How to Give Feedback on AI-Assisted Programming Work
Give feedback tied to an observable programming decision or check, then make the student’s next action clear without inferring hidden traits from AI use.
TraceYield Insights
6 min read · Updated October 6, 2026
Start from the work you can point to
Feedback on AI-assisted programming is most useful when it names a specific part of the submitted work or a recorded check, explains why it matters for the task, and gives the student a feasible next action. A functioning result may still contain an untested assumption; a failed attempt may provide a useful point for feedback. Neither observation, by itself, explains the student’s motivation or general ability.
This is a teaching practice, not a way to score AI use. The ACM Task Force report includes educator-contributed approaches to programming instruction and assessment; the example below is a constructed TraceYield application, not a research finding about feedback effectiveness.
Constructed example: an ambiguous date parser
Weak feedback: “The code works.” This describes a result too broadly for the stated input rule. “You relied too much on AI” is also unsupported: the code alone does not establish how much help the student used or what they understood.
Evidence-linked feedback: “The assignment requires ISO-formatted dates. Your parser accepts `03/04/2026`, which can mean different dates under different conventions, while your current test covers only `2026-04-03`. Add a test for the ambiguous input and decide, from the stated contract, whether the parser should reject it or accept a specified format.”
The observable point is the parser’s behavior and the scope of the current test. The comment does not infer why the student chose that parser. Its next action is a bounded test and a decision tied to the assignment requirement.
Make the next action proportionate
A next action might be to add one boundary test, reproduce a failure with a smaller input, verify an API claim against documentation, or explain why a design fits the stated constraint. It should match the learning goal and the risk of the task; feedback need not demand exhaustive testing for every small exercise.
When feedback concerns an explanation or a trajectory record, be precise about what prompted it. For how to select process evidence, see Process Evidence. The lecturer-role guide discusses how feedback can connect to teaching decisions.