Short answer: no, and the official projections say the opposite. The US Bureau of Labor Statistics expects postsecondary teachers to grow 7 percent between 2025 and 2035, adding about 98,200 positions to a 2025 base of 1,378,200, with 2025 median pay of $85,330. That is double the 3.5 percent projected for total US employment over the same decade. What is genuinely under threat is not the professor but a particular model of teaching: the lecture that transmits information, the assessment that measures recall, and the tutorial that answers questions a chatbot answers instantly. Those were always the most replaceable parts of the job, and they are going first.
The Part That Is Actually Being Automated
Universities do many things, and only some of them are information transfer.
Information transfer is the piece under real pressure. A student wanting the second law of thermodynamics explained can now get it at two in the morning, pitched at their level, with follow-up questions answered. That used to be a reason to attend a lecture. It is no longer a good one.
What is not under pressure: certifying that someone actually knows something. Supervised practice with expert feedback. The social structure that gets people to finish hard things. Research supervision. The judgement about what is worth teaching at all. Those are the load-bearing functions of a university, and none is an information problem.
The uncomfortable implication is that the lecture was never the valuable part. It was the delivery mechanism for the valuable part, and that mechanism is now being unbundled.
Where the Pressure Actually Lands
The employment effects will not be evenly distributed, and the projection for the occupation as a whole hides a lot of variation.
| Segment | Pressure level | Reason |
|---|---|---|
| Large introductory lecture courses | High | Content is standardised, so delivery alone no longer differentiates |
| Fully online asynchronous programmes | High | The differentiator was convenience, which AI tutoring also offers |
| Adjunct teaching of survey courses | High | Lowest institutional attachment, most substitutable content |
| Laboratory and studio teaching | Low | Supervised physical practice with real-time correction |
| Clinical and professional programmes | Low | Accreditation, licensure and supervised placement requirements |
| Graduate supervision | Low | Apprenticeship, judgement and accountability for a person’s work |
| Small-group seminars | Low | The value is the discussion, not the content |
Read that table as a description of what students pay for. Wherever the answer is “access to information”, pressure is high. Wherever the answer is “access to a person who will correct me and certify that I can do this”, pressure is low.
BLS also published AI exposure categories alongside these projections, sorting occupations into Low, Moderate, High and Very high relative exposure using five external datasets. It states directly that exposure “does not imply job loss, productivity gains, automation probability, or wage effects.” Teaching involves substantial language work, so it registers on exposure measures. It is still projected to grow well above average. Both facts are real, and the gap between them is exactly the gap between tasks and jobs.
The unbundling has happened before
Universities have survived this pattern several times. The printing press made lectures technically unnecessary, since a book could transmit the same content more reliably. Recorded lectures and open courseware were supposed to end the residential university in the 2010s. In both cases the institution absorbed the technology and kept its position, because what it sells is not primarily content.
What is different this time is speed and interactivity. A textbook cannot answer a follow-up question or adapt to a confused student. A tutoring system can. It does so at a marginal cost near zero, with a patience no overworked lecturer can match at 2am in week nine.
So the honest version is neither “nothing changes” nor “it is over”. It is that the weakest teaching is now clearly worse than the free alternative, and the gap between weak and strong teaching has become visible to students in a way it was not before. That is uncomfortable for the sector and mostly good for students.
What to Know Before You Draw Conclusions
Assessment is the immediate crisis, not employment. The take-home essay as a measure of individual understanding is in serious trouble, and no institution has fully solved it. This is a problem about what a degree certifies, and it is far more urgent than any question about job numbers.
The funding model matters more than the technology. Enrolment demographics, state funding and international student policy will do more to postsecondary employment over ten years than any model release. Academics worrying about AI are often worrying about the wrong variable.
Adjunct and tenured positions face completely different risks. A tenured position is contractually protected. An adjunct teaching three sections of an introductory course is exposed to any change in how that course is delivered. The occupation-level projection averages the two together and hides the gap.
Research supervision is durable for a structural reason. A doctoral student needs someone accountable for their training who can judge whether work is original and defensible. That accountability cannot be delegated to software.
Institutional inertia cuts both ways. Universities change slowly, which protects jobs in the short run and makes the eventual adjustment sharper. A department that spends three years debating policy while its students quietly restructure how they study is not preserving standards, it is deferring a decision.
Students still want the credential. Demand for postsecondary education is demand for signalling and certification as much as for learning, and no current technology issues an accredited degree.
What Good Teaching Looks Like Now
The academics adapting well have made the same three moves, and none of them involves banning the technology.
- Assessment moved toward process and performance. Oral examinations. In-class writing. Practical demonstrations. Iterative drafts with documented feedback. Defence of your own work. All expensive in staff time, and all more reliable than what they replaced.
- Contact time moved toward the hard part. If explanation is available on demand, contact time is better spent on practice, correction and discussion than on delivering content. This is the flipped classroom argument, finally with a real forcing function behind it.
- The technology became a subject as well as a tool. Every discipline now has a version of the question “what does this tool do to our field, and what does it get wrong?” Teaching that explicitly is more useful than pretending it does not exist.
A worked example of the second move. A quantitative methods course previously ran as two hours of lecture and one hour of tutorial. The lecturer now records short explanations, uses the two hours for supervised problem work where students attempt questions with staff circulating, and uses the tutorial for students to present and defend an analysis they ran themselves. Content coverage is identical. What changed is that the lecturer now sees, every week, exactly which students cannot do the thing, which was previously invisible until the examination.
Institutions that responded to generative AI mainly with detection software are having a worse time than those that redesigned assessment, because detection is unreliable and adversarial while redesign addresses the actual problem.
A Decision Framework for Academic Careers
- Tenured or tenure-track. Your position is secure and your teaching model is not. The productive response is redesigning assessment in your own courses now, which is also the fastest route to institutional influence, because most departments have nobody who has actually done it.
- Adjunct or contingent, teaching survey courses. This is the most exposed position in the sector. Move toward courses with a practical, clinical or laboratory component, or toward the assessment and course-design work institutions urgently need.
- Doctoral student deciding whether to continue. The academic job market’s problems long predate AI and are mostly about funding and supply. Evaluate on those grounds. Build skills that transfer, because most doctoral graduates work outside academia regardless.
- Administrator or programme lead. Your near-term problem is assessment integrity and your medium-term problem is explaining what your degree certifies. Both are solvable and neither is solved by procurement.
The useful question for any academic: if a student could get every explanation you give from a machine tonight, what would still make your course worth attending? A confident answer is a secure position. No answer is the thing to work on.
Building Fluency Rather Than Taking a Position
The academics with the most influence in this debate right now are the ones who have used these systems seriously enough to describe their failure modes precisely. That is a very different posture from either enthusiasm or refusal, and it is the one that carries weight in a curriculum committee.
Getting there means understanding how these systems generate output, why they fabricate confidently, what they do well in your specific discipline, and how to design a task they cannot complete for a student. That is a short, structured course of study rather than a research programme, and doing it deliberately is much faster than absorbing it from institutional guidance documents. Pairing it with a certificate gives you something concrete when a department asks who should lead its policy. If you want a structured route in, explore Coursiv AI lessons and check current plan details on the official site.
FAQ
Are professor jobs declining?
Could AI teach a course on its own?
What happens to essays as assessment?
Will class sizes grow because of AI tutoring?
Which academic roles are most at risk?
Where to Start This Term
Take one assessment in one course and redesign it so that a student who used a model to produce it would still have to demonstrate understanding in person. A five-minute oral defence of a submitted piece is usually enough. Run it once, note what it reveals, and you will have something most departments currently lack: direct evidence about what students actually understand, and a defensible answer to what your course certifies.