World model labs won't say what they're building, and their supplier is 'in the dark'. The design question that tests for the same gap.
TechCrunch's explanation is that world models are versatile as an idea, with possible directions running from robotics to interactive video to self-driving, and that while fundraising is easy there is, in the reporter's words, "no particular pressure to focus on one." Staying quiet also delays competition, which the piece calls "a dark forest scenario." That is the reporter's reading, not something the companies said. But de Vigan's complaint stands apart from the strategy. A capability with no named user is hard to build for, and everyone downstream ends up making something generic. A lab with funding can postpone that choice. A product manager in an interview cannot, and one common design prompt exists to find out whether the candidate notices. It hands over a feature name with no user attached and waits to see who picks one.
Why they ask it
The prompt, as candidates have reported it, is How would you design 'Hot Topics' on Instagram? It contains very little. It doesn't say whether Hot Topics is a tab, a feed module, a search surface or a notification. It doesn't say who it is for or what problem it solves. The name describes a capability, that the platform can detect what people are talking about, and nothing else.
That is deliberate. The interviewer isn't grading the feature. They are watching the order in which the candidate makes decisions. They want to see a user and a problem chosen before a screen gets drawn, and they want to hear why that user was picked over another. For senior candidates it is also a test of something managers do constantly, which is taking a vague mandate from leadership and returning it as something a team could start on Monday.
The trap
The trap resembles TechCrunch's description of World Labs' Marble, a platform that "seems more designed to demonstrate capabilities." Weak answers tour everything Hot Topics could be. They describe a trending tab for casual viewers, alerts for creators, a news module and a brand-safety view for advertisers. Each gets a minute of sketching and none gets chosen. The tour sounds thorough, but it leaves the interviewer where de Vigan is. They know something is being built and can't say what.
The opposite failure is just as common. The candidate goes straight to the interface and places a carousel at the top of Explore, with no user behind it. A third version borrows a trending list from another platform and ports it across, without asking what is different about a product built on photos, video and followed accounts. All three skip the same step, because nobody has yet decided who the feature is for.
Applying STAR-T
A design prompt isn't a behavioural question, but the same spine holds it up, and it keeps the answer from sprawling.
Situation. Set the context in a few sentences, stating assumptions aloud instead of asking the interviewer to supply them. A candidate might say that the product is organized around followed accounts and a recommendation system's guess at individual taste, and that what it lacks is a shared sense of what everyone is discussing right now. Then name the plausible users, such as casual viewers, creators and people who come for news, and say plainly that they want different things.
Task. Choose a user. A strong answer commits to one. For example, the candidate designs for mid-sized creators who want to join a conversation while it is still live. The reasons are that creators produce what everyone else consumes and that they usually discover a topic after it has peaked. The goal is to shorten the distance between a topic emerging and a creator posting into it. The candidate should also say what is out of scope, and say so early.
Action. Design for that person only. Decide where it surfaces, which would be inside the creation flow and not the consumption feed. Decide what qualifies as a topic, using velocity and not raw volume so the list isn't the same celebrities every day. Decide how it explains itself, so a creator can see why a topic is being suggested. Then set the guardrails, meaning which topics are never promoted (tragedies, harassment pile-ons, unverified claims) and who makes that call.
Result. There is no real outcome to report, so define one. A primary metric would be the share of creators shown a topic who post into it and beat their usual reach. A counter-metric would be reports and hides on topic-tagged posts, because a trending surface that rewards outrage will hit its primary number while damaging the product. The candidate should say what the first experiment would need to show and what would stop the rollout.
Trade-off. Name the cost of the choice. Designing for creators means casual viewers get nothing at launch. It also carries a homogenization risk, in which everyone posts the same thing in the same hour. The candidate should say why that cost is acceptable and what evidence would reopen the decision.
The follow-up that breaks weak answers
The first follow-up is simple: Why that user and not the others? The harder version changes the constraint. Leadership now wants this for news consumers instead, so what changes? A candidate who toured the options has nothing to defend and nothing to revise. A candidate who chose can walk through it. The surface moves, the topic threshold tightens, the guardrails get heavier and the metric changes.
Then the question often turns behavioural: Tell me about a time you were handed something this vague for real. This is where a prepared story matters. It would cover a mandate that arrived as little more than a feature name, the users that were considered, the one that was chosen, what was declined, and what happened afterward. A generic example might begin with a request that arrived as a single line asking for a dashboard. The story would then cover the week spent finding out that only the support leads would open it every morning, and the features cut for everyone else.
AMI Labs can wait until it is ready to talk. A candidate has the length of the interview to make the same decision.
Score your answer against the director’s bar
Q: How would you design 'Hot Topics' on Instagram?
Bank the story of a vague mandate that became a specific product, and rehearse it against the follow-ups. Try it free →