Natural-language app builders can make software creation feel like a direct move from prompt to product. In practice, a teacher’s professional intent passes through several consequential transformations: compilation into a specification, code generation, automated checks, packaging, repair and approval.
With Wei Ting Liow, Sumbul Khan and Lay Kee Ang, I examined those transformations in a new preprint, “Beyond Prompt-to-App: Accountable Translation in Teacher-Facing Agentic Authoring.”
A bounded trace study
The study analyses six eligible build attempts across three accounts. A separate corpus of 37 workshop units from 23 display names provides context without person-level linkage. This is deliberately bounded evidence: it follows what happened in retained traces rather than making claims about all teachers, systems or deployment settings.
When technical passage is not faithful translation
The traces show why passing a technical threshold is not the same as faithfully representing an educator’s brief. Compiled specifications added governance requirements, while later representations sometimes normalised case-specific learning relationships. Two drafts met a stored package-and-security threshold despite analyzer reservations and unresolved correspondence to their briefs. Four attempts in one account produced no usable payload, and repair messages did not translate internal terms into domain-legible revisions.
Accountable translation
We propose accountable translation as a way to keep consequential changes attributable, inspectable, properly scoped in validation and contestable by the professional whose intent is being transformed. The point is not simply to expose more technical logs. It is to help educators see what changed, what was checked, what remains uncertain and where professional judgement can intervene.
What the paper does not claim
This is a bounded trace study and a preprint. It does not establish classroom learning outcomes, general system reliability or peer-review acceptance. Those limits are part of the argument: evidence and uncertainty should remain visible when an agentic authoring system transforms professional intent.
I welcome conversations with educators, HCI researchers, end-user-development scholars and educational-AI teams interested in teacher agency, traceability and repairability.
Read the preprint: arXiv · DOI
Publication note: this is a preprint and has not been peer reviewed.
Leave a Reply