A ChatGPT inventor built an AI model that outputs no language at all. The question that tests whether you know your audience.
Set aside whether Jev holds up. The diagnosis is the interesting part. Nothing in it says the intelligence was poor. It says the intelligence arrived in a form its receiver could not act on. The receiver was a program that needed a number, and it was being handed paragraphs. The fix TechCrunch describes is not a cleverer sender. It is a model in which "users define the outputs in advance" — the receiver decides the shape of the message before it is sent. That is a behaviour, not a technology, and it is the behaviour behind a question managers are asked constantly: Tell me about a time you had to tailor your communication for different audiences.
Why they ask it
A manager sits at a junction. The same fact — a slipped date, a failed experiment, a reorganization — has to reach engineers, finance partners, executives, and sometimes customers, and each of them has to do something different with it. The interviewer is checking whether the candidate starts from the receiver: what does this audience have to decide, and what form of information lets them decide it?
Armin Ronacher, the CTO of Earendil, gave TechCrunch a neat description of what a usable message looks like from the receiving end. A coin-toss probability gets disregarded. "But if it’s 95%, sure, then I can do something with it." That last clause is the test. An audience that cannot do something with what it was told has not been communicated with, however accurate the content was.
The question also checks something simpler: whether the candidate notices audiences at all. Plenty of senior people have a single register and deliver it everywhere, at the same depth, in the same order.
The trap
The common weak answer is about style. The candidate explains that they removed the jargon for the business side, used an analogy, and cut the slide count. That is translation, not tailoring. It treats the non-technical audience as a less capable version of the technical one, receiving the same message at lower resolution. A finance lead does not need a simpler account of the architecture. They need a different message entirely, built around the decision that is theirs to make.
The second trap is the opposite failure: a story in which each audience heard a different truth. Tailoring changes form, order, and emphasis. It never changes the facts. Interviewers listen for this, because a manager who tells the executive team a rosier story than the engineers is describing a liability, not a skill.
The third is an answer with no evidence the message landed. Sending is not the accomplishment.
Applying STAR-T
Situation. Name the single fact and the audiences, and define each audience by what it had to decide rather than by its job title. A billing platform migration was going to miss its date because a vendor dependency slipped. The engineering team had to re-sequence its work, the finance lead had to revise a forecast, and the executive sponsor had to choose between cutting scope and moving the date.
Task. State the responsibility as an outcome. The job was not to inform everyone of the delay. It was to get the right decision out of each group in the same week, without any of them learning the news from another.
Action. This is where the answer is won, and the strongest move comes before any message is written: asking each audience what it needs. Then show the differences concretely. The engineers received a written document with the dependency graph and a proposed new sequence, circulated early enough to be challenged. The finance lead received the dates on which the forecast would change, a range rather than a point, and a plain statement of how confident the team was in each end of it, with no architecture at all. The sponsor received a single page: the options, the cost of each, and a recommendation. All of it pointed back to one source document, so that anyone comparing notes would find the same facts underneath.
Result. Report what each audience did, because behaviour is the only proof of comprehension. The sponsor chose within the week. The forecast was revised before the close rather than after it. The finance lead could restate the risk in their own terms, unprompted, in a later meeting.
Trade-off. Tailoring has costs, and naming them is what separates a considered answer from a tidy one. It took several times the effort of a single announcement. It also required omission: the sponsor’s page left out a lower-probability risk that the engineers were tracking closely. Say how that call was made, and where the omitted detail lived for anyone who went looking. Ronacher’s caveat about Jev applies here too. "At the end of the day, it delegates the hallucination problem a little bit to the user," he told TechCrunch. A compressed message delegates judgment to its receiver in the same way, and the sender owns that choice.
The follow-up that breaks weak answers
The follow-up is usually some version of this: What did you leave out for each audience, and how did you decide it was safe to leave out?
A style-only answer collapses here, because nothing was left out; the same content was simply reworded. A candidate who truly tailored the message can list the omissions immediately, since choosing them was most of the work. They can also say what would have caused them to put a detail back — the threshold at which the executive page would have needed the engineers’ risk on it.
A second follow-up often arrives close behind: How did you know it landed? The weak answer reports that nobody had questions. The strong answer points to a decision made, a number changed, or a person repeating the message accurately to someone else. Bryo AI’s CTO told TechCrunch that what interested him most about Jev was that "it is the only one that hands back a real probability which makes it ideal for automating workflows!!" The enthusiasm is for an output shaped to be acted upon. An interviewer is looking for the same property in a manager.
Score your answer against the director’s bar
Q: Tell me about a time you had to tailor your communication for different audiences.
Bank one story in which the same facts went to different audiences in different forms, and rehearse it until the omissions are as clear as the message. Try it free →