AI is rapidly changing the conversation about postsecondary value because it changes the production function of postsecondary education and academic research.

By “value,” I mean more than the individual return on investment to students, though that matters enormously. Postsecondary value also includes the costs and benefits for institutions, communities and states, the research enterprise, and society more broadly. It includes whether students gain the education and credentials they need to build meaningful lives and careers; whether institutions can deliver that education and sustain their research and public missions; whether states see returns in economic growth, workforce capacity, civic participation, and public budgets; and whether society benefits from knowledge production, democratic participation, and human flourishing.

A flagship research university, regional public university, community college, liberal arts college, and online adult-serving institution will not answer these questions in the same way. Each needs to be clear about the value it promises, the students and communities it serves, and the costs it can justify.

The value question was already hard

That broader value proposition was already under pressure. Families are asking whether college is worth the cost. States are asking whether public investments in higher education are producing clear economic and social returns. Employers are asking whether graduates are prepared for work that is changing quickly. Institutions are trying to answer those questions while navigating financial strain, political scrutiny, demographic change, and fragile public trust.

For individual students, the evidence still supports the enduring value of postsecondary education, but it also points to real variation in risk and return. The Federal Reserve Bank of New York recently estimated that the return to college for the median graduate remains about 12.5 percent, well above the threshold for a sound investment, while emphasizing that the return varies significantly by time to completion, cost, wages, and major.[1] The College Board’s Education Pays reports that higher levels of education are associated with earnings and employment advantages as well as differences in health-related behaviors, reliance on public assistance, civic participation, and social mobility.[2]

Postsecondary attainment can create public value as well as private return. For example, my Ithaka S+R colleagues examined the relationship between postsecondary attainment and state finances in It’s Complicated, finding that higher attainment is generally associated with higher state tax revenues and lower spending on Medicaid and welfare programs, though the size of those effects varies substantially by state.[3] My colleagues Elizabeth Pisacreta and Cameron Childress have also applied this framework in state-specific work, including an analysis of postsecondary investment in Mississippi estimating that raising attainment among 25-to-64-year-olds from 42 percent to 60 percent could generate more than $376 million annually through increased tax revenue and reduced social services spending.[4]

Academic research adds another dimension. The value of higher education extends beyond teaching students to producing, validating, preserving, and disseminating knowledge. US higher education institutions reported more than $108 billion in research and development expenditures in FY2023.[5] Those investments support discovery, innovation, graduate training, regional economic development, and the knowledge infrastructure on which public policy, medicine, technology, and democratic life depend.

Notwithstanding this record, public trust in higher education has fallen significantly and is near an all-time low. Gallup and Lumina found that confidence in higher education rose to 42 percent in 2025 after sitting at 36 percent in 2023 and 2024, but remains well below the 57 percent recorded in 2015.[6] Part of the reason is that the distribution of risk matters as much as the average return. As more of the cost of postsecondary education has shifted to students and families, the downside risk of noncompletion, delayed completion, weak labor-market outcomes, or poorly aligned programs has become more consequential. Add to that secular trends questioning the legitimacy of longstanding institutions, populist challenges to scientific expertise, and partisan critiques of higher education’s dominant norms and policies, and it is a recipe for skepticism.

AI changes the production function

In a recent essay in Persuasion, Nils Gilman argues that AI is exposing the fragility of what Clark Kerr called the “multiversity”: the modern research university as a bundle of teaching, research, credentialing, social formation, and knowledge production. His thesis is in the title: “The university as we know it is finished.”[7] I do not think we need to accept the strongest version of Gilman’s claim to take the provocation seriously. AI is forcing institutions to ask which parts of that bundle remain mutually reinforcing, which are under strain, and how postsecondary education can preserve its deepest sources of value while adapting to new technological and economic realities.

Put another way, AI challenges the production function of postsecondary education and academic research. (I’m grateful to the economist Betsey Stevenson for introducing me to this phrase and concept.) AI changes the inputs, processes, costs, and outputs of teaching, learning, administration, and scholarship. It may make some activities cheaper or faster. It may make some familiar measures of quality less reliable. It may shift value away from the production of outputs and toward judgment, trust, interpretation, relationship-building, and human purpose in the choice and process of creating those outputs. And it may expose places where institutions have depended on inherited structures without clearly explaining what value those structures create.

AI touches the basic assumptions that underlie postsecondary value: what students need to learn, how they learn, how institutions operate, and how universities contribute to the production and dissemination of knowledge.

The question extends beyond how existing institutions should use AI. It also asks which functions still need to be bundled together, which can be shared across institutions, and which may need to be rebuilt outside the traditional institutional frame.

AI touches the basic assumptions that underlie postsecondary value: what students need to learn, how they learn, how institutions operate, and how universities contribute to the production and dissemination of knowledge. The immediate questions it raises are urgent and real. Students need AI literacy and training on work-relevant AI tools. Faculty need to rethink assignments and assessment. Institutions are trying to understand how AI improves (or complicates) operational workflows. Researchers need to understand how AI will affect methods, workflows, funding, and infrastructure. Libraries and publishers need to manage new tools for discovery, access, rights, and trust.

The harder questions are more consequential. What forms of human judgment become more valuable when routine cognitive work is automated? How can the learning experience be designed to retain friction, feedback, and human interaction? What parts of the institutional bundle remain worth paying for? How should universities sustain independent knowledge production when the data, models, compute, and talent needed for frontier AI work increasingly sit outside the academy? And what standards of provenance, integrity, rights, and trust should govern AI-mediated discovery?

Those questions shape the rest of this essay: what students need to learn, how they learn, how institutions operate, and how knowledge is produced and trusted.

What students need to learn

At the most immediate level, students need to understand when and how to use AI tools effectively: how to prompt effectively, assess outputs, identify errors, understand limitations, and apply judgment about when not to use AI. This is why AI literacy has become a core institutional priority. Ithaka S+R’s work in this area has engaged institutions in defining and implementing AI literacy, including through cohorts that bring together libraries, centers for teaching and learning, and other campus leaders to ask how AI literacy can become part of undergraduate education.[8]

That facility with AI also needs to be grounded in experience with the kinds of AI tools students are likely to encounter in real work settings, not only general-purpose chatbots. If graduates have practiced only with simplified or classroom-contained tools while workplaces are using integrated systems for coding, analysis, design, writing, and project management, they will be underprepared for how AI is actually changing knowledge work.

But AI literacy is not enough.

PwC’s 2026 Global AI Jobs Barometer found that entry-level roles most exposed to AI are now seven times more likely to require traditionally senior-level human-intensive skills, such as leadership, creativity, and face-to-face interaction.[9] Strada’s recent employer survey similarly found that employers value critical thinking and communication more highly than AI literacy in entry-level hires, and that related work experience, internships, and project-based experience matter more than a perfect GPA with no formal work history.[10]

This creates a central paradox for postsecondary education. The work AI is least able to do well—forming goals, building trust, exercising judgment, persuading others, managing relationships, and acting responsibly in a community—is often learned through experience. Historically, many early-career workers developed judgment by doing more basic work like drafting, summarizing, fact-checking, and being corrected through human feedback. If AI takes over too much of that entry-level work, how will new graduates develop the understanding, judgment, and professional maturity needed to supervise, evaluate, and direct AI-enabled work?

This is one of the most important value questions in the age of AI. Institutions cannot simply teach students to use tools and assume they will develop judgment later. They will need to create learning environments where students practice judgment, persuasion, collaboration, and management before they enter the workforce. As Gilman puts it, the curriculum must be organized around what AI can do and around the human capacities that become more valuable because AI cannot do them.[11]

They will also need to rethink the relationship between institution and learner. If AI keeps changing the nature of work, postsecondary education cannot be understood only as an event that happens once, mostly between the ages of 18 and 24. It will need to become a continuing relationship: a way for learners to return, update skills, test new knowledge, and reconnect with communities of learning over the course of their lives.

One promising approach is to connect liberal education and applied, work-relevant skill development more deliberately and adaptively over time. A strong liberal arts education can cultivate judgment, communication, ethical reasoning, adaptability, and the ability to work across difference. Integrating experiential learning that positions students to connect those capacities to specific AI-related or industry-relevant skills is a potentially powerful combination.

How students learn

AI also changes how students learn.

Here, too, the immediate story is familiar. AI can provide tutoring, feedback, translation, accessibility support, and always-available assistance. It can help instructors design assignments, generate examples, and support students who need more practice. UNESCO has noted that AI may create new opportunities for access and personalized learning, while also warning that these benefits require attention to inequality, privacy, and safety.[12]

Carefully designed AI tutors can be part of that positive story when they scaffold learning rather than simply provide answers. Tools that prompt students to reason through problems, offer targeted feedback, and keep the learner engaged in the work can improve outcomes, as Terri Taylor discusses in EdSurge.[13]

At the same time, AI makes cheating easier and raises immediate questions about academic integrity. A recent Cornell summary of research published in Science reported that AI misuse creates problems for assessment validity and the credibility of university credentials.[14] Many dominant forms of assessment were designed for a world in which producing a memo, essay, problem set, or summary was itself strong evidence of learning. AI weakens that assumption.

That makes it important to be clear about what assessment is for. If assessment functions mainly as a mechanism for assigning grades, AI creates obvious incentives to produce the grade with less work. If assessment is meant to motivate practice, certify capability, and help students understand what they can actually do, then institutions need designs that make the learning itself harder to bypass and more visibly worth doing.

If AI removes too much friction from learning—drafting the first version, finding the source, struggling through the problem, forming the argument, making the mistake—it may make it harder for students to develop the understanding they need to evaluate AI outputs later.

This is where the paradox deepens. If AI removes too much friction from learning—drafting the first version, finding the source, struggling through the problem, forming the argument, making the mistake—it may make it harder for students to develop the understanding they need to evaluate AI outputs later. Cognitive outsourcing can be useful. But if students outsource too much, too early, they may lose the very experience that builds judgment.

Emerging evidence points in that direction. A recent arXiv preprint, “Faster Completion, Less Learning,” uses a large panel of ALEKS learning interactions and finds that generative AI reduced study time on AI-susceptible math problems while also reducing durable learning outcomes.[15]

To adapt, both the learning experience and assessment will need to focus more on process, and less on outputs alone. Students should have to explain choices, defend claims, test outputs, revise in response to evidence, and work with others to produce something of value. In some cases, that will mean more oral examination, more iterative assignments, more project-based work, more collaborative problem solving, and more careful integration of AI into the assignment itself.

This shifts the role of faculty as well. The professor becomes less a conveyor of information and assessor of student outputs, and more an interlocutor, designer of learning environments, assessor of judgment, and guide to disciplinary standards. That is a more demanding role.

The redesign should also pay attention to peer learning and social capital. If a private AI assistant becomes the easiest study partner, students may have fewer reasons to form study groups, ask classmates for help, or build the informal networks through which much learning and opportunity have long flowed. Julia Freeland Fisher and Anna Arsenault caution that AI-enabled navigation and guidance tools may expand access to information while making isolation more convenient unless institutions explicitly design for human connection and network-building.[16] The answer is to make collaboration, mutual accountability, and social capital part of the learning model rather than leaving them to chance.

This is also where the AI response becomes a civic project. Deliberation, disagreement, collaborative problem-solving, and careful attention to evidence are not incidental classroom habits. They are democratic capacities. If AI makes it easier to avoid difficult thinking or difficult people, colleges will need to make the opposite move: creating structured settings where students practice reasoning with others, listening across difference, and working through hard problems together.

How institutions operate

AI will also affect institutional operations.

At the surface level, the opportunity is straightforward: AI may reduce the cost of producing some institutional services. Colleges and universities can use AI to improve administrative workflows, student support, research administration, advising, library operations, communications, and routine analysis. My colleague Claire Baytas and I have also written about how AI is being incorporated into the workflows of admissions, financial aid, and enrollment management.[17] Ithaka S+R’s AI in Action cohort for academic libraries is focused on exactly this kind of practical experimentation: how academic libraries can explore, test, and assess AI applications in areas such as metadata creation, patron services, workflow optimization, and collections management while remaining aligned with library values.[18] On the research side, my colleagues Dylan Ruediger and Ruby MacDougall, with several collaborators, recently examined AI adoption in research administration at emerging research institutions, showing both the promise of AI for reducing administrative burden and the need for careful attention to data security, trust, workforce development, and institutional strategy.[19]

We should not assume that AI will immediately reduce costs. New technologies often require new staffing, training, governance, procurement, security, and support. In some use cases, the cost of compute is high relative to the cost of existing human-led processes. In the short run, AI may add costs even as it creates the possibility of longer-term efficiencies.

But eventually, there will be efficiencies, and those efficiencies matter. Postsecondary value depends partly on cost. If institutions can reduce administrative burden, improve service quality, or lower the cost of routine work, they may be able to redirect resources toward student support, instruction, research capacity, or affordability. In a sector where many institutions face structural financial pressure, operational efficiency is not a distraction from mission. It can be a condition for sustaining it.

But the deeper question is what institutions do with the savings, and what they are willing to protect.

The parts of education that may matter most in an AI world are the hardest to automate: sustained faculty interaction, mentoring, rigorous feedback, collaborative problem-solving, research apprenticeship, community formation, and the relationships that help students develop judgment and confidence. These require time, attention, and trust.

That creates a real strategic choice. Efficiencies in administrative and routine instructional work could free resources to invest in the human-intensive dimensions of education that AI makes more important. Or the drive for efficiency could hollow them out, producing lower-cost experiences that are easier to deliver but less capable of building the judgment, belonging, and applied understanding students need.

The challenge is to use AI and shared infrastructure to reduce costs where they should be reduced, while protecting and expanding access to the forms of human interaction that matter most.

There is a risk that what different institutions can offer, and what different students can access, will exacerbate inequality. Students with means may continue to have access to seminars, mentoring, internships, research experiences, advising, and high-touch support, while students with fewer resources are offered automated substitutes. It is incumbent on institutions to redesign advising, coaching, mentoring, applied learning, and feedback so that more students, including working adults and part-time learners, can access the human support that matters most.

The challenge is to use AI and shared infrastructure to reduce costs where they should be reduced, while protecting and expanding access to the forms of human interaction that matter most. Different institutions serve different students, missions, and communities. But all will need to make clearer choices about where human interaction is essential, where AI can responsibly improve service, and where collaboration or shared infrastructure can reduce costs without hollowing out value.

How knowledge is produced and trusted

AI also poses deep questions for academic research and scholarly communication.

At the immediate level, researchers across disciplines are already finding ways to deploy AI tools in their work: scanning and synthesizing literature, generating hypotheses, writing code, analyzing data, identifying patterns, drafting text, and supporting peer review. A recent Nature survey of researchers found strong interest in using AI to make research faster, easier, and more accessible, while also surfacing concerns about accuracy, ethics, transparency, and the need for clearer guidance.[20] Ithaka S+R’s work on generative AI and higher education has similarly emphasized that AI is beginning to affect every stage of the research cycle, from methods and workflows to scholarly communication.[21]

But the deeper question is whether AI changes how research is conducted and what counts as a research contribution.

The research paper has long served as the central artifact of scholarly communication: the place where evidence, method, argument, and contribution are made legible to a field. AI may complicate that model. Early efforts such as Orchestra’s Agent-Native Research Artifacts and related work on machine-executable research packages,[22] point toward a world in which research outputs are designed for human readers and for AI agents—or in some cases become agents[23]—that can reproduce, extend, and interrogate the data and findings. These approaches may or may not become dominant. But they highlight a real shift: research outputs may increasingly become dynamic, executable, and machine-mediated.

That creates enormous opportunity. AI may help researchers work across larger literatures, test more hypotheses, surface hidden patterns, and accelerate discovery. But it also raises serious questions about trust. If AI systems retrieve, summarize, synthesize, or extend scholarship, how will users know what sources were used, whether a finding has been peer reviewed, whether a paper has been retracted or corrected, whether rights have been respected, and whether provenance has been preserved?

Publishers and scholarly infrastructure providers are now wrestling directly with those questions. My colleague Todd Toler has framed this as the need for “scholar-ready AI”: systems that reflect integrity, transparency, rights awareness, provenance, and institutional trust. As Todd and Lisa Janicke Hinchliffe discuss in The Scholarly Kitchen, AI retrieval, provenance, rights signaling, and shared infrastructure are becoming central issues for scholarly communication.[24] The deeper institutional question is whether universities can continue to play their historic role as centers of independent knowledge production when some of the most important resources for AI-enabled research sit outside the academy. Recent research from the University of Chicago’s Becker Friedman Institute documents a growing migration of AI talent from universities to industry, driven in part by large and widening compensation gaps.[25] Access to data, the latest models, and compute resources likely also play a role in pulling researchers away from the academy. The Stanford AI Index reports that industry produced nearly 90 percent of notable AI models in 2024, while academia remained the leading producer of highly cited AI research.[26]

That shift raises foundational questions for research universities. If frontier research capacity depends on infrastructure owned by private companies, datasets unavailable to most scholars, and compensation levels universities cannot match, what role should universities play in producing independent, public-interest knowledge? How can they sustain open research when more discovery may occur inside firms whose incentives are proprietary? Public efforts such as the National Science Foundation’s National Artificial Intelligence Research Resource are one response, aiming to provide researchers and educators with access to compute, data, models, software, and expertise.[27] But the broader challenge remains: universities must rethink what infrastructure, partnerships, and governance are necessary to sustain the research enterprise in an AI era.

Libraries are wrestling with this transformation from another angle. At the immediate level, they are evaluating AI-enabled discovery tools, supporting AI literacy, negotiating licenses, preserving access, advising on information integrity, and experimenting with AI in their own operations. Ithaka S+R’s 2025 US Library Survey shows library leaders navigating pressure around technology, funding, institutional alignment, and value.[28]

The deeper question is what libraries become when discovery, synthesis, and interpretation are increasingly mediated by AI. Their role has never been only to provide access to content. Libraries help institutions make knowledge discoverable, usable, preserved, interpretable, and trustworthy. In the age of AI, that work becomes more complex and more central. Libraries may need to serve as stewards of collections and of trust: helping institutional stakeholders understand provenance and authority and navigate opaque commercial systems.

The value of higher education depends on what students learn and on whether universities remain trusted places for producing, validating, preserving, and disseminating knowledge.

A value agenda for the age of AI

If AI changes the production function of education and research, the value agenda cannot be limited to AI adoption. It has to ask what should become cheaper, what should become better, and what must remain human.

Postsecondary value will depend on whether institutions help people do what AI cannot do alone: form purposes, make judgments, build trust, contribute to communities, and advance knowledge in service of human flourishing. It will also depend on whether institutions can deliver that value at a price students, families, states, and society can sustain.

AI can (eventually) reduce the cost of some forms of production: routine administration, content generation, information retrieval, basic analysis, and certain kinds of student support. Institutions should use those efficiencies. But they should use them to protect and expand access to what AI cannot provide: sustained interaction with faculty and mentors, rigorous feedback, collaborative problem-solving, research apprenticeship, community, and the experience of developing judgment through practice.

That is the first test of a serious AI strategy: whether it reduces cost in ways that strengthen value, or simply delivers a thinner version of education.

The second test is whether institutions redesign learning around the work students will actually need to do. If AI changes entry-level knowledge work, then students need more than exposure to AI tools. They need repeated opportunities to frame problems, evaluate evidence, defend choices, work with others, and revise their thinking. Learning with AI should resemble responsible work with AI: iterative, social, evidence-based, and accountable.

The third test is whether higher education becomes a continuing relationship rather than a single episode early in adulthood. A degree will still matter. So will the formative experience of college. But in a world where knowledge, tools, and work are changing quickly, institutions will create more value if they help learners return, reorient, and continue learning over time. That requires new models for advising, credentials, alumni engagement, employer partnerships, and public funding.

The fourth test is whether institutions can cooperate where cooperation is necessary. No college or university can solve these problems alone. AI infrastructure, credit mobility, shared assessment models, research capacity, trustworthy scholarly records, and continuing education pathways all require collaboration across institutions, systems, states, libraries, publishers, employers, and funders. This should include cooperation to secure fair and affordable access to advanced models, compute, and AI infrastructure, especially for institutions with fewer resources.

Public policy matters here, especially state policy. States can shore up postsecondary value by investing in shared infrastructure, supporting credit mobility, aligning education and workforce data, funding evidence-based student supports, and creating incentives for institutions to serve adult and part-time learners well. They can also help ensure that AI adoption does not widen the gap between institutions and students with abundant resources and those with fewer.

Collaboration with employers is essential, but it should not mean asking students or taxpayers to subsidize training whose benefits flow mainly to firms. Public investment should prioritize capabilities that increase learner mobility, public value, and long-term adaptability.

The right answer will differ by mission. A research university, community college, liberal arts college, and online adult-serving institution will not make the same choices because they create value for different learners, communities, and public purposes. Some institutions will need to preserve an intensive residential experience. Others will create value through flexible pathways, employer-connected learning, low-cost credentials, research capacity, or trusted knowledge infrastructure. Some will find ways to combine these characteristics in ways that meet their constituents’ needs. What matters is that each institution becomes clearer about the value it is trying to create, the costs it can justify, and the human commitments it is unwilling to sacrifice.

Ultimately, the question is not whether AI makes postsecondary education and academic research more or less valuable. The question is whether institutions in the ecosystem use this moment to become clearer about the value they are uniquely positioned to create.

Endnotes

  1. Jaison R. Abel and Richard Deitz, “Is College Still Worth It?” Liberty Street Economics, Federal Reserve Bank of New York, April 16, 2025, https://libertystreeteconomics.newyorkfed.org/2025/04/is-college-still-worth-it/. ↑
  2. Matea Pender, Jennifer Ma, Xiaowen Hu, and Ashley Edwards, “Education Pays 2026: The Benefits of Higher Education for Individuals and Society,” College Board, 2026, https://research.collegeboard.org/trends/education-pays. ↑
  3. James D. Ward, Benjamin Weintraut, and Elizabeth D. Pisacreta, “It’s Complicated: The Relationship between Postsecondary Attainment and State Finances,” Ithaka S+R, January 19, 2021, https://doi.org/10.18665/sr.314677. ↑
  4. Cameron Childress, James D. Ward, and Elizabeth D. Pisacreta, “Strengthening Mississippi’s Economic Future Through Postsecondary Investment,” Ithaka S+R, January 17, 2023. https://doi.org/10.18665/sr.318143. ↑
  5. “Higher Education Research and Development (HERD) Survey 2023,” National Center for Science and Engineering Statistics, https://ncses.nsf.gov/surveys/higher-education-research-development/2023. ↑
  6. Jeffrey M. Jones, US Public Trust in Higher Ed Rises From Recent Low,” Gallup News, July 15, 2025, https://news.gallup.com/poll/692519/public-trust-higher-rises-recent-low.aspx. ↑
  7. Nils Gilman, “The University as We Know It Is Finished: That’s a Good Thing,” Persuasion, June 17, 2026, https://www.persuasion.community/p/the-multiversity-is-finished; Jeffrey M. Jones, “US Public Trust in Higher Ed Rises From Recent Low,” Gallup News, July 16, 2025, https://news.gallup.com/poll/692519/public-trust-higher-rises-recent-low.aspx. ↑
  8. “AI Cohorts for Higher Education,” Ithaka S+R, https://sr.ithaka.org/ai-cohorts-for-higher-education/. ↑
  9. AI Reshapes Global Labour Market Into Two Distinct Paths, Rewarding Human Skills: PwC 2026 Global AI Jobs Barometer, PWC, June 15, 2026, https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-ai-jobs-barometer.html. ↑
  10. Andrew Hanson and Molly Cook Escobar, “Entry-Level Hiring in the AI Era: What Employers Are Thinking (and Doing),” Strada, May 19, 2026, https://www.strada.org/news-insights/entry-level-hiring-in-the-ai-era-what-employers-are-thinking-and-doing. ↑
  11. Nils Gilman, “The University as We Know It Is Finished: That’s a Good Thing,” Persuasion, June 17, 2026; Jeffrey M. Jones, “US Public Trust in Higher Ed Rises From Recent Low,” Gallup News, July 16, 2024, https://www.persuasion.community/p/the-multiversity-is-finished. ↑
  12. “AI and Education: Protecting the Rights of Learners,” UNESCO, last updated September 25, 2025, https://www.unesco.org/en/articles/ai-and-education-protecting-rights-learners. ↑
  13. Terri Taylor, “AI Won’t Replace Educators. But It Is Changing How Students Learn,” EdSurge, June 12, 2026, https://www.edsurge.com/news/ai-wont-replace-educators-but-it-is-changing-how-students-learn. ↑
  14. Patricia Waldron and Ann S. Bowers, “Widespread AI Misuse Means Higher Ed Must Rethink Assessment,” Cornell Chronicle, May 21, 2026, https://news.cornell.edu/stories/2026/05/widespread-ai-misuse-means-higher-ed-must-rethink-assessment. The article referenced is available at: Igor Chirikov, Ivan Smirnov, and René F. Kizilcec, “Generative AI Use and Misuse Call for Assessment Reform in Higher Education,” Science, May 21, 2026, https://www.science.org/doi/10.1126/science.aec5115.↑
  15. Sina Rismanchian, Hasan Uzun, Jeffrey Matayoshi, Eric Cosyn, and Eyad Kurd-Misto, “Faster Completion, Less Learning: Generative AI Reduced Study Time on Math Problems and the Knowledge They Build,” arXiv, last updated June 12, 2026 (v2), https://doi.org/10.48550/arXiv.2605.21629. ↑
  16. Julia Freeland Fisher and Anna Arsenault, “Navigation & Guidance in the Age of AI: 5 Trends to Watch,” Christensen Institute, January 14, 2025, https://www.christenseninstitute.org/publication/navigation-guidance-ai/. ↑
  17. Claire Baytas and Martin Kurzweil, “AI-Assisted Admissions and Enrollment Management,” in AI Applications in Online Higher Education Administration Strategies for Maximizing Returns and Improving Outcomes, eds. Kathleen S. Ives, Marie Cini, and Ray Schroeder (Routledge, 2026), https://www.routledge.com/AI-Applications-in-Online-Higher-Education-Administration-Strategies-for-Maximizing-Returns-and-Improving-Outcomes/Ives-Cini-Schroeder/p/book/9781032954806. ↑
  18. Tracy Bergstrom, “AI in Action: Announcing a New Cohort Project for Academic Libraries,” Ithaka S+R, April 2, 2026, https://sr.ithaka.org/blog/ai-in-action/. ↑
  19. Dylan Ruediger, Ruby MacDougall, Stefanie Brachfield, Doug Dechow, Jonathan Parker, and Jana Remy, “AI Adoption in Research Administration at Emerging Research Institutions,” Ithaka S+R, March 30, 2026, https://doi.org/10.18665/sr.325252. ↑
  20. Miriam Naddaf, “How are researchers using AI? Survey reveals pros and cons for science,” Nature, February 4, 2025, https://www.nature.com/articles/d41586-025-00343-5. ↑
  21. “Generative AI and Higher Education, Ithaka S+R, https://sr.ithaka.org/generative-ai-and-higher-education/. ↑
  22. See the blog: “The Last Human-Written Paper: Agent-Native Research Artifacts,” Orchestra, 2026, https://www.orchestra-research.com/ara; and preprint: Jiachen Liu et al., “The Last Human-Written Paper: Agent-Native Research Artifacts,” arXiv, last revised May 19, 2026 (v3), https://doi.org/10.48550/arXiv.2604.24658. ↑
  23. Jiacheng Miao, et al, “Paper2Agent: Reimagining Research Papers As Interactive and Reliable AI Agents,” arXiv, last revised October 16, 2025 (v2), https://doi.org/10.48550/arXiv.2509.06917. ↑
  24. Lisa Janicke Hinchliffe, “Building Scholar-Ready AI: A Conversation with Todd Toler,” The Scholarly Kitchen, June 10, 2026, https://scholarlykitchen.sspnet.org/2026/06/10/building-scholar-ready-ai/. For more on Todd Toler’s field-building work see: Todd Toler, “Why I Joined Ithaka S+R,” Ithaka S+R, June 8, 2026, https://sr.ithaka.org/blog/why-i-joined-ithaka-sr/. ↑
  25. Ufuk Akcigit, Craig A. Chikis, Emin Dinlersoz, and Nathan Goldschlag, “Attention (And Money) Is All You Need: Why Universities Are Struggling to Keep AI Talent,” The University of Chicago Becker Friedman Institute for Economics, March 16, 2026, https://bfi.uchicago.edu/insights/attention-and-money-is-all-you-need-why-universities-are-struggling-to-keep-ai-talent/. ↑
  26. Nestor Maslej et al, “The AI Index 2025 Annual Report,” AI Index Steering Committee, Institute for Human-Centered AI, Stanford University, April 2025, https://doi.org/10.48550/arXiv.2504.07139. ↑
  27. “National Artificial Intelligence Research Resource,” US National Science Foundation, https://www.nsf.gov/focus-areas/ai/nairr. ↑
  28. Ellen Carroll, Tracy Bergstrom, and Ioana G. Hulbert, “US Library Survey 2025: Under Pressure,” Ithaka S+R, May 14, 2026, https://doi.org/10.18665/sr.325599. ↑