Do you need to know what you want before you act?

About 22,500 people who couldn’t make up their minds went to a website and flipped a coin. Heads: make the change. Quit the job, end the relationship. Tails: stay.

Steven Levitt’s study (Review of Economic Studies, 2021) found that heads made people noticeably more likely to change, and that six months later the people the coin told to change were happier than the people it told to stay. His conclusion: people “may be excessively cautious when facing life-changing choices.” He also calls the results “merely suggestive”, and he’s right to.

The question underneath

A Big Think article by Simone Stolzoff (18 September 2026) offers five ways to become “high agency”, the skill Sam Altman reportedly named as the most important for the AI age. The article cites no studies, and its first and last tips pull against each other. Tip one: work out what drives you. Tip five: act in spite of uncertainty, illustrated with Sylvia Plath’s fig tree, where Esther starves because she can’t choose a fig.

So which comes first, clarity or action?

The case for acting

Our deliberation isn’t neutral. It leans towards staying put.

• We overestimate refusals. In Flynn and Lake’s studies (JPSP, 2008), Columbia students guessed they’d need to ask 7.2 strangers to get one to walk them to the gym. It took 2.3. Charity volunteers expected to approach about 210 people each and needed about 120.
• We overestimate the audience. In Gilovich and colleagues’ spotlight-effect study (2000), students in an embarrassing T-shirt guessed about half a room noticed. About a quarter did. (Figures from secondary summaries.)
• For small decisions, it barely mattered which way people went. In Levitt’s data the coin moved behaviour on minor questions far more than on big ones, and made no detectable difference to later happiness. On those, the agonising was the only cost.

The case for clarifying

• The coin study selected people who’d already done the thinking. Levitt recruited people “having a difficult time making a life decision”; on average they put their own odds of changing at about 42%. They were stuck, not starting. The sample was self-selected, “young, male, and highly educated”, follow-up fell from about 14,000 to 8,000, and happiness was self-reported. For big decisions the happiness difference wasn’t clear at two months and was imprecise at six.
• Choice overload mostly hits people without preferences. A meta-analysis of 50 experiments (Scheibehenne, Greifeneder and Todd, 2010) found an average effect of about zero. The authors read the exceptions as people who didn’t know what they wanted. Know what you want, and more options help.

There’s a twist in the fig tree, though. In the novel, Esther doesn’t lack preferences. She wants every fig, and choosing one means losing the rest. That’s not a clarity problem. More thinking wouldn’t fix it.

Two kinds of door

Jeff Bezos’s 2015 letter to Amazon shareholders splits decisions into one-way doors, which should be made “methodically, carefully, slowly”, and two-way doors, which “can and should be made quickly”. His warning is that big organisations drift into using the slow process for everything, and get “slowness, unthoughtful risk aversion, failure to experiment sufficiently.”

People do the same. Flynn and Lake’s students were treating a two-block walk like a one-way door.

But the same door isn’t the same for everyone. In Bowles, Babcock and Lai’s experiments (2007), roughly a thousand people rated invented job candidates who did or didn’t ask to negotiate. Asking cost both men and women, but in the first experiment the penalty for women was more than twice as large, and male evaluators penalised women more than men. Lab ratings aren’t a real hiring panel, but the point stands: a two-way door for one person can be close to one-way for another.

And “the age of AI”?

The nearest study is a CHI 2025 survey of 319 knowledge workers. The more confident they were in the AI, the less critical thinking they reported; the more confident in themselves, the more. It’s self-reported, and the authors say it doesn’t establish causation. Nothing tests whether “high agency” protects anyone from anything.

---

Check which kind of door you’re at, and whether it’s the same door for you as for the person giving the advice. If it’s one-way, take your time. If it’s two-way, and most are, walk through. Think until you’re stuck, then stop: the evidence suggests that at the fence, people lean too far towards staying. You don’t learn what you want from a fig tree. You learn it from eating a fig.

---

Sources and reporting notes: mostly American samples; as far as I found, nothing was run in New Zealand. The Altman remark is from the article, with a secondhand check. The Gilovich figures are from secondary summaries. Levitt’s happiness scale is not given here because accounts of it differ.