Strategy question

Anthropic wants PMs to uncover non-obvious use cases for frontier models and feed those insights back into research. How would you structure opportunity discovery, compare adjacent versus novel use cases, and choose the first experiments to run?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Ability to structure opportunity discovery for a frontier model and connect it back to research priorities.

Set up structured channels to surface use cases: customer interviews, support and sales signals, and hands-on internal experimentation with new capabilities.
Separate adjacent use cases, which extend something the model already does well, from novel ones that need new capability or evaluation.
Score candidate use cases on user value, technical feasibility given current model limits, and what they would teach research about model behavior.
Choose a small first batch of experiments that spans both adjacent and novel cases, so learnings are not skewed toward the easy wins.
Define what evidence from each experiment feeds back to research, such as failure modes or unexpected capability edges.
Set a short cycle time so findings reach research while the underlying model version is still current.

Keeps adjacent and novel use cases in separate tracks since they need different validation approaches and timelines.
Prioritizes experiments that generate research-relevant signal, not just ones likely to succeed commercially.
Builds a lightweight, repeatable intake process instead of one-off exploration, so discovery keeps running after this project ends.
Names a concrete first experiment and the specific question it would answer for research.

Only chasing obvious adjacent use cases and missing genuinely novel ones.
Treating discovery as a one-time project instead of an ongoing feedback loop with research.

How would you prioritize if a novel use case has high value but low current feasibility?
What signal would convince research to prioritize a capability gap you found?
How often would you refresh this process as models improve?