Assess the reasoning behind an AI-polished product
If AI can produce working software, what evidence should establish a student’s engineering competence? An ongoing postgraduate course offers a practical design response. By week eleven, 87 students in 17 teams had produced extensive repository records within an assessment system combining products, process, verification, explanation, reflection and individual assurance. The report describes how these channels can be collected, cross-checked and escalated for human review. It also exposes the difficulty: code, documentation and reflections can all be generated together, so several artifacts may not be independent evidence of understanding. This is an implementation account with useful design lessons, rather than proof that a particular assessment model is valid or that student learning improved.
Why it matters for leaders
Pair an AI-permitted product with a short individual explanation or changed problem. Give teachers clear criteria for resolving conflicting evidence, and estimate moderation time before requiring large portfolios of logs and reflections across subjects.
The evidence
The 13-week course uses six evidence channels. The report discusses collection, interpretation and workload but does not yet analyze student-level achievement or establish comparative effects.
Who was studied
University of Sydney software-intensive engineering course; indirect evidence for school assessment.
What the evidence cannot tell us
One ongoing postgraduate course, no comparison group and no validated measurement model. More traces may increase compliance work without strengthening the inference about competence.
Read the original sources
Summary and leadership analysis from the Week ending 2026-10-30 brief. The edition’s selection notes explain source access and evidence limitations.