This study is to investigate the effects of AI-driven feedback generated by My Access on EFL college students' academic writing performance and student writing assessment literacy (SWAL). Adopting a mixed-methods research design with a quasi-experimental quantitative strand and a complementary qualitative strand, this study addressed four research questions concerning between-group differences in writing performance, within-group writing improvement, between-group differences in SWAL, and students' perceptions of AI-driven feedback. Participants were fifty-three first-year vocational students assigned to either an experimental group (AI-driven feedback) or a control group (teacher feedback). They also completed different writing tasks and revised drafts across feedback cycles. Quantitative analyses examined pre- and post-test writing performance across six traits and SWAL across four dimensions. Results indicated there was no significant difference between the groups after testing, whereas the experimental group showed significant in-group improvement. MANCOVA revealed a statistically significant group effect in favor of the experimental group for SWAL, where stronger behavioral engagement was found to be more pronounced, and marginal gains were observed regarding assessment literacy. Qualitative data further indicated greater self-awareness, revision, and metacognitive engagement, but with caution regarding the utility of feedback. Overall, the findings indicate AI-driven feedback improves engagement and assessment literacy in EFL writing.