<![CDATA[A preliminary observation of 31 eleventh-grade students in the Network and Application Engineering (SIJA) program at a vocational school in Cimahi found that all students (100%) had come to rely on generative AI such as ChatGPT whenever they encountered coding difficulties, with 86.7% reporting an inability to complete programming tasks independently without AI assistance. This AI dependency risks hindering the development of students' critical reasoning. This research aims to design, implement, and test an automated feedback system named AutoFeed to support the Problem-Based Learning (PBL) model in mitigating AI dependency while enhancing students' critical thinking skills in Informatics subjects at vocational high schools (SMK). The research methodology applies the Waterfall development model, encompassing requirements analysis, system design, software implementation, and testing phases. The application was built using a technology architecture consisting of React on the frontend, FastAPI (Python) on the backend, and MySQL as the database, integrating the Google Gemini 2.5 Flash model as the processing engine for hybrid (corrective and reflective) feedback through Socratic questioning techniques. The results of expert judgment based on LORI standards showed an average usability score of 4.60, workspace-PBL feature alignment scored 4.25, while the validation of AI pedagogical response accuracy scored 4.00. Functional testing further confirmed that all system features operated as intended but still require refinement. Thus, the AutoFeed system is confirmed feasible as a digital learning assistant (co-thinker), designed to minimize delayed feedback constraints and support students' critical reasoning, with its actual learning impact recommended for further empirical validation.]]>
