Graduate teaching assistants

Three parts of your work where AI is already doing real lifting, and what stays yours.

• Chalk1876–1945

  Chalk, printed textbook, and oral examination (German seminar model)

  The teaching assistant of the research-university era worked with the same tools as the professor: a blackboard, a textbook, and direct conversation. The seminar method imported from German universities required the fellow to guide discussion, challenge student arguments, and demonstrate mastery of primary texts. There was no paper-grading machinery, no standardized rubric, and no administrative layer between the TA and the student. Assessment happened largely through oral examination and written essays returned with handwritten marginal notes. The craft of teaching was transmitted by observation and imitation of senior faculty, with little formal preparation.

• Mimeograph1945–1975

  Mimeograph and overhead projector (mass university era)

  The GI Bill enrollment surge forced universities to process instruction at scale. Mimeograph machines, and later Ditto machines and spirit duplicators, let TAs reproduce problem sets, reading guides, and quizzes for sections of many students. The overhead projector (widely adopted in classrooms through the 1950s-1960s) let a single TA deliver visual material to a room. These were genuinely labor-saving relative to hand-copying, but grading still happened entirely by hand, and a TA responsible for many students in three sections could spend several hours per week marking papers. The mimeograph era defined the TA's experience for a generation: low pay, high paper volume, and isolation from the scholarly life promised in the fellowship brochure.

  Effect on the work

  The GI Bill doubled US higher education enrollment from 1940 to 1960; universities absorbed this demand primarily by enlarging lecture sections and adding TA-led discussion sections rather than hiring additional tenure-track faculty. Teaching assistants became structurally load-bearing in ways they had not been before World War II.

• Photocopier1975–1998

  Photocopier, telephone office hours, and word processor

  The Xerox-era photocopier, commonplace in university departments by the mid-1970s, freed TAs from mimeograph chemicals and cut distribution time significantly. Word processors (first departmental, then personal computers through the 1980s) changed how TAs prepared written feedback: longer, more structured marginal comments replaced cramped handwritten annotations. Telephone office hours, while not a formal tool, became standard: a TA's posted phone number became the first real-time help channel available to students outside classroom hours. None of these tools reduced the core labor of reading and marking written work, which remained the dominant time demand of the TA role.

• Learning Management Systems1998–2012

  Learning Management Systems (WebCT 1996, Blackboard 1997, Moodle 2002)

  WebCT (acquired by Blackboard in 2005) and Moodle gave TAs their first digital infrastructure for course delivery: online grade books, threaded discussion boards, digital assignment dropboxes, and email-broadcast lists. For the first time, a TA could return marked assignments without handing back physical paper, hold asynchronous "office hours" via discussion board, and track individual student engagement over time. The grade book alone reduced end-of-semester arithmetic by hours. The LMS also introduced a new source of TA labor: administering the course shell, answering routine student emails generated by 24/7 platform access, and troubleshooting technology problems that faculty often delegated downward.

• AI-assisted grading infrastructure2012–2023

  AI-assisted grading infrastructure (Gradescope 2014, Turnitin 1998 with AI layer added 2023)

  Gradescope, founded in 2014 at UC Berkeley and acquired by Turnitin in 2018, was the first tool to directly accelerate the grading workflow for large courses: its answer-grouping algorithm clusters similar student responses so a TA can apply one rubric decision to many answers simultaneously rather than reading each individually. For STEM courses with short-answer or multiple-choice components, Gradescope reduced grading time in documented deployments. Turnitin's plagiarism detection tool, which had been in use since 1998, added an AI-writing-detection layer in 2023, introducing a new category of TA judgment work: interpreting AI-similarity scores and deciding whether to escalate to faculty.

  Effect on the work

  At UC Berkeley and similar large research universities, adoption of Gradescope across STEM courses reduced total TA grading hours for structured assignments, while increasing the time TAs spent on rubric design and cluster-review judgment tasks.

• Generative AI tutoring2023–present

  Generative AI tutoring and course-content tools (Khanmigo 2023, MagicSchool AI 2023, NotebookLM 2023)

  Khanmigo (Khan Academy, March 2023), MagicSchool AI (2023), and Google NotebookLM (2023) represent the first generative-AI tools specifically designed for the education workflow. Khanmigo provides 24/7 Socratic tutoring that can handle the routine conceptual questions that previously consumed TA office-hour time, freeing TAs to focus on complex reasoning, mentorship, and the course feedback loop. MagicSchool AI generates first-draft problem sets, rubrics, and differentiated explanations from course material, reducing prep time for new TAs. NotebookLM synthesizes large reading piles into cited summaries, accelerating the research-support tasks many TAs perform for supervising faculty. The practical effect, still unfolding as of 2026, is a rebalancing: routine factual Q&A and first-draft materials move toward AI, while the irreplaceable human contribution of the TA concentrates on facilitation of live discussion, mentorship, nuanced feedback on original student thinking, and academic integrity judgment.