AI Product Thinking
The judgement half of the interview — what to build with a model, and what not to.
5 topics
Designing for Nondeterminism
The same input can produce different output and sometimes a wrong one, so the product has to be built around being occasionally wrong rather than around being right.
core20 minAI UX Patterns
The handful of interaction shapes that keep recurring — inline suggestion, command palette, chat, agent-with-approval — and why chat is usually the laziest of them.
core20 minTrust, Transparency & Citations
Trust has to be calibrated rather than maximised — users should believe the system exactly as much as it deserves, which means showing your work and admitting limits.
core15 minWhen Not to Use an LLM
Slower, more expensive, and occasionally wrong is a bad trade against a regex, a query, or a lookup table — and recognising that is a senior signal.
core15 minShipping & Iterating on AI Features
The delivery loop for a feature whose quality you cannot fully predict — narrow scope, real traffic, tight feedback, and the willingness to pull it.
core20 min