Unit 15: AI-Assisted Development

An AI coding interface can produce an answer faster than you can review it. That speed is useful only if your review process is stronger than your urge to accept plausible-looking code.

Separate the model from the authority

A coding assistant may combine a language model with project context, retrieval, editor integration, or other tools. The interface can therefore do more than simple text completion. None of those capabilities turn its output into proof.

For this course, every AI-produced suggestion begins in the same state:

CANDIDATE

It becomes project code only after you compare it with the requirement, inspect the exact changes, execute appropriate tests/checks, and decide to accept, revise, or reject it.

You will practice bounded requests, diff-first review, generated-test skepticism, debugging assistance, explanation prompts, privacy boundaries, and rejection. If no AI tool is available, supplied candidate patches and responses exercise the same review skill.

The artifact is the review, not the prompt

Keep one candidate change and document what you requested, what changed, what evidence you ran, what you rejected or revised, and why the final decision matched the specification.

Using AI is optional. Owning the code is not.