AI-assisted development

AI-assisted development uses coding assistants (Copilot, Cursor, ChatGPT, Claude, etc.) to draft, refactor, explain, or test code. Interviewers care whether you treat assistants as **accelerators under engineering discipline**—not as a substitute for ownership, review, or security judgment.

| Practice | Why it matters | |----------|----------------| | Own the diff | You merge what you understand and can maintain | | Review like a teammate | Check correctness, edge cases, style, and dependencies | | Validate with tests | Generated code often compiles but fails on real inputs | | Prompt hygiene | Clear context, constraints, and examples reduce garbage output | | Privacy boundaries | Do not paste secrets, PII, or unreleasable proprietary code into untrusted tools |

					Workflow: small scoped prompt → read full diff → run targeted tests →
refactor for clarity → document non-obvious choices in PR
				

Common pitfalls: blind merge because "it compiles"; skipping tests on boilerplate; leaking credentials into prompts; accepting architecture you cannot explain; using assistants on compliance-sensitive paths without policy.

On interviews: describe how your team uses AI without lowering the bar for review, testing, or data handling. Mention when you would **not** use an assistant (incident response, subtle concurrency, regulated data).

The trade-off is speed of first draft versus time spent verifying and simplifying generated complexity.

Checklist:

• Policy for what context may leave the repo.
• Mandatory human review on production paths.
• Tests or focused manual evidence before merge.
• Ability to explain every merged line.