Commencement speakers are getting booed for mentioning AI. Faculty op-eds are calling it an existential threat to education. Meanwhile, it's already on every job site the graduating class is about to enter.
According to the 2026 State of Higher Ed Report, most institutions are responding to AI from a reactive posture, caught between permissiveness and prohibition, without a clear framework for what to do about a technology that's already reshaping the labor market their students are entering.
The two dominant institutional responses, fear and avoidance, are both inadequate on their own terms, and neither prepares students for the workplace they're walking into.
The institutions that figure out AI fluency will produce more employable graduates. The ones that don't will send students into a labor market they're not equipped to navigate.
The fear response treats AI as a threat to academic integrity, intellectual development, and the value of credentials. This concern is not baseless. AI can facilitate plagiarism, can substitute for genuine intellectual work, and does raise legitimate questions about what a degree is certifying when students can use AI to produce assignments that would previously have required significant original effort.
The avoidance response treats AI as a passing phenomenon or a domain primarily relevant to technical fields. Some institutions have handled AI by not handling it: delaying formal policy, avoiding the conversation in curriculum, or leaving faculty to manage it individually with inconsistent results.
Kevin Prentiss, Head of Product and Technology at the NSLS, described the problem with both approaches in the State of Higher Ed webinar, comparing AI to a chop saw: dangerous, powerful, present on every job site, and requiring skill, respect, and training rather than fear or ignorance. A construction program that refuses to teach students how to use a chop saw because it's dangerous isn't protecting students. It's sending them into job sites unprepared.
Students graduating from institutions that lead with fear or avoidance will enter a workforce where their peers from more forward-thinking institutions already have a competency they don't.
Amy Everson, Senior Director of University Recognition and Institutional Events at the American Public University System, was direct about the structural failure of the current dominant approach during the webinar, explaining that AI fluency cannot be confined to a single course.
The one-course AI model, an "AI in Business" elective or an "Introduction to Machine Learning" requirement, addresses a moment in time. The technology changes weekly. What students learn as sophomores will be substantially outdated by the time they graduate, and a course that was current when designed may teach tools that no longer exist by the time it's required.
Sustainable AI fluency requires integration across the student experience: threading AI literacy into coursework, co-curricular programming, orientation, advising, and career preparation. Not one course that certifies competency at a point in time, but an ongoing relationship with a category of tools that updates as the tools update.
This is a curriculum design challenge more than a technology challenge. The relevant question isn't which AI tools to teach, but how to build students who can learn any AI tool quickly, evaluate its outputs critically, and apply it appropriately in their field. That answer runs through critical thinking, information evaluation, and disciplinary judgment, competencies higher education has always developed, now applied to a new domain.
Prentiss's chop saw analogy provides a clear framework for institutional response. A chop saw is dangerous if you don't know how to use it, and construction programs don't respond to that danger by keeping students away from the equipment. They respond by teaching proper technique, safety protocols, appropriate use cases, and the judgment to know when a different tool suits the task better.
The same framework applies to AI. Institutions need to teach technique, giving students enough practical literacy to prompt effectively and understand model limitations. They need to teach safety, since failure modes like hallucination, bias, and overconfidence in output quality are real and require students who can recognize them rather than trust AI uncritically or avoid it entirely. They need to teach appropriate use, since AI is more or less suited to different professional contexts, and students need judgment about when human evaluation is irreplaceable. And they need to teach the judgment layer itself, the ability to assess AI outputs critically and integrate AI-generated content with genuine intellectual contribution, which is the skill that makes AI a tool rather than a replacement.
One of the more practical frameworks for institutional implementation comes from how Prentiss described cultural adoption in the State of Higher Ed webinar: the dance floor model. When a new song comes on, a few confident early adopters move to the dance floor first, and their visibility gradually draws more participants. The people at the edge of the floor aren't unwilling to dance; they're watching for the signal that it's safe to try.
Institutional AI adoption follows a similar pattern. Early adopters, faculty integrating AI thoughtfully, departments with clear frameworks for AI use, programs that have built AI literacy into their curriculum, create visibility that gives others permission to engage. Institutions that make early adoption visible, celebrate it, and build on it will shift their culture faster than institutions that either mandate top-down adoption or leave everyone to figure it out alone.
This doesn't require large-scale investment. It requires intentional visibility: spotlighting the faculty members doing this well, sharing what's working, and making explicit that engaging with AI seriously is valued and supported.
The most effective approach is assignment design rather than prohibition. Assignments that require genuine intellectual contribution, synthesis, argument, judgment, and original analysis, can be supported by AI tools without being replaced by them. The goal is not to prevent AI use but to design learning experiences where AI assistance doesn't eliminate the learning objective.
The better question is whether AI literacy should be integrated across requirements rather than added as a separate one. A standalone AI requirement may be outdated before it's implemented, while AI literacy woven through existing requirements can update alongside the technology. The institutions with the most resilient approach are threading it across curriculum rather than siloing it.
It treats AI as content to be learned at a point in time rather than a capability to be developed continuously. A course that was current when designed may teach tools that are obsolete by graduation. Sustainable fluency requires ongoing engagement with a category of tools, not certification of competency at a specific moment.
Co-curricular programming can fill gaps in curricular coverage. Structured programs that encourage students to experiment with AI tools in the context of professional development, with reflection built in, develop fluency without requiring specialized faculty. The goal is student engagement with the technology, not institutional mastery of it.
The institutions that produce the most employable graduates in the next decade will be the ones that taught students to use AI effectively, not the ones that protected students from it. For the full data on AI fluency, employer expectations, and what higher education can do to prepare students, read the 2026 State of Higher Ed Report.