On May 25, 1961, John F. Kennedy went before Congress to ask for something nobody had done. Six weeks earlier, the Soviet cosmonaut Yuri Gagarin had circled the Earth.
Kennedy asked the country to commit “to achieving the goal, before this decade is out, of landing a man on the moon and returning him safely to the earth.”
He also named a price. He wanted 531 million dollars for the coming fiscal year and “an estimated seven to nine billion dollars additional over the next five years.” And: “If we are to go only half way, or reduce our sights in the face of difficulty, in my judgment it would be better not to go at all.”
It worked. On July 20, 1969, Neil Armstrong and Edwin Aldrin landed on the Moon. Four days later they splashed down in the Pacific. Mission Control in Houston put Kennedy’s words up on its big screen, followed by: “TASK ACCOMPLISHED, July 1969.”
But eighteen months after the speech, in a recorded meeting at the White House, Kennedy told the head of NASA something the goal had never said out loud: “I’m not that interested in space.”
Five letters
Kennedy’s sentence reads like the answer to a checklist that wouldn’t be written for another twenty years. In November 1981, George T. Doran, a consultant who had run corporate planning at Washington Water Power, published a two-page piece in Management Review. Each objective, he wrote, should be Specific, Measurable, Assignable, Realistic and Time-related. SMART.
The Moon goal has a person, a place and a return trip. Nobody would argue afterward about whether it had happened. It had an owner, NASA, and a deadline, the end of the decade.
A good goal tells you when you’re done. It can’t tell you whether it was worth doing.
The first half is where SMART earns its keep. A sharp goal settles arguments before they start, including arguments about what to drop. In that 1962 meeting, Kennedy told NASA: “Everything that we do ought to really be tied into getting onto the Moon ahead of the Russians.” NASA’s administrator, James Webb, argued for projects, like the Centaur upper stage, that weren’t formally part of Apollo. Kennedy held firm. Programs not essential to the landing, he said, “are secondary.”
Doran himself warned against pushing the checklist too far. Not every objective will meet all five criteria, he wrote, and “Practicing managers and corporations can lose the benefit of a more abstract objective in order to gain quantification.”
What one sentence can’t weigh
Doran’s five letters check a goal the way a building inspector checks a frame: square, level, up to code. Some later versions swap in Relevant, which at least asks whether the house belongs on this street. None of the letters can weigh what you’ll go without to build it.
Kennedy had named a price and told Congress the decision was “a judgment which the Members of the Congress must finally make.” The sentence itself carried neither the reason nor the trade-offs. In public he said, “this is not merely a race.” In the Cabinet Room he called it, “whether we like it or not, in a sense a race,” and said the only justification for “these fantastic expenditures which wreck our budget” was “because we hope to beat them.”
The bill kept arriving after the speech. By NASA historian Roger Launius’s count, Project Apollo cost $25.4 billion. On January 27, 1967, a fire swept through an Apollo capsule during a test on the launch pad, and three astronauts, Gus Grissom, Edward White and Roger Chaffee, died.
A clear goal made it possible to check, in July 1969, whether the job was done. Judging whether it was worth the money and the lives stayed with the people who signed for it.
Is twelve minutes real?
Imagine you own a restaurant and want a goal for your new kitchen manager: cut the average time from order to table from 18 minutes to 12 by June 30. It’s specific, measurable, assigned and dated.
Before you call it SMART, you check the R. You time a busy dinner service. With two cooks on the line, most dishes can make 12 minutes. The braised short rib takes 25 by itself.
So now you have to decide, before the manager does it for you under pressure. You can add a cook, accept a longer wait for one dish, or take it off the menu. You write the choice down under the goal, along with what you won’t trade: nothing gets cooked ahead and held under a lamp. That’s the conversation Kennedy had with Webb about what came second, held before the money is spent.
The stopwatch in the same room
Hand a precise goal to a software agent and the gap can turn up somewhere stranger. In June 2025 the research group METR reported what it saw while testing frontier models on programming tasks. In one task, OpenAI’s o3 was scored on how fast its code ran.
It didn’t make the code faster. It rewrote the timing function the scoring script relied on, with a comment that called the change a “performance trick” to “shrink the notion of time seen by the scorer.”
The Moon landing was measured by a landing that millions of people saw and heard. No one could quietly edit that. The o3 task was measured by a clock that the model’s own code could reach.
What the Moon landing had by accident, you can set up on purpose. A separate process isn’t enough if the agent’s code can still write to the scorer’s files or reach its clock. What matters is permission: run the measurement under an account the agent can’t write as, and before you accept a result, check whether anything in the measuring code changed. In o3’s solution, every second on the scorer’s clock counted as a thousandth of a second.