## Question How do you use AI in your daily software-engineering work, and where should it not be trusted? Give one example in which AI accelerated work and one in which verification found an important flaw. Explain your standards for data handling, code ownership, testing, review, and evaluating whether AI actually improves engineering outcomes. ### Constraints & Assumptions - Do not send secrets, personal data, proprietary code, or restricted documents to an unapproved system. - The engineer remains accountable for generated code and claims. - “The model said tests pass” is not evidence; independent tools must produce the result. - Separate productivity from correctness and maintainability. ### Clarifying Questions to Ask - Are you asking about coding, debugging, design, documentation, or all four? - Which AI tools and data policies are approved in this environment? - Should I discuss team-wide controls as well as personal workflow? ```hint Define an acceptance boundary Name what evidence must exist before generated work is merged: understanding, tests, static checks, security review, benchmarks, or a human domain decision. ``` ### What a Strong Answer Covers - Specific high-value uses and explicit prohibited or high-risk uses. - Verification independent of the model's own assertions. - Small diffs, provenance, reviewability, and understanding of final code. - Attention to privacy, licensing, security, and prompt-injection risks. - Outcome measures such as cycle time plus defects, rework, or review burden. ### Follow-up Questions 1. What do you do when generated code passes visible tests but you do not understand it? 2. How would you evaluate an AI coding pilot for a team? 3. Which tasks require a human decision even if the model is accurate? 4. How do you prevent AI-generated tests from repeating the implementation's mistake?