As I continue looking for research to support some of the premises in a book about the future of colleges in the age of artificial intelligence, I have been looking for evidence that helps separate serious signals from speculation. There is no shortage of commentary about AI and work. Some of it predicts massive displacement. Some of it predicts extraordinary productivity. Some of it dismisses both predictions as hype.
The difficulty for colleges is that we do not have the luxury of waiting to see which prediction comes true. Changes are happening now. Students are making decisions, families are paying tuition, faculty are redesigning assignments, employers are changing expectations, and college presidents and boards are being asked to make strategic decisions today.
A recent Stanford Digital Economy Lab working paper caught my attention. The paper by Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen is titled Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. The researchers updated a paper they first published in August 2025. Their dataset includes high-frequency administrative payroll data from ADP covering millions of U.S. workers through June 2026 to examine labor market changes after the widespread adoption of generative AI.
The title is well chosen. The authors are not claiming that AI has already transformed the entire labor market. They are looking for early signals, the canaries, that might indicate where AI is beginning to alter employment patterns first. For those of us in higher education, the paper matters because the canaries appear to be young workers.
The authors found a substantial divergence for workers ages 22 to 25 in occupations more exposed to AI. The employment rate among these young workers in highly AI-exposed occupations now stands about 19 percent below where it would be if it had kept pace with employment among similarly aged workers in less-exposed occupations. Experienced workers show no comparable gap.
That finding should get the attention of every college president, provost, trustee, dean, career services leader, and faculty member. The 19 percent lower employment does not prove that AI is the sole cause. The researchers carefully avoid declaring an absolute conclusion about that. They describe their findings as descriptive indicators rather than causal estimates. They note that some trends predate ChatGPT, that the results are sensitive to certain education controls, and that the pattern is more pronounced in their ADP analysis sample than in some national survey benchmarks.
Those caveats are important. It would be irresponsible for colleges to use the findings in this paper to tell students that AI has destroyed entry-level work. However, colleges would be equally irresponsible if they ignored the possibility that the first rung of the professional career ladder is already changing.
The Most Important Finding Is Not the 19 Percent
The 19 percent figure is striking and will receive attention. The more important finding is how the divergence appears to be happening. The researchers find that the employment gap for young workers in AI-exposed occupations is driven primarily by reduced hiring, not by increased separations.
If AI were mainly causing layoffs, colleges would be thinking about displaced workers, mid-career retraining, and adult education. Those issues are important, and adult-serving colleges should consider meeting the needs of those groups. Reduced hiring is a different problem. It affects the entry point. It affects newly minted graduates attempting to launch.
A college graduate does not experience “reduced hiring” as an abstract labor market statistic. She experiences it as fewer interviews. He experiences it as a longer job search. They experience it as more competition for internships, analyst roles, junior software jobs, marketing assistant positions, customer support roles, research jobs, and other jobs that have traditionally allowed young people to begin building work experiences.
The first job is not everything. Many successful adults did not begin in the perfect job. I lasted in my first permanent job only five months after I was told the six months training program was extended to a two-year training program. Two years of training wasn’t what I bargained for. Career paths twist and turn. But the first substantial job often matters more than colleges admit. It provides income, confidence, feedback, supervision, professional habits, a network, and a story of competence. If the first rung becomes narrower, students need help climbing onto it.
College leaders should pay close attention to this finding. Too many institutions still treat the transition to work as something that happens after the real education is finished. The Stanford paper suggests that this separation is becoming harder to defend.
Codified Knowledge and Tacit Knowledge
The most interesting part of the paper is the researchers’ discussion of possible mechanisms. They suggest that AI may more easily substitute for knowledge that has been digitized and codified, the kind of formal, standardized, documented knowledge that can be taught through textbooks, procedures, and written materials. By contrast, AI may complement the tacit, experience-based knowledge held by more senior workers.
Much of formal education is built around codified knowledge. Colleges teach students concepts, theories, methods, equations, frameworks, vocabulary, disciplinary histories, and procedures. That learning continues to be important. I do not believe colleges should abandon content or lower academic expectations. A student cannot exercise judgment in a field without knowing something about the field. If generative AI is good at reproducing and applying codified knowledge, however, then colleges must ask what else students need.
The answer to that question is that students need knowledge used in richer contexts. Students need tacit knowledge. They need practice. They need feedback. They need mentors. They need projects. They need to learn what good work looks like in situations where the answer is not fully specified. They need to understand how a professional frames a problem, notices a weak assumption, reads a room, revises an argument, manages a client, interprets ambiguity, and decides when an AI-generated answer is not good enough.
If college is known as the place where codified knowledge is delivered, AI looks like a competitor. If college is understood as a community where students practice applying knowledge under human guidance, AI becomes a tool within a broader educational purpose.
What the Paper Does and Does Not Say
The researchers deserve credit for the care with which they present their findings. They do not claim widespread job displacement or that AI is the only explanation for the patterns they observe. Nor do they predict that the trends will continue indefinitely. They explicitly state that future impacts will depend on technological progress, firm adoption, policy responses, adjustment by firms and workers, and broader macroeconomic forces.
The paper is positioned as an early warning signal. The authors note that the labor market appears to be changing unevenly, with young workers in AI-exposed occupations facing a measurable employment shortfall relative to less-exposed peers.
Colleges often wait too long to respond to labor market changes because they are understandably wary of chasing every trend. That caution is healthy. Universities should not redesign the curriculum every time a new technology appears. But AI is not simply another tool. It is changing how knowledge work is performed, how employers think about entry-level tasks, and how students demonstrate readiness. The right institutional response is disciplined redesign, one of the points I intend to make in my book and a point recently posited by the MIT faculty.
The Entry-Level Bargain Is Changing
For generations, entry-level jobs served as a training ground. Employers hired new graduates who had strong general preparation but limited practical experience. Those graduates learned by doing the routine tasks of a profession: drafting, coding, analyzing, summarizing, scheduling, researching, supporting customers, preparing presentations, cleaning data, writing reports, or assisting senior colleagues.
Much of that work was formative, if unglamorous. It allowed young workers to learn how organizations operate. It gave them proximity to more experienced workers. It exposed them to tacit knowledge. It gave them supervised mistakes. AI may be changing the availability of some of that vital work.
If a senior employee can use AI to draft the first memo, summarize the documents, prepare the code, answer the customer, analyze the dataset, or generate the presentation outline, the organization may need fewer junior people for those tasks. In some workplaces, AI may increase the productivity of experienced workers more than it increases the demand for beginners. That does not mean employers will stop hiring young people. It could mean that employers expect new graduates to arrive with more evidence of their ability to contribute quickly.
The National Association of Colleges and Employers’ (NACE) 2026 research points in the same direction. Employers report that more than one-third of entry-level jobs require AI skills, nearly triple the share from six months earlier, and 28 percent of employers say they are seeking early-career talent who can use AI in their work. At the same time, slightly more than half of graduating seniors say they are not currently building AI skills for the future.
That gap should worry colleges. It is not enough to tell students, “AI is important.” Students need structured opportunities to learn how AI impacts careers in their field. They need to know when to use it, when not to use it, how to check it, how to disclose it, and how to remain responsible for the work produced with it.
Students who graduate from college without AI fluency will be at a disadvantage. Students who graduate with superficial AI fluency may also be at a disadvantage. Employers will not need graduates who can merely prompt a chatbot. They will need graduates who can combine AI tools with judgment, communication, ethics, domain knowledge, and practical problem-solving.
What Should Colleges Do?
The Stanford paper does not provide a college strategy. That is not its purpose. But it points toward several educational imperatives.
1. Colleges should stop treating career preparation as a senior-year service.
The career question should begin in the first year. Students should be helped to understand their strengths, interests, values, and possible pathways. They should learn how different fields are changing because of AI. They should meet alumni earlier. They should begin building evidence of capability long before they apply for full-time jobs.
2. AI fluency should become part of the core curriculum.
Not every student needs to become a computer scientist, but they should understand how AI tools work at a basic level, what they are good at, where they fall short, how they shape work across different fields, and how to use them ethically and effectively.
A student majoring in English needs to understand AI’s effect on writing, authorship, editing, and source evaluation. A business student needs to understand AI’s effect on analysis, marketing, finance, operations, and strategy. A biology student needs to understand AI’s role in research, health care, data analysis, and ethics. A computer science student needs to understand not only how systems are built, but how they affect human beings.
3. Colleges should move from coverage to capability.
A course organized mainly around content coverage will be less distinctive in a world where AI can explain content quickly and repeatedly. Content remains essential, but the course must ask what students can do with it. Can students analyze? Can they create? Can they critique? Can they work with others? Can they make judgments under uncertainty? Can they use AI responsibly? Can they defend their reasoning? Can they apply knowledge in a context that matters?
4. Colleges should build more project-based and work-based learning into the student experience.
If tacit knowledge matters more in an AI era, students need more opportunities to acquire it before graduation. They need internships, undergraduate research, community-based projects, employer challenges, clinical experiences, design studios, capstones, and alumni-sponsored projects.
Strada’s work-based learning research is relevant here. Strada reports that 73 percent of graduates who completed a paid internship had a first job requiring their degree, compared with 44 percent of those who did not complete an internship. Strada also emphasizes that work-based learning provides learners with access to networks, skills development, and real-world work scenarios.
If internships and work-based experiences help students launch, then access to those experiences should not depend on family connections, unpaid summers, transportation, or a student’s ability to navigate the hidden curriculum. Colleges should make applied learning a designed part of the educational model.
5. Students need mentors and networks.
The Stanford paper’s codified-versus-tacit distinction reinforces the importance of human guidance. Tacit knowledge is often learned through apprenticeship, observation, feedback, and repeated practice. It is acquired in relationship with people who know what good work looks like.
That is why colleges should build mentoring infrastructure. Faculty mentors, staff advisers, peer mentors, alumni mentors, and employer partners all have roles to play. Students do not need a specific, perfect mentor. They need a constellation of people who help them interpret their experiences and prepare for what comes next.
The Transcript Is No Longer Enough
The Stanford paper also supports another premise I have been developing, notably, that the traditional transcript is inadequate for the AI era. The transcript tells us what courses a student completed, what grades were earned, and what credential was awarded. It remains important. But it does not show enough.
A transcript does not show whether the student can use AI responsibly. It does not show whether the student completed a project for a real client. It does not show whether the student can communicate with non-experts. It does not show whether the student can work in a team, revise in response to criticism, defend a judgment, or apply knowledge in an ambiguous situation.
If entry-level hiring becomes more competitive in AI-exposed occupations, new graduates will need better evidence. These graduates will need portfolios, project artifacts, internships, and research experiences. They will need capstones. They will need faculty and alumni who can speak credibly about their work. They will need evidence that they can do more than complete courses.
Colleges that help students build and explain that evidence will have stronger value propositions than colleges that send students into the labor market with only a transcript and a résumé assembled in senior year.
Why This Matters Especially for Private Colleges
Although the Stanford paper is about the labor market, it has special relevance for private colleges. Most private colleges charge more than public alternatives. They justify that premium through promises of personal attention, small classes, mentorship, community, alumni networks, and career preparation. In an AI era, those promises will be tested like never before.
Families will want to know: If AI can explain content, provide tutoring, improve writing, support coding, and help students prepare for job searches, then what are we paying the private college to do?
The answer cannot be vague. The answer must be specific. Students are known, challenged, and mentored. Students learn to use AI responsibly. Students work on real projects. Students build evidence of capability. Students connect with alumni and employers. Students learn to exercise judgment in situations where AI may provide an answer but cannot assume responsibility for it.
That is the community premium. The Stanford paper strengthens the argument for that premium by suggesting that the labor market may increasingly reward exactly what private colleges should be able to help students develop: tacit knowledge, applied capability, judgment, mentorship, networks, and evidence.
Having the potential to do something is not the same as executing that potential. Not all small colleges are mentoring colleges. All residential colleges are not close-knit communities. A low faculty-student ratio does not form relationships. Career centers are not career formation systems. Alumni databases are not networks. Private colleges must design these things deliberately.
The Risk of Waiting
Some college leaders may be tempted to wait for more definitive evidence. That cautionary instinct is understandable. The Stanford researchers emphasize that their analysis is descriptive and that more research is needed to determine how much of the observed pattern is attributable to AI rather than other forces.
Waiting for perfect evidence may be risky. Colleges do not need to know precisely how AI will reshape every occupation to know that students need AI fluency. Perfect causal proof isn’t necessary to know that students benefit from internships, applied projects, mentoring, and clearer career pathways. Labor market trends don’t prohibit colleges from preparing their graduates with capabilities beyond the transcript.
The prudent response to AI-driven changes is to ask whether the college is preparing students for a world in which AI is altering workforce expectations and actions.
Questions College Leaders Should Be Asking
The Stanford paper prompts a set of strategic questions for college leaders.
- Are we tracking how AI is changing the fields our students enter?
- Do we know which majors lead into occupations most exposed to AI?
- Are students in those majors receiving discipline-specific AI fluency?
- Are faculty being supported to redesign assignments and assessments?
- Are students completing projects that require them to use knowledge in real contexts?
- Are internships and work-based learning accessible to all students, not just the well-connected?
- Do students graduate with portfolios or evidence of capability?
- Are alumni helping students understand how work is changing?
- Are employers telling us what they now expect of entry-level talent?
- Are we treating career preparation as an institutional responsibility or leaving it to a career services office at the end?
- Can students explain what they can do, not just what they studied?
These questions should not be separate from the academic mission. They are central to it. If AI is changing the relationship between formal education, codified knowledge, and early-career work, then colleges must help students bridge the gap between knowledge and capability.
A Careful but Serious Warning
The researchers use the phrase “canaries in the coal mine” carefully. They are not saying the mine has collapsed. They are saying there may be early warning signs. The findings are a reason to:
- Take the college-to-career transition more seriously.
- Help students acquire tacit knowledge before graduation.
- Make AI fluency a humane and practical part of the curriculum.
- Connect students earlier and more intentionally with alumni, employers, research, projects, and mentors.
- Make student capability visible.
The old career contract was often viewed as “complete college, earn your degree, and the labor market will recognize your preparation.” That contract has been weakening for some time, and AI may be accelerating the change.
The new career contract is more demanding. Students need degrees, but they also need evidence of their competencies. They need knowledge, but also judgment. They need AI fluency, but also human fluency. They need credentials and networks. They need exposure and practice. They need mentors, not just advisers. They need projects, not just assignments.
The Future College Must Build the First Rung
The Stanford paper supports the premise that colleges should help students build the first rung of the career ladder before graduation. That shouldn’t reduce college to really expensive job training. I spent enough of my professional years in higher education to believe strongly that college must do more than prepare students for their first job. It should prepare them for citizenship, ethical judgment, meaningful work, continued learning, and lives of purpose and contribution. The larger mission is not served when students graduate unable to enter the kind of work where their education can grow.
If AI is reducing hiring for some entry-level roles, colleges need to help students graduate with more than potential. They need demonstrated competence, experience, references, AI fluency, confidence, people who can open doors, and ultimately, a body of work all their own.
The Stanford paper should lead colleges to design:
- Courses that build capability, not just coverage.
- Assessments that reveal judgment, not just output.
- AI fluency that is ethical, practical, and discipline-specific.
- Project marketplaces that connect students to real work.
- Mentoring systems that help students interpret experience.
- Alumni networks that function as opportunity networks.
- Dashboards that show whether students are known, challenged, and connected.
- Pricing and value propositions that families can understand and trust.
AI will continue to evolve. The labor market will continue to adjust. Some early signals will prove durable. Others may fade. But the direction of the educational response seems clear. Colleges should not become less human because AI is more capable. They should become more intentionally human, more experiential, more evidence-rich, and more connected to the world students are entering.
The Stanford paper issues a warning to colleges and provides an incentive to adapt. If young workers are the canaries, colleges should not merely listen for the alarm. They should change their curriculum to include AI fluency and experiences that will enhance their graduates’ chances of obtaining new entry-level jobs. Being the last college to change will not be rewarding for the college or its graduates.
