Three years ago, I wrote an article called “AI and the Adjuncts: The Converging Trends Reshaping Higher Education.” At the time, the argument felt speculative, though not far-fetched. AI was beginning to move into teaching, tutoring, grading, writing support, and academic administration. Adjunct labor had already become one of the hidden foundations of American higher education. The question was whether these two trends would eventually collide. They have. The collision is now visible in one of higher education’s most sacred public rituals: graduation season.
Every graduation ceremony is a promise dressed up as a ritual. The robes, the music, the procession, the proud families leaning into the aisle with their phones, the president’s remarks about courage and purpose, the handshakes repeated hundreds or thousands of times. All of it says the same thing: you came here to be prepared, and now you are ready.
That promise has always been partly aspirational. Colleges do not merely certify completed coursework. They certify readiness. They tell graduates that the late nights, loans, papers, exams, internships, office hours, awkward seminar discussions, and hurried meals between classes have added up to something durable. A person has been formed. A mind has been strengthened. A future has been made more possible. During the 2026 commencement season, that promise started to sound more fragile.
At several ceremonies this spring, speakers who praised artificial intelligence were booed. Not the polite cough of disagreement. Something closer to collective suspicion. At the University of Central Florida, a speaker told arts and humanities graduates that AI “represents the next industrial revolution.” The crowd booed until she paused. At the University of Arizona, former Google CEO Eric Schmidt faced a similar reception when he told graduates they would help shape artificial intelligence. At Middle Tennessee State University, AI drew the same negative reaction. A Marquette University graduate who majored in digital media put the contradiction plainly: students are discouraged from using AI and penalized for using it, and then the commencement speaker is a champion of AI.
The students were not being childish. They were noticing a contradiction before many institutions had found the language to describe it.
The contrast at Carnegie Mellon was instructive. Nvidia CEO Jensen Huang mentioned AI and received no audible pushback. The difference appeared to lie in the audience’s relationship to the technology. Engineering students see AI as a tool they will build. Humanities graduates see it as a threat to the careers they were being celebrated for pursuing.
Higher education is telling graduates that AI fluency will matter. That is true. It is also telling faculty to police AI use, redesign assignments, preserve academic integrity, and somehow maintain the value of the degree. That is also true. At the same time, colleges continue to rely on a vast contingent labor force to do much of the actual teaching. Contingent faculty now represent roughly 68 percent of all instructional appointments in U.S. higher education, and the people doing the most teaching are often the least protected: median adjunct pay runs about $1,166 per credit hour, and only about a third of institutions offer adjuncts health benefits. That is the hidden architecture behind many commencement promises.
The older article argued that AI and adjunct labor were not separate stories. They were converging pressures on the same system. Three years later, the convergence is no longer theoretical. AI has entered the classroom. Student use has become widespread. Employers are questioning old signals of competence. Grades are becoming harder to interpret. Graduates are anxious about the first rung of the labor market. The people most likely to manage the day-to-day consequences are often those with the least institutional security.
The danger is not that robots will suddenly replace professors. That was always the shallow version of the story. The more plausible danger is quieter and more bureaucratic. AI may become another layer placed on top of an already stressed teaching system. It may ask adjuncts to handle larger classes, more student anxiety, more academic integrity concerns, more software, more policy ambiguity, and more institutional contradiction, all without more pay, more time, or more voice in governance. That would not be innovation. It would be extraction with better branding.
The central question for higher education is not whether AI belongs in the classroom. That question has already been answered by student behavior. Global student AI usage jumped from 66 percent in 2024 to 92 percent in 2025. Students are using these tools to brainstorm, summarize, draft, translate, code, debug, study, and sometimes cheat. The technology is already woven into the ordinary texture of student work. The real question is whether higher education can redesign itself around that reality without hollowing out the human center of education.
Three questions should guide the next phase.
What Can Be Done to Improve Outcomes?
Colleges have spent too much energy on AI permission policies and not enough on AI learning design. A permission policy says what students may or may not do. A learning design asks what students must actually learn now that AI exists.
That distinction matters. A college can ban AI in a syllabus and still produce students who are unprepared for the world they are entering. A college can allow AI everywhere and still produce graduates who cannot think without it. The goal is not prohibition or surrender. The goal is disciplined use.
Better outcomes will require verified learning. Students should show the process, not just the product. A final essay matters less when machines can produce polished prose. Drafts, source trails, revision memos, oral defenses, live presentations, code walkthroughs, lab notebooks, portfolios, and reflective explanations matter more. A student should be able to say not only what the answer is, but how the work was done, what assistance was used, what was rejected, what was verified, and what judgment the student brought to the process.
That is not anti-AI. It is anti-fakery.
The research supports this. A Carnegie Mellon study of more than 670 seventh-grade students found that students in a human-AI tutoring program outperformed students tutored by AI alone, advancing more than a third of a grade level further by year’s end. The lesson is not complicated: AI support works best when it is paired with human guidance. Institutions often miss obvious lessons when money is tight, but this one should be impossible to ignore. AI should not be used to remove the human relationship from education. It should be used to make that relationship more available, more focused, and more powerful.
The best version of AI in education will give students more feedback, not less. It will help multilingual students navigate difficult texts. It will support disabled students. It will offer practice, tutoring, examples, and low-stakes correction. It will help faculty reduce routine burdens so that more time can be spent on mentoring, judgment, and the human work of teaching.
None of that happens automatically. And none of it happens if the people doing the teaching are excluded from the design.
What Can Be Done to Solidify Value Over Time?
Here is a number that should trouble every provost in the country: a UC Berkeley study of more than 500,000 grades at a large research university found that A grades in courses with AI-exposed tasks rose by 13 percentage points after ChatGPT’s release. That is roughly 30 percent more A’s relative to the 2022 baseline. The effect was largest where homework carried greater weight, consistent with AI substituting for student work rather than enabling deeper learning.
A transcript is too thin for the AI age.
It says a student completed courses and earned grades. When those grades are inflating because machines are doing portions of the work, the transcript is no longer a reliable record of what a student can do. Employers are already responding: the share of employers on Handshake requiring a minimum 3.5 GPA jumped to nearly 25 percent in 2026, from 9 percent in 2020. That is not a vote of confidence. It is an attempt to compensate for a signal they no longer trust, and it punishes students who actually learned without AI assistance.
The response should not be nostalgia. Blue books alone will not save higher education. Neither will surveillance software or unreliable AI detectors. The better response is to make the degree a richer record of demonstrated capability. Colleges need portfolios, capstones, authentic assessments, oral demonstrations, community projects, internships, research experiences, and competency records that show what graduates can actually do. The degree must become less like a receipt and more like a body of evidence.
This is where something counterintuitive emerges from the data. While public confidence in higher education has dropped sharply, with 70 percent of Americans telling Pew Research that the system is headed in the wrong direction, employers have not abandoned the degree. The AAC&U’s 2025 employer survey found 70 percent of employers expressing strong confidence in higher education. Eighty-five percent said colleges are preparing students well for the workforce. Seventy-three percent agreed a degree is worth the financial investment. But here is the condition: 90 percent said it is important for graduates to develop AI-related skills while in college.
Employer confidence is not unconditional. It is a bet that institutions will adapt. What employers want is evidence that graduates possess judgment, adaptability, communication, ethical reasoning, and the ability to learn new tools. Those are not trivial skills. They are the deepest purpose of education.
A college that merely delivers content will be in trouble. Content is abundant now. A college that cultivates judgment, craft, trust, adaptation, ownership, systems thinking, and meaning still has a future. But the window for proving that is not indefinite.
When Choices Must Be Made, What Values Belong at the Top?
That is the question colleges have been avoiding. AI forces institutional values into the open. Every adoption decision is a ranking exercise. Do we value efficiency over dignity? Scale over trust? Affordability over mentorship? Student convenience over learning integrity? Administrative control over shared governance?
Three values should anchor the response.
The first is learning integrity. Does the decision increase real student capability, or does it merely improve the appearance of performance? That is the anchor. A university that loses this value loses its reason for existing. When 92 percent of students are using AI and grades are inflating at the rates Berkeley documented, calling academic dishonesty a student-behavior problem misframes the scale. This is a systemic condition requiring systemic design. The question is not whether students are cheating. The question is whether the institution has built a learning environment where the distinction between genuine and performed competence still holds.
The second is human dignity. Students are not throughput. Faculty are not content-delivery units. Adjuncts are not disposable labor patches. Education is an encounter between people, texts, problems, tools, traditions, and futures. Technology can enrich that encounter. It can also cheapen it. The AAUP’s 2025 report on AI and academic professions documented what cheapening looks like in practice: top-down technology decisions made without faculty input, work intensification without compensation, surveillance risks, and erosion of intellectual property rights. At Boston University, administrators suggested faculty use AI to cover for striking graduate students. Faculty unions at CUNY won contract language requiring human instruction for every scheduled course. The Rowan College union secured language stating AI shall not cause the replacement, displacement, or reduction of any unit member’s base workload. These are not Luddite gestures. They are attempts to draw a line between innovation and exploitation.
The third is what might be called faculty sustainability, though the phrase is too bloodless for what it describes. Teaching quality is not separable from teaching conditions. The person grading the AI-assisted essay, redesigning the assessment, responding to the anxious student, and preserving the meaning of the course needs time, support, training, and economic stability. More than a quarter of adjuncts earn under $26,500 annually, below the federal poverty line for a family of four. Seventy-six percent of part-time contingent faculty are on short-term, nonrenewable contracts. An adjunct can lose all scheduled work with seven days’ notice. A system that depends on this workforce for 68 percent of its instruction while excluding them from AI governance is designing failure into its own future. The research gap itself is telling: when Lexi searched for data on whether adjuncts receive AI training, instructional design support, or paid course development time, nearly nothing existed. Adjunct labor is not included in the institutional AI conversation because adjuncts are not included in institutional conversations, period.
These three values do not stand alone. They carry others with them. Equity of access matters: AI could democratize tutoring, translation, and feedback, or it could build a two-tier system where affluent students get human mentorship plus AI while everyone else gets automation and thin staffing. Durable economic value matters: with the underemployment rate for recent graduates reaching 42.5 percent in late 2025, the highest since 2020, education must prepare students not just for today’s tools but for the capacity to keep learning when the tools change. Public purpose matters: college is not only job training, and graduates become citizens, neighbors, voters, and institutional participants in a democracy that needs people who can evaluate claims, argue honestly, and make judgments that are not merely efficient but wise. Institutional trust matters: families are making enormous sacrifices, students are borrowing against uncertain futures, and public confidence has weakened to a 15-year low. Trust is rebuilt through transparency, rigor, and honest evidence of learning, not through commencement rhetoric alone.
But learning integrity, human dignity, and the conditions required for good teaching are the load-bearing walls. Everything else depends on whether those hold.
The Promise
This is why the commencement backlash matters. The students were not rejecting technology itself. Many will use AI every day. Some will build it. Some will depend on it. Some will be harmed by it. What they seemed to reject was the breezy inevitability of the sales pitch. They are right to be suspicious of inevitability.
The future is not a weather system. It is a set of choices. AI will continue. That much is clear. But how it is governed, who benefits from it, who bears its costs, and what values it serves remain open questions.
A good college should be able to say to a graduate: you are not prepared because you avoided AI. You are not prepared because you used AI. You are prepared because you learned how to think with powerful tools without surrendering your judgment to them. You learned how to produce work that can be explained, defended, revised, and trusted. You learned from human beings who had the time and support to teach you well. You learned that efficiency is not the highest human value.
That is the promise worth making. That is the promise higher education now has to keep.