When should products use AI?

In a product-focused interview, you are asked to discuss the adoption of AI within software products.

Describe how you would decide whether a particular feature should use AI or a conventional deterministic implementation. Your response should cover:

1. What makes a problem a good fit for AI.
2. What types of problems should generally not use AI.
3. What an AI-native application is, and how it differs from a standard product that merely adds an AI feature.
4. The practical trade-offs around accuracy, latency, cost, explainability, safety, and user trust.

Support your reasoning with concrete examples.

Overview:

This question assesses product-focused judgment on AI adoption by gauging proficiency in machine learning product strategy, weighing trade-offs like accuracy, latency, cost, explainability, safety, and user trust, and telling AI-native architectures apart from standard add-on AI features.