The problem with your AI output started before you typed a word
Every team I work with has the same complaint about AI: the output is generic. They try better prompts, more context, a different model. Sometimes it helps. But the problem usually started earlier, at the moment they decided what to ask the AI, without first being clear on what a useful answer would look like.
This workshop works through that moment in two passes, using a shared fictional feature as the working material throughout. In the first, you will take a vague hope about AI and turn it into a specific, testable question for that feature. What would success look like? What would make the output unsafe? Could you run this evaluation by Friday? Most teams skip this step entirely, which is why their evaluations end inconclusively.
In the second pass, you will take that now-specific question and use it to construct a better prompt: narrower, more contextual, asking for brainstorming material rather than finished work. You will see how clarity before the prompt changes what comes back.
The final exercise gives you the actual AI output that prompt produced, printed on cards. No laptops, no wifi. Your job is to evaluate those scenarios against the question you developed in the first pass.
You will leave with a template for building both: the question that makes evaluation possible, and the prompt that makes AI output worth having.