Given the popularization of artificial intelligence in education, this study explores this technology's effects on emergent bilinguals' writing and revision. Large Language Models have become prevalent as supports for teachers and students in revising. Yet, research has not examined the intricacies of using these tools with emergent bilinguals. This study uses a LLM (ChatGPT4o) to assess barriers and affordances, and their role in the cultural reproduction of language instruction in schools. Through revising five published bilingual children's books, this study employs different prompting styles as a heuristic for traditional revision processes. Employing a critical content analysis of the LLM's revisions and taking up translanguaging and decolonial lenses, several themes are explored: standardizing translanguaging, standardizing structure, and homogenizing culture. Further, the paper discusses how the LLM can reify deficit-based framings, calling for the examination of technology's capabilities and barriers, and how teachers and students can (dis)engage for liberation.