Generative artificial intelligence (GenAI) can support programming education but may also encourage cognitive offloading when learners request complete solutions instead of developing problem-solving competence. This paper aims to frame natural language programming (NLPg) as a metacognitive scaffold prioritising problem framing, decomposition and verification over answer production. The contribution is primarily pedagogical, positioning NLPg as a competence-oriented practice for human-centred AI-supported learning in digital and cyber humanities contexts.
The study reports an exploratory pilot in a master’s digital humanities course. Thirteen student prompt logs (439 prompts) were analysed using cue dictionaries and three trace-based indicators capturing metacognitive regulation, delegation versus co-reasoning and decomposition with explicit validation. Trace data were complemented by rubric-based outcomes from oral project defences.
Prompt logs reveal a characteristic “build-and-check” pattern: planning and monitoring cues are frequent, while reflection is the least stable NLPg component. Rubric evidence shows weaker evaluation in the baseline cohort and higher scores for framing, decomposition and validation in the GenAI-enabled cohort despite increased task complexity.
The study is exploratory and context-specific; findings rely on a small cohort and do not support causal claims, calling instead for longitudinal and cross-institutional validation.
Assessment in AI-rich learning environments should prioritise traceable reasoning through specifications, decomposition artefacts, validation evidence and brief reflection notes.
Competence-oriented NLPg can strengthen learner agency and reflective engagement in AI-mediated learning and knowledge environments.
The study links NLPg routines, competence frameworks and trace-based evidence to support competence-oriented GenAI pedagogy aligned with DigComp 3.0 and emerging cyber humanities perspectives.
