In October 1964 a new Labour government took office in Britain, set a review of Concorde in hand, and decided to withdraw. The supersonic airliner was being built with France under a treaty signed in 1962. According to a historian writing on the UK government’s history blog, the aviation minister, Roy Jenkins, carried the news to Paris, and the French response was to withdraw all diplomatic communication with London.
On January 20, 1965, Jenkins gave the verdict in the House of Commons. “We had, and we still retain, some doubts about the financial and economic aspects of the project,” he said.
Then he announced the opposite of what his government had set out to do. The Prime Minister had told Paris “that we stand by the treaty obligations into which the last Government decided to enter.” Britain was going ahead.
The plane that named a mistake
Money spent that you can’t get back is gone either way. Let what comes next decide.
Economists call money you can’t get back a sunk cost, and their textbooks tell students to ignore it. What you could still recover, by a refund or a resale, isn’t sunk, and the costs of getting out, like exit fees or a lawsuit, still count. By 1980 the Oxford zoologist Richard Dawkins had a name for the opposite habit. He called it the Concorde fallacy: “investing further in a project simply because one has invested in it heavily in the past, rather than because of potential future return on investment.”
Britain’s own cabinet papers complicate the nickname. When ministers weighed withdrawal again in 1969, the reasons on paper pointed forward: a “fifty-fifty chance” of “unacceptable costs” if France sued, and 16,000 workers building the plane. A lawsuit and lost jobs are future costs, and they belong in the decision. Whether the money already spent also pulled on ministers, the papers can’t show.
The clean version of the experiment came later, on paper.
A radar-blank plane
In 1985 Hal Arkes and Catherine Blumer, psychologists at Ohio University, asked students to play the president of an airline. You’ve put $10 million into building a plane that radar can’t detect. The project is 90 percent done when a rival starts selling a radar-blank plane that is faster and cheaper to run. Do you spend the last 10 percent? Forty-one of 48 said yes.
Another group read the same market news with nothing spent yet: an employee proposes putting the last $1 million of research money into the plane. Ten of 60 said yes.
They also ran a field test. For the Ohio University Theater’s 1982 to 1983 season, the first 60 people who came to buy season tickets were randomly charged the full $15, or got $2 or $7 off. In the first half of the season, full-price buyers used more of their tickets. In the second half, the difference between the groups was no longer statistically significant.
Arkes and Blumer’s explanation was a wish not to appear wasteful. Quitting turns the money already spent into waste you have to admit. Carrying on keeps open the hope that it wasn’t. Students who, in the made-up scenario, had sunk money into the plane also rated its chances of success higher, and having taken economics courses did not lessen the effect.
Weaker than the textbook says
Not every test finds a large effect. In a study published in 2007, a team of economists led by Daniel Friedman built a computer game with real payments, in which players decided whether to keep digging for treasure on an island or pay to move to another. They reported “a surprisingly small and erratic sunk cost effect.” They also argued that many famous examples can be explained without any error. What you’ve already put in can tell you something real about the road ahead, and quitting can carry costs of its own, like a treaty partner who sues.
The useful question is narrower: which of your reasons would still stand if the past spending had been free?
Eight months into a tool you could buy
Imagine you run a team eight months into building its own scheduling tool, two months past the original plan. A vendor now sells something that covers most of what you need. Every review starts with the eight months.
Andy Grove faced a bigger version in 1985, after Intel had spent almost a year trying to rescue its memory business. He asked Gordon Moore what a new chief executive would do if the board threw them both out. Moore said the new CEO would get Intel out of memory chips.
Put Grove’s question to your tool. A newcomer would see two columns: what it costs from today to finish against what the vendor charges, and what each gets you. The eight months appear in neither column, except as what they taught you about the hard parts. Sometimes the columns still say finish. Then you finish for a reason you can say out loud.
An hour into the wrong session
AI work has its own small Concordes. You’ve spent an hour steering a coding agent through a bug. You’ve corrected it twice, and the fix still isn’t right. Starting over feels like throwing the hour away.
Anthropic’s guide to Claude Code covers this moment. “If you’ve corrected Claude more than twice on the same issue in one session, the context is cluttered with failed approaches,” it says, and recommends clearing the session and starting fresh with a more specific prompt that incorporates what you learned. The same guide notes that sometimes you should let context build up, when you’re deep in one hard problem and the history is useful.
You can check which case you’re in. Before clearing, write down what the hour taught you: which file, which approach failed, which test proves the fix. Open a new session with only that note, and judge what it produces against the same requirements and the same test you started with. One fresh start going better is a clue, not proof that the old session was the problem.