Why explaining beats answering
Ask whether AI will replace teaching assistants, postsecondary, and the honest answer depends on which half of the job you mean. Grading short answers, writing practice problems and drafting review sheets are language tasks, and current models handle a first pass well. Running a Friday discussion section with twenty tired sophomores is a different kind of work.
Two duties show the split clearly. A teaching assistant marking a problem set is reading for a known answer, and software can score that quickly. A teaching assistant holding office hours is reading a student: where the confusion started, whether they studied, whether they are too embarrassed to say they are lost. That second reading drives what you say next, and nothing on a screen has to sit with a crying freshman in week ten.
Lab and studio sections push the same way. Supervising equipment, catching an unsafe step before it happens and signing off on a student’s technique all take a person in the room. Across education jobs, the pattern is similar: the paperwork and the practice material erode first, while the live teaching hour holds.
What AI handles, what it assists, what people keep
The tasks AI can take on its own cluster in marking and material prep. Scoring objective items, flagging likely plagiarism, generating extra drill questions and turning lecture notes into a study guide all fit. Our coverage score, the share of task time AI can handle today, is measured here and prints as 29 out of 100.
The assist group is bigger in practice. Drafting feedback comments that a teaching assistant then edits, summarizing which questions a class got wrong, answering repeat logistics questions in the course forum, and prepping a worked example before a review session. Share of task time where AI helps rather than replaces: 23%.
What stays with people is the teaching itself: leading sections and recitations, running office hours, supervising lab work, judging borderline work and handling academic integrity conversations. People hold this share of task time: 71%.
Good to know: the grading part of the job is often the part a department is quickest to automate, because it is measurable and already runs through the learning platform.
How strong the evidence is
Here is the limit worth knowing. Our evidence grade for this job is D, and a D grade means no study has tested AI against postsecondary teaching assistants on their own work. So we publish no parity number for this occupation. Parity is explained in how we judge quality, where 50 means a typical qualified professional.
What would settle it is specific: a graded comparison of AI feedback against assistant feedback on the same student papers, marked blind by faculty; a term-length trial of AI tutoring against staffed office hours on the same course outcomes; and audits of how often automated scoring agrees with a trained human on partial-credit answers. Until something like that is published, the number would be a guess. Our full approach is set out in the scoring method.
Labor data gives useful context in the meantime. BLS counted about 164,090 postsecondary teaching assistants in the United States with median pay of $42,910 (BLS, 2025), and projects employment to grow 2.7% from 2025 to 2035. That is slow growth, not contraction.
When the balance could shift
Most likely between 2035 and 2052 (8 in 10 of our scenarios). What that window measures is explained on the replacement year page.
Two things could pull it earlier. First, grading and tutoring features are being built straight into the course platforms universities already pay for, so adoption needs no new purchase decision. Second, software licenses are cheap next to staffing extra sections, and budget pressure makes that comparison tempting when enrollment in a large lecture course climbs.
Two things hold it back. The physical share of the work, lab supervision and in-room instruction, needs dexterous general-purpose robots, which are not close to doing a chemistry bench safely. And the role is partly a funding mechanism: graduate assistantships pay tuition and stipends, so cutting them changes how a department recruits doctoral students, not just how it grades. Hiring at the entry edge is still worth watching, which is what our entry-level tracker follows.
How to stay needed in this role
Lean into the parts of the job that take presence and judgment. Run sections where students talk rather than listen. Make office hours the place students bring confusion they can’t type into a prompt. Own lab or studio supervision, including safety and technique sign-off, because that is the hardest piece to hand over.
Two skills travel well from here. One is assessment design: writing problems and rubrics that test reasoning rather than recall, and spotting work a model produced. The other is using AI tools openly as a first-draft grader and example generator, then showing faculty where the tool got it wrong. That makes you the person who supervises the software instead of the person it duplicates.
If you are weighing a longer path, close jobs are worth comparing: Tutors, Teaching Assistants, Special Education and Education Teachers, Postsecondary. You can put any two of them side by side on our compare page, or browse neighboring roles in the same job family. Our headline figure for this job is 71 out of 100 (higher is safer), and what that score counts is published in full.