There was a time when students would spend long hours hunched over thick textbooks, moving between library shelves, trying to piece together answers to problems that refused to give in easily. Then came the “just Google it” era, which made information faster to access but still required some effort to search, filter, and understand.At the University of California, Berkeley, a striking academic pattern has emerged in spring 2026 computer science courses, raising difficult questions about artificial intelligence, student preparedness, and the future of technical education.Failure rates in key computing classes have surged well beyond historical norms, marking a sharp departure from departmental expectations and triggering concern among faculty about how students are learning and what they are actually retaining.According to data reported by Berkeleytime and cited by The Daily Californian, 35.3% of students in CS 10 (The Beauty and Joy of Computing) received failing grades in spring 2026. In CS 61A (The Structure and Interpretation of Computer Programs), the failure rate stood at 10.6%.These figures are especially significant given that in spring 2024 and spring 2025, failure rates in both courses remained below 10%. The EECS department’s grading guidelines typically anticipate around 7% D and F grades in lower-division courses, making the latest results a notable outlier.Even more concerning, both courses recorded average grades equivalent to a C+, roughly a 2.3 GPA, below the department’s expected range of 2.8 to 3.3.

AI, academic integrity, and a shifting learning curve

A central concern raised by instructors is the growing influence of generative artificial intelligence tools in student workflows.UC Berkeley teaching professor Dan Garcia, who taught both CS 10 and CS 61A, told The Daily Californian that a “primary driver” behind the unusually high failure rates was what he described as a “vast increase in academic dishonesty” linked to large language models such as ChatGPT, Claude and Google Gemini.According to Garcia, nearly 30 students in CS 10 were “caught cheating on take-home exams” in spring 2026, with additional cases referred to the Center for Student Conduct.However, Garcia also pointed to a more subtle issue than outright misconduct: overreliance on AI tools that may allow students to complete assignments without fully internalising the underlying concepts.As quoted by The Daily Californian, Garcia said students are “leaning a little too hard on LLMs to do their work for them, and then at exam time just really aren’t ready.”The result, he suggested, is a widening gap between coursework completion and actual competency, one that becomes visible when students are no longer able to rely on AI assistance during in-person assessments.

Fixed standards and visible gaps

Unlike many university courses that rely on grading curves, Garcia’s classes used fixed grading thresholds, where letter grades are determined by clearly defined performance standards.Speaking to The Daily Californian, Garcia said he is “a strong, strong opponent” of systems that cap the number of high grades and prefers transparent benchmarks that allow students to reach an A without institutional limits.He argued that curved grading can obscure deeper instructional problems by distributing grades regardless of absolute performance levels, whereas threshold-based systems make learning gaps more visible.

The Mathematics readiness question

Beyond AI-related concerns, faculty members have also highlighted a persistent issue: uneven mathematical preparedness among students entering advanced computing courses.Associate teaching professor Gireeja Ranade told The Daily Californian that her EECS 127 course (Optimization Models in Engineering) had become “differently challenging” in spring 2026, as students struggled with foundational concepts in linear algebra, vector calculus, and mathematical proofs.The course recorded a 16.8% failure rate, significantly above the EECS department’s typical benchmark of around 5% D and F grades for upper-division classes.Ranade noted that some students lacked prerequisite mathematical fluency despite progressing into advanced coursework. In discussions with students during office hours, she learned that at least one had previously taken a linear algebra course that permitted open-internet and open-AI use for assignments and examinations, according to her remarks reported by The Daily Californian.

Teaching under pressure

Structural constraints have further complicated instruction. Ranade told The Daily Californian that staffing shortages forced the removal of a major project component from EECS 127, a segment that had previously offered guided, hands-on learning with teaching assistant support.According to EECS Department Chair Jelani Nelson, as cited in a post on X referenced by The Daily Californian, the university has had to reduce both undergraduate computer science enrolment and the number of undergraduate teaching assistants due to rising costs associated with TA wages.

Declining engagement in learning spaces

Perhaps as telling as the grade distributions is what faculty are observing outside the classroom. Ranade said office hours, once “overflowing” with students, have seen “very low engagement” in spring 2026 despite repeated encouragement for attendance, according to The Daily Californian.Garcia reported a similar trend, describing office hours that were at times entirely empty, an experience he called surprising after years of steady student participation.“I used to have full office hours, and for the first time, I was having nobody come to my office hours,” Garcia said, as quoted by The Daily Californian. “It was just so surprising to sit in my office alone.”The shift raises broader questions about how students are seeking academic support and whether AI tools are quietly replacing traditional forms of learning interaction.

Rethinking the classroom in the AI era

Both professors are now reconsidering how their courses should evolve in response to these changes.Garcia told The Daily Californian that he plans to explicitly address the spring 2026 outcomes with future students and explore ways to identify those who need additional foundational support.Ranade, meanwhile, argued that the solution is not to simplify instruction but to deepen it. She emphasized that students must be prepared for a more competitive and complex world, where analytical and critical thinking skills remain essential even as AI tools become ubiquitous.“We really need to make sure that we are preparing our students to be solid, contributing citizens and leaders,” Ranade said, as quoted by The Daily Californian. “We need to—and we want to—teach them how to… take on new challenges.” Garcia echoed a similar sentiment, reflecting on what he called the essential difficulty of true learning.“Confusion is the sweat of learning,” he said, as quoted by The Daily Californian. “A lot of students, I think, are not putting in the sweat.”

A larger warning for higher education

What is unfolding at Berkeley may be an early signal of a broader transformation in higher education.Artificial intelligence is no longer an external tool used sparingly in academic settings, it is increasingly embedded in how students complete assignments, approach problem-solving, and even conceptualise learning itself.The challenge now facing universities is not simply how to regulate AI, but how to preserve intellectual struggle in an environment where answers are always instantly available.At Berkeley, the data suggests that when that struggle is reduced or outsourced, performance gaps become sharply visible. Whether that represents a temporary adjustment period or a longer-term structural shift in education remains an open question.What is clear, however, is that the traditional relationship between assignment completion and learning outcomes is under pressure, and institutions are only beginning to understand the consequences.