Designing AI Behavior: Why Prompts Aren't Enough
How we shape, evaluate, and improve generative experiences beyond the prompt.
TicketsHow we shape, evaluate, and improve generative experiences beyond the prompt.
TicketsPrompting is often treated as the whole job: write the right instruction, and the AI does the rest. In practice, a single prompt rarely produces a generative experience people can rely on. What actually shapes behavior is the system around the prompt — the examples, guardrails, evaluation criteria and feedback loops that decide whether an AI response is useful, safe, and on-brand.
Drawing on her work building AI experiences at eBay, Barbara walks through how to move from a clever prompt to a generative feature that holds up in production: how to define what “good” output actually looks like, how to evaluate AI behavior systematically rather than by gut feeling, and how to iterate once the feature is live and real user feedback starts coming in.
You'll leave with a clearer picture of where prompt engineering ends and product and content design work begins — and a few concrete methods for evaluating and improving AI behavior you can apply on your own product.
For content designers, product designers, PMs and anyone shaping AI-powered features who wants to go beyond “just write a better prompt.”