What actually fills a math instructor’s week
Whether AI will replace mathematical science teachers depends less on solving equations and more on the rest of the job. A chatbot can finish a calculus problem set in seconds. It cannot read a lecture hall, notice that half the room lost the thread at the substitution step, and start over with a different example. That live reading of a room is the core of the work, and it is why most of this job stays with people for now.
Two tasks show the split well. Preparing course material, practice problems and slides is work software now shares. Teaching the class itself, including answering the question a student did not know how to ask, is not. The same goes for advising: helping a student decide between a statistics track and a proof-heavy analysis sequence involves the student’s history, confidence and plans, not just a transcript.
There is also the assessment problem. When students can get answers instantly, instructors are rewriting how they test. Oral checks, in-class proofs and problems built around a student’s own prior work all take judgment to design and to grade fairly. That work has grown, not shrunk.
Tasks AI handles, shares, or leaves alone
Start with the tasks software can take on with little supervision. Drafting routine exercises, producing worked solutions, writing first-pass slide decks and scoring mechanical computation all sit here. The share of task time in that group is 6% of the job as we score it, and how that share is measured is set out in how coverage is scored.
Next come the tasks where AI assists a person who stays in charge. Giving feedback on written proofs and keeping course records and syllabus documents in order both fall in this group: the tool speeds up the first draft, the instructor checks the mathematics and the tone. That assisted share is 58% of task time.
Then there is the work that still needs a person. Lecturing and leading discussion, supervising student projects and theses, advising on course and career choices, and serving on department and curriculum committees all sit here. That group is 36% of task time, and it is the reason the headline figure for this job lands where it does. Our full headline score method explains how the three questions combine.
How strong is the evidence?
Thin, and we say so. Our evidence grade for this occupation is D on an A to D scale, and a D means no study in our set has tested an AI system against qualified postsecondary math instructors on this job’s real tasks. So we publish no parity number here. Benchmarks showing models solving competition problems are not the same test: solving a problem and teaching a room of first-year students to solve it are different skills.
What would settle it: course-level trials that compare AI-led sections with instructor-led sections on blind-graded exams and later course performance; measured learning gains from AI office hours versus human office hours; and graded, repeated tests of AI feedback on student proofs judged by experienced faculty. Until something like that exists, treat confident claims in either direction as opinion. You can see what the major assistants say about this kind of work in what the AIs say, and the wider approach in our scoring methodology.
When the picture could shift
Most likely between 2034 and 2045 (8 in 10 of our scenarios). What that window measures, and why it is a range rather than a date, is explained on the replacement-year method page.
Two things could pull it earlier. Budget pressure is one: course sections are expensive to staff, and the Bureau of Labor Statistics puts median pay for this occupation at $79,940 (BLS, 2025), so cheap tutoring software looks attractive to administrators. Enrollment shifts toward large online sections are the other, because scale is where automated grading and automated help desks save the most.
Two things hold it back. Accreditation and credit-hour rules expect a qualified instructor of record, and changing that is slow. And no physical automation is needed or available here: our robotics tier for this job is “None needed,” so nothing about the hands-on side of campus work changes the math. Trust is the quieter blocker. A department will not hand grading of a thesis defense to software it cannot audit.
Good to know: BLS reported 47,670 people employed in this occupation and projects employment growth of 1.7% from 2025 to 2035 (BLS, 2025), which is slow but not shrinking.
Staying needed in a math department
Lean into the parts of the job the task list puts firmly with people. Supervise undergraduate and graduate projects, where the value is in guiding a messy, original piece of work. Take advising seriously, because course-path decisions shape whether a student finishes. And own assessment design: build problems and oral checks that reward reasoning you can see, not answers that can be pasted in.
Two skills pay off. First, designing AI-resistant and AI-aware assessment, including how to let students use tools and still show their thinking. Second, fluency with the tools themselves, so you can show a class where a model’s proof quietly skips a step. Entry-level and adjunct hiring is the pressure point worth watching, and our entry-level hiring tracker follows that across occupations.
Nearby jobs face a similar mix. Compare the task splits for Computer Science Teachers, Postsecondary, Physics Teachers, Postsecondary and Mathematicians. For the wider picture, see the postsecondary teachers family and the education sector page, or put two jobs side by side with our compare tool.