Translations:Agile inception deck/6/en

• What keeps us up at night? in AI projects usually includes: model hallucinations, biases in the training data, inference cost at scale, and dependence on external providers (OpenAI, Anthropic, Google).
• The NOT list should explicitly include which data will not be used to train or fine-tune the model, especially customer data or sensitive information.
• Size it up requires defining quality metrics for the AI output (accuracy, coherence, rejection rate) in addition to the usual business metrics.