World model labs won't tell their own data supplier what they're building. The underperformer question that tests for the same silence.
On September 18, 2026, TechCrunch published its reporter's account of moderating a panel on world models at the All In conference. The piece describes a field in which the best-known companies, Yann LeCun's AMI Labs and Fei-Fei Li's World Labs, have plenty of buzz and funding and very little to say about what they are building. Michael Rabbat, a co-founder of AMI Labs and its VP of World Models, told the reporter over email that the company is "in a research and building phase" and is "not talking publicly about any product plans or timeline." TechCrunch is fair about this: AMI is less than a year old, and the article argues the quiet has a logic, since a clear path to market would invite well-funded rivals. But the secrecy, it reports, reaches the companies' own suppliers. Alex de Vigan, CEO of the data supplier Physicl, said he knows his data has been useful and is still in the dark about what it is useful for. "I wish they would tell us more," he told TechCrunch. "We could build more useful data if we knew what they were working on."
Set the strategy aside and look at de Vigan's position. He is delivering, his work is being accepted, and by his own account he is working below what he could do. The gap has nothing to do with ability. Nobody has told him what the work is for, so he cannot aim it. For the labs that may be a cost worth paying. Inside a team it is rarely a decision at all, and it produces the same result: a capable person turning out adequate, slightly wrong work, while the picture of what good looks like stays in the manager's head. A good deal of what gets labelled underperformance starts there, which is why a story about secretive AI labs leads to the interview question: "Tell me about a time you had to manage an underperformer and how it impacted the team."
Why they ask it
Every manager has this story, so the interviewer is not checking whether it exists. They are checking the diagnosis. A performance problem can come from skill, from will, from circumstances, or from a missing expectation, and each calls for a different response. The candidate who treats them all the same way will run the same playbook on every struggling report they inherit.
The second clause of the question carries as much weight as the first. An underperformer is never a private matter between a manager and a report. Peers absorb the slipped work, watch how long the manager takes to act, and watch how the person is treated. The interviewer wants to know whether the candidate saw that happening while it was happening.
The trap
The common weak answer is a process recital. The report missed deadlines, the manager gave feedback, wrote a plan, held check-ins, and the person either improved or left. Everything in it is defensible and none of it is informative. It skips the only interesting moment, which is how the manager worked out why the performance was poor, and it treats the team as scenery.
The subtler failure is the answer in which the manager has no part in the cause. If the report was unclear on priorities, whose job was it to make them clear? Interviewers notice when a candidate describes a person who did not know what was expected and never asks who left them in the dark. The strong answer names the manager's own contribution early and without drama, then shows what changed.
Applying STAR-T
Situation. Keep it short and make the gap observable. A senior engineer joined from a team with different norms and spent a quarter shipping work that was technically sound and consistently beside the point: thorough solutions to problems the roadmap had already moved past. Reviews dragged, and peers began quietly rerouting the urgent tickets around them.
Task. State the real task, which is diagnosis before correction. The manager needed to find out whether this was a capability problem or an information problem, and needed to do it before the team's workaround hardened into a verdict on a colleague.
Action. This is where the answer is won. The manager asked the engineer to describe what they thought the team was for and what a good quarter looked like, and the description was a fair account of the previous team. Nobody had replaced it. So the manager wrote the expectations down, walked through recent work against them, and named which pieces had missed and why. They set a short window with specific outcomes and said plainly what would happen if the window closed without them. Separately, they told the team that the work distribution was being fixed, without discussing the individual, and moved the rerouted tickets back.
Result. Give the outcome for the person and for the team, in that order. The engineer's next projects landed on the roadmap rather than next to it, and review time dropped back to the team's norm. The peers who had been carrying extra work stopped carrying it, which the manager confirmed by asking them rather than assuming.
Trade-off. Name the cost. The diagnosis took weeks the team paid for, and the manager should say they would now set expectations in the first days of onboarding rather than discovering their absence a quarter later.
The follow-up that breaks weak answers
The follow-up is almost always a version of this: at what point would the manager have decided it was not going to work, and what would they have done then? A candidate whose story rests on process has no answer, because the process was the plan. A candidate who did the diagnosis can say exactly what evidence would have shifted the conclusion from unclear expectations to a genuine capability gap, how long they would have waited for it, and how they would have handled an exit so that the rest of the team saw fairness rather than either delay or haste. The interviewer is listening for a threshold that existed at the time, not a principle assembled in the room.
Score your answer against the director’s bar
Q: Tell me about a time you had to manage an underperformer and how it impacted the team.
Bank the underperformer story with the diagnosis, the expectation conversation, and the team's cost in it, then rehearse it against the follow-up. Try it free →