Start with the problem, not the technology. Understand what customers genuinely need, then choose the most reliable way to solve it; AI may be useful, but it is not the default. section 4.11
A practical test
For an AI assistant, ask whether you can describe the task clearly and validate its output easily. Translating hotel reviews may tolerate occasional hallucinations; a task where errors are unacceptable may require another approach. An existing open-source tool may also beat an AI-based reinvention. section 4.11
A real customer problem changes how a product should be judged:
- Check whether users care about accuracy, explainability, scalability, or small details before building. section 4.11 section 5.9
- Treat a prototype as evidence-gathering when its purpose is to test whether a real need exists, not as the final product. section 5.9
