AI-Assisted Teaching Tools and Teacher Burden in K-12 Informatics: A Burden-Augmentation Synthesis

A Vietnamese K-12 informatics teacher in academic year 2025-2026 operates under a structural workload that the 2018 General Education Programme (CT GDPT 2018) intensified — informatics is now a compulsory subject from lớp 3, lập trình content from lớp 6, with class sizes routinely exceeding forty students and most teachers carrying preparation, delivery, grading, parent-communication, and reporting load without dedicated teaching-assistant support. The same window has produced a class of AI-assisted teaching tools — generative-AI assistants (ChatGPT, Claude, Gemini, Vietnamese-language wrappers), AIaugmented tutoring systems (Khan Academy Khanmigo), AI code-feedback systems (GitHub Copilot, Codex-lineage tools), AI grading and short-answer assessment (the Wilson, Heffernanand-Heffernan, Burrows-et-al. lineages), and AI lesson-plan and parent-communication generators — marketed with the implicit promise that teacher burden will fall. This paper asks whether the promise is real for K-12 CS instruction, or whether the burden is reduced on the front-end authoring task and re-introduced as a hidden verification, correction, and quality-assurance load on the back-end. The synthesis integrates the AI-tutoring empirical literature (Bastani et al. 2024 GPT-4 vs GPT-4-Tutor field experiment; Kasneci et al. 2023 ChatGPT-in-education survey; Mollick 2024 Co-Intelligence; Khanmigo deployment reports), the AI-grading and code-feedback literature (Wilson et al. on automated scoring; Burrows et al. survey; Chen et al. 2021 Codex), the cognitive-offloading and verification-cost basis (Risko and Gilbert 2016; Kahneman–Klein 2009; Zhang et al. 2023 hallucination survey), the hybrid-intelligence and teacher-in-theloop literature (Holstein, McLaren and Aleven 2017–2019; Molenaar 2022), the teacher-workload literature (OECD TALIS 2018; Sellen 2024–2025), and the Vietnamese policy context (MOET 2018 CT GDPT; MOET AI-in-education guidance 2022–2024; Vietnamese educator commentary 2024–2026). The substantive contributions are (a) a seven-category K-12-informatics teachertask taxonomy with per-task burden-impact (reduce, shift, add); (b) a five-criterion burden-reduction rubric (time-saved-per-task, quality-preserved, instructor-judgement-required, teacher-skilluplift, hidden-cost); (c) a net-burden assessment matrix with delegation recommendation per task; (d) deployment guidance mapped onto the program's Python Pathway, From-Scratch-toPython, and Competitive Programming Pathway series at the teacher-assist layer; (e) honest treatment of the Vietnameselanguage AI-tool maturity gap. The conclusion is a structured it depends: AI reduces burden on a bounded subset of tasks (problem-set draft generation, first-pass grading on objective rubrics, parent-communication boilerplate, exam-preparation variant generation), shifts burden on a larger subset (lessonplan creation, code-review, individualised feedback) by trading