Direct answer
The best AI development partner should clarify the workflow, data requirements, technical risk, MVP boundary, ownership model, and commercial goal before writing production code.
The questions good partners ask
A good partner asks what user decision improves, what evidence exists, how quality will be measured, what data is missing, and which parts of the workflow should stay human-reviewed.
The warning signs
Be careful with vendors who start with model names, demos, or large fixed scopes before they understand the workflow and adoption risk.