Expanding Access to High-Quality Feedback with Pensive

## Overview Instructor: John DeNero, Electrical Engineering and Computer Science, University of California, Berkeley John teaches quantitative, problem-solving-focused courses with more than 1000 enrollments. As is common in technical disciplines, these courses require frequent assessment and timely feedback, placing substantial demands on instructional staff. Seeking ways to improve the **quality and timeliness of feedback** without increasing grading workload John began integrating Pensive into his courses. While efficiency gains were the initial appeal of Pensive, John ultimately found the greatest value in Pensive’s ability to improve access to academic support and provide students with more consistent opportunities to engage with course material outside of class. ## A Long History of AI Innovation John has a long history with AI, starting as a student. His dissertation focused on machine translation, one of the earliest, foundational examples of artificial intelligence. Following the completion of his doctoral work, John was hired as teaching faculty, with a focus on AI, at Berkeley. In 2015 he founded Lilt, an AI-driven translation company that includes an opportunity for human intervention in its output, similar to the Pensive grading process. On the academic side, John’s contributions to undergraduate education have been recognized with multiple campus‑wide teaching awards, including Berkeley’s Distinguished Teaching Award and the Jim and Donna Gray Award for Excellence in Undergraduate Computer Science Education. In addition to his teaching, John has published and presented research on scalable instructional systems and AI‑supported learning environments, including presenting work at the ACM SIGCSE Technical Symposium and won the [Google Academic Research Award](https://research.google/programs-and-events/google-academic-research-awards/google-academic-research-award-program-recipients/?filtertab=2024) for his work on AI tutoring in 2024. In 2026, John will be presenting both on instructors’ perception of LLM models for providing feedback and on his experience using AI Tutors for programming courses. Taken together, John’s experience as an instructor, educational leader, and researcher situates him as a credible authority on the opportunities, affordances and potential pitfalls of using AI to support learning at scale. ## Instructional Challenges Prior to adopting Pensive, John encountered several challenges typical of large courses: - Students required more frequent and detailed feedback than could realistically be provided through manual grading alone. - Many students were unable or unwilling to attend office hours, or desired feedback during the times they were working on assignments. - Grading open-ended or computational assignments demanded substantial time, often delaying feedback. - Delayed or minimal feedback reduced early interventional opportunities for both John and his students to correct misunderstandings earlier. > “Students do not always recognize when they need help, or they may hesitate to ask. That can limit how effective traditional support structures are.” > > — John DeNero John sought an approach that would allow students to receive guidance and feedback continuously, rather than only during scheduled instructional time. ## Implementation John incorporated Pensive’s AI-assisted grading and AI tutoring tools into his existing course workflow. With Pensive, he was able to: - Streamline grading while still providing instructor oversight and judgment - Ensure that **all students received substantive, individualized feedback** - Provide students with on-demand support aligned with course expectations and learning objectives Rather than replacing established teaching practices, Pensive functioned as a complementary layer of support, extending instructional reach beyond the classroom. ## Expanding Access to Academic Support A central benefit of Pensive for John was its capacity to make academic support more accessible. The AI tutor enabled students to: - Seek clarification at any time, including outside normal instructional hours - Work through problems incrementally using guided prompts - Receive immediate feedback when encountering conceptual or procedural difficulties This proved particularly beneficial for students who were less inclined to participate during class or attend office hours. > Lowering the barrier to asking questions can make a meaningful difference. Students are more likely to engage when support feels readily available. > > — John DeNero John observed more sustained engagement throughout the term and fewer instances of students falling behind without intervention. ## Enhancing Feedback Quality John also emphasized improvements in the depth and consistency of feedback. - AI-generated comments were specific to student work and aligned with course objectives - Feedback focused on reasoning and process, not solely correctness - Faster turnaround allowed students to apply feedback to subsequent assignments While John continued to review and adjust feedback as needed, the AI-generated responses provided a reliable baseline that ensured all students received meaningful guidance. > Even when time is limited, students benefit from clear explanations. This approach makes that feasible at scale. > > — John DeNero ## Effects on Teaching and Learning By reducing the time required for grading and routine student support, Pensive enabled John to reallocate effort toward higher-impact instructional activities: - Identifying common misconceptions and addressing them during class - Designing assignments that encouraged deeper engagement - Supporting a learning environment in which students felt more confident attempting challenging problems John also noted improvements in student preparation and reduced anxiety around exams, which he attributes to clearer expectations and more timely feedback. ## Broader Academic impact John situates his experience with Pensive within a broader and growing body of research on the role of AI-supported systems in higher education, particularly in computer science. In 2025, John published a [proceedings paper, presented at the ACM SIGCSE Technical Symposium](https://dl.acm.org/doi/epdf/10.1145/3641555.3705244), that highlights how AI-supported tutoring environments can meaningfully enhance student collaboration, satisfaction, and access to support in under-resourced courses. This research emphasizes several findings from John’s classroom experience: - **Scalability of instructional support**: results demonstrate that AI-supported systems can extend the reach of instructors and teaching assistants, enabling effective small-group tutoring even in large-enrollment courses. - **Positive student perceptions**: Survey data indicate strong student preference for AI-augmented tutoring interfaces, particularly when AI support complements human instruction John sees these findings reflected in his own courses, where AI support helped students engage more consistently with course material and reduced barriers to participation. Importantly, the research underscores that the effectiveness of such systems depends on careful integration into pedagogical practice, with instructors maintaining oversight over assessment and learning outcomes. ## Comparison to Alternatives Before adopting Pensive, John relied on conventional grading, office hours, and peer support mechanisms. While effective in some respects, these approaches were difficult to scale without increasing workload. > Most systems are designed to record grades. What was missing was a way to support learning continuously. > > — John DeNero The integration of grading assistance and instructional support distinguished Pensive from other tools he had used. ## Conclusion For John, Pensive has served as a means of extending instructional support while preserving academic rigor. By improving access to feedback and guidance, the platform has helped create a more responsive and inclusive learning environment. > Effective teaching depends on sustained engagement. Tools like Pensive make it possible to support students more consistently, even in resource-constrained settings. > > — John DeNero Through its combination of structured feedback and on-demand support, Pensive has enabled John to better align assessment practices with student learning needs across his courses.