Most university teaching assistant appointments are built around a simple number: roughly twenty hours a week for a half-time position, averaged across the semester. On paper, that sounds manageable. In practice, TAs in writing-heavy courses will tell you the average is meaningless, because grading weeks eat far more than their share while quiet weeks barely register. A midterm essay stack for a 120-student lecture course can consume an entire week's allotted hours by itself, leaving nothing for office hours, section prep, or the research the TA is actually supposed to be advancing toward their own degree.

University workload policies acknowledge this unevenness explicitly, generally advising TAs to bank lighter weeks in anticipation of heavier grading periods. That guidance is reasonable in theory. In a semester with three separate essay assignments and a final paper, there often aren't enough light weeks to bank against.

The mismatch creates a quiet but persistent tension between instructors of record, who need essays graded promptly and consistently across sections, and TAs, who are contractually limited in how many hours they can spend doing it. When that tension isn't managed well, something usually gives: turnaround time slips, grading gets rushed and inconsistent across sections, or TAs simply work unpaid hours to keep up, which departments have every reason to want to avoid.

The consistency problem multiplies with more graders

A single professor grading their own class has one internal standard, even if it drifts slightly over time. A large lecture course with four or five TAs grading different discussion sections has four or five standards running in parallel, and research on inter-rater reliability in essay scoring shows that without deliberate calibration, those standards diverge meaningfully. A student in one section can end up graded more harshly than an identical essay submitted in another, purely as a function of which TA happened to be assigned their work.

• Build a shared rubric with concrete, example-anchored descriptors before the semester starts
• Run a brief norming session where all TAs grade the same sample essays and compare results
• Rotate a small percentage of essays across TAs for spot-check comparison mid-semester
• Give TAs a first-pass tool that applies the shared rubric consistently, so human review time goes toward judgment calls rather than mechanical scoring
• Track average grades by section to catch drift before it reaches students as a fairness complaint

What actually reduces the crunch

Spreading assignment due dates across sections rather than stacking them on the same day is one of the simplest fixes, and one of the most underused. When every section's essay is due Friday, every TA hits a grading wall the same weekend. Staggering deadlines by even a few days smooths the workload without changing how much writing students actually produce.

Rubric-based AI grading tools have started to find a real foothold here, specifically because the problem is volume and consistency rather than pedagogical judgment. A tool that generates a rubric-aligned first-pass score and draft comment gives a TA something to react to and adjust rather than a blank page to fill from scratch on every one of forty essays. The TA still makes the final call on every grade; the mechanical first pass is what gets faster.

A workload conversation departments should have earlier

Faculty supervisors are generally receptive when TAs raise workload concerns directly, but many TAs are reluctant to bring it up until they're already behind. Departments that build a mid-semester check-in on grading workload into the standard TA orientation catch problems while there's still time to redistribute sections or adjust deadlines, rather than after a TA has quietly worked well past their contracted hours for weeks.

None of this requires an overhaul of how graduate teaching works. It requires treating grading load as a real, measurable part of the job, the same way section size and office hours already are, and building tools and schedules around that reality instead of around an average that rarely matches any actual week.