer the question again.
A to-do list for an agent
Software agents face the same gap between a goal and a step. In July 2025, Yichao “Peak” Ji, a co-founder of the agent company Manus, wrote that a typical Manus task takes around 50 tool calls, a long loop in which the model can drift off-topic or forget earlier goals. For complex tasks, he wrote, the agent creates a todo.md file, checks off items as it goes and keeps rewriting it. “By constantly rewriting the todo list, Manus is reciting its objectives into the end of the context,” pushing the plan into the model’s recent attention span.
That’s the company’s description of a design choice, not a measured result. And the list had a price. In October 2025, Lance Martin, who had just held a webinar with Ji, posted his notes from it. According to his notes, Manus found that roughly a third of all actions went to updating the todo list, and it moved planning to a separate planner agent.
Allen’s Two-Minute Rule rests on a similar sum: keeping an item on a list costs something too.
You can try a version yourself. Before an agent starts a vague job, ask it to write the plan as steps, each naming a file, a command or a person, plus how you’d check it was done. When it reports back, pick two ticked items and check them by their effects, the changed file or the test that ran, not by the tick.
As for the captain, by her account, when her letter arrived, his wife told him to sit down and write a nice letter back. He picked up the phone instead.