In April 2025, Duolingo’s chief executive Luis von Ahn sent a memo declaring the company AI-first. Employees would be evaluated, in part, on how much they used AI.
A year later, he killed the rule.
Speaking on a podcast in April 2026, von Ahn said the company had backtracked because staff had started asking whether Duolingo just wanted them to use AI for AI’s sake. The policy had produced compliance. He also pointed to his own product, where he said AI-written code can be difficult to debug and is not consistently reliable for writing Duolingo’s stories.
Here is what makes it interesting. Von Ahn did not change his mind about AI. He still believes in it, still wants to hire people who use it well. What he changed his mind about was a specific instruction he had given publicly, and the evidence that led to the change came from his own staff and his own product.
Now compare that with a different reversal from the same period.
In May 2026, both Sam Altman and Dario Amodei walked back their predictions of an AI jobs apocalypse. Amodei had warned a year earlier that AI could eliminate half of entry-level white-collar roles and push unemployment to 10 or 20 per cent. Altman said he had been “pretty wrong” about the economic impact.
On the evidence, they had good reason. The Yale Budget Lab had found no significant change in occupational mix or unemployment duration in high-exposure jobs. But both companies were also approaching public listings, and a great many people read the reversal as positioning rather than learning.
Two changes of mind, weeks apart, both defensible. One read as a leader responding to evidence. The other read as a leader responding to bankers.
This is important: the question facing you is almost never whether to change your mind. It is whether the people watching can tell which kind of change it was, and what they conclude about how decisions get made here. Get that wrong and you pay twice, because you take the cost of the reversal and lose the credit for the learning.
By the end of this article you will have a four-part test for the difference, and the words to use when you have to stand up and say you were wrong.
Learning, or flinching?
Investors have a working distinction for this. They ask whether a leader is learning or flinching.
Larry Fink is the clearest illustration, because he did both.
When Fink changed BlackRock’s stance on crypto, the markets broadly approved. When BlackRock rolled back its environmental and diversity commitments, and Fink commented that the pendulum had swung too far, two Dutch pension funds divested a combined €17bn in direct response.
The same chief executive, equally willing to reverse himself, and the verdicts came out opposite.
This is happening at scale. Around one in eight large companies has weakened its diversity policies, and roughly one in five has partly or wholly abandoned its net-zero commitments. Many of those reversals will be read as flinching whether they deserve it or not, because they arrived without an explanation.
Remember: your organisation is not judging whether you were right the first time. It is judging what your reversal tells them about how decisions get made here. That is a much harsher standard, and it is the one you are actually being marked against.
The four-part test
Ok, here is the practical part. Four questions separate learning from flinching, and you can apply them to your own reversal before anyone else does.
1) What specifically changed? Learning names the new information. Von Ahn could point to what his staff were telling him and to code he could not reliably debug. Flinching cannot name anything, because what changed was the pressure rather than the evidence, so it reaches for vague language about the environment or the pendulum.
2) Does the new position cost you something? Nobody believes a reversal that happens to make life easier. Von Ahn abandoned a policy he had announced publicly, which cost him. If your new position is more comfortable, more profitable and better received all at once, expect to be read as flinching even if you are not.
3) Do you own the old position? Learning says I decided this and I was wrong. Flinching says the memo was misinterpreted, or the market shifted, or we are evolving our approach. The organisation can tell the difference immediately, and what your reactions teach them outlasts the decision itself.
4) Is the direction still the same? Von Ahn reversed a tactic while holding his strategic view. That reads as learning because the destination did not move. A reversal that also changes where you are going is a different and much bigger announcement, and pretending otherwise is how leaders lose the room.
What to actually say
When you have to reverse a decision you announced, say something close to this:
In March I decided we would do X, and I was wrong. What I have learned since is Y, specifically. That changes the decision but not the direction: we are still building towards Z. Here is what we are doing instead, and here is what I will be watching to know whether this one is right.
You name the decision as yours, you name what changed, you protect the direction, and you commit to a standard for the new position. Every one of those is missing from a reversal that reads as flinching.
Three things to leave out. Do not say the original decision was misunderstood, unless it was. Do not blame conditions unless conditions are what changed. And do not quietly stop doing the thing and hope nobody raises it, which costs you more than either, because it teaches everyone that decisions here simply evaporate.
What this is not
You might be thinking, “So I should be willing to reverse anything.” It is the opposite.
A leader who changes position every time the evidence wobbles has not learned anything; they have just stopped deciding. Your organisation needs to know that most of what you commit to will hold, or nobody can plan. A genuine reversal carries weight precisely because it is rare.
There is also a version of this that is simply cowardice, described in the language of learning. Reverse a difficult commitment the moment it becomes unpopular, call it new evidence, and you get the worst of both: you lose the position and you lose the credibility, because people can usually see what actually changed.
So make reversals rare, so that they mean something when they come. Evidence them, so that somebody can check your reasoning. And own them, because otherwise everyone will guess whether you noticed.
What next?
Think of a decision you announced in the last year that you now suspect was wrong.
Not one you have already quietly abandoned. One that is still officially in place, still being executed by people who believe you meant it, and which you have been avoiding revisiting because reopening it would be uncomfortable.
Run the four questions on it, in the same way you would run any hard conversation with your team. If you can name what changed, if reversing would cost you something, if you are prepared to own the original call, and if your direction holds, then you are not flinching and you should say so this week.
The alternative is to let it fade. That protects your record and costs you something far more valuable: the ability to be believed the next time you commit to anything.
Say it plainly, and say it early.