Distinguish explanation, completion and multi-step workspace work.
1. Interaction modes
Code completion suggests local text, chat can explain or propose changes, and agent workflows can inspect and modify multiple files. The level of action should match the task.
2. Context quality
The agent needs relevant source, project conventions and a concrete problem statement. More context is not always better if it includes stale or unrelated instructions.
3. Engineering ownership
Generated code can compile and still be wrong. You remain responsible for behavior, architecture and compatibility. Review the resulting evidence rather than trusting confident prose.
Worked scenario
An agent fixes a visual issue by removing a state update that tests depend on. The screenshot improves while behavior regresses.
Apply it
Compare an explanation request and an implementation request for the same bug, specifying the expected output of each.
Check your understanding
You can choose an interaction mode based on the work required. Explain the decision and show evidence from your implementation or design. If you cannot demonstrate it yet, revisit the relevant section before continuing.