From Code Completion to Correct Programs: Human-Machine Differences and Hierarchical Debugging

Abstract

Ai-Assisted Programming poses a recurring problem of coordinating performance with evidence quality, resource limits, and transfer across settings. The review develops an evidence-centered account of treating code generation as an iterative reasoning and execution process rather than one-shot text prediction. The source set brings together 2 focal papers with 13 independently retrieved publications reviewed against traceable publication metadata. The analysis is organized around programmer behavior, error localization, execution feedback, repair hierarchy, and evaluation leakage. Instead of assuming that reported outcomes as directly interchangeable, the review compares study questions, technical premises, and validation scope. Across the literature, the central lesson is that advances in AI-assisted programming become credible when representation, objective, and evaluation protocol are evaluated together and when uncertainty about distribution shift is reported explicitly. The resulting account aligns method selection to implementation risk, making transfer failures visible, and proposes a research agenda centered on auditable baselines, controlled perturbations, and replicable records.