My first decent job – my first job where I wasn’t at risk of being attacked by dogs while delivering newspapers or of being scolded for scooping too big – was as a waiter at Oliver’s Bakery Restaurant, a North Toronto institution where my one of my older cousins had worked behind the bakery counter, selling the wonderful breads and cakes baked in the back. It was affirming to get hired and don the uniform: yellow button-down, the owner’s striped school tie, and an apron as white as the snow outside. The sights, sounds, and smells (freshly baked rolls, coffee) were exhilarating. But nothing topped the taste of the place: before shifts when we tried the daily specials, and after, if there was – say – just one piece left of the incomparable Café Royale Truffle Torte.
I learned a lot at Oliver’s. I learned that every Saturday at 7 a.m., blue haired Lois would sit at her favorite table and expect a cup of freshly brewed coffee + blueberry muffin straight from the oven; that I should smile and take it as a compliment when an elderly customer gave me a once-over and said in a creepy way that I had really nice skin, like she wanted it for herself; and that when Torontonians ordered, they were polite, saying things like “could I please have the capellini” or “I’d like the capellini.” In my years as a server, I don’t think I ever heard a customer order by saying: “I’ll do the capellini.”
Walk around any restaurant these days and you’ll hear it. “I’ll do the filet,” “I’ll do the Cobb salad.” I heard it last week walking by an airport Subway:
Subway Guy: I’ll do the six-inch tuna with spinach.
Then, two days later, in a Chipotle in Midtown Manhattan:
Chipotle Guy: [pointing at unidentified meat] I’ll do that.
And as Chipotle Guy shuffled down to the cashier and ordered a beverage:
Chipotle Guy: I’ll do a large.
When did ordering food or drink shift from the modest “I’d like” or “I’ll have” to the braggadocious “I’ll do”? What exactly are Subway Guy and Chipotle Guy doing? They’re asking someone to make them lunch. Then they’re going to eat it. They’re not doing anything.
The “I’ll do” epidemic is a symptom of the dumbing down of work. The threshold for what constitutes work seems to have gotten lower. We’re seeing it in education. While grades go up, students are being asked for less and less. In K-12, they’re no longer asked to read entire books. A report from AEI estimates the average full-time college student only learns for about 25 hours per week. More important at the dawn of AI, learning is not the same as doing, at least not most of the time. And that’s about to become a serious problem.
***
At Oliver’s, we had sauces upon request: a neat caddy of ketchup, mustard, vinegar, Tabasco, Worcestershire for oysters, and A1. But there was little risk of confusing A1, a tangy steak sauce, with AI, as new Secretary of Education Linda McMahon did at last month’s ASU-GSV conference in San Diego, a gaffe that launched a thousand steak sauce memes.
While A1 isn’t impacting education outside of restaurant training programs, AI is all everyone wants to talk about. There are already dozens of newsletters bringing educators up to speed on how to use it to help students learn. And while everyone is still experimenting, the good news is we’ve already come up with myriad use cases to accelerate learning. Students are using AI to organize their time, prioritize topics for study, and then produce explanations, visualizations, opposing viewpoints, flash cards, practice quizzes, games, and podcasts to improve understanding and facilitate mastery. Meanwhile instructors are using AI to generate lesson plans, create presentations, and build assessments. Some are already using it to provide feedback on student writing.
At the same time, it seems like every exciting innovation is counterbalanced by fear that less learning is occurring as a result of shortcuts or downright cheating: asking for summaries of lengthy reading assignments; breaking down STEM problems and copying the results; prompting AI to create first drafts of written assignments rather than struggling with concepts and structure. The most widely shared article from New York Magazine last week was Everyone Is Cheating Their Way Through College.
But it couldn’t be clearer that today’s students will graduate into a world that expects AI experience. Something like 80% of companies are already deploying AI and about 75% of knowledge workers are using it on the job. More and more entry-level jobs descriptions mention AI skills.
So schools are understandably torn. Should they ban the use of AI or partner with OpenAI or Anthropic? Will students be workforce-ready if they’re not encouraged or even allowed to use ChatGPT? But what if this is a false dichotomy? What if the great AI-in-school debate is a distraction from a more fundamental problem?
As my British friend Tom Bewick noted recently in his Skills Agenda newsletter:
Our children still learn in classrooms, many built during Queen Victoria’s reign. Like a Fordist production line, pupils are enrolled and taught in batches… Teachers supervise this knowledge-production process on the shop floor… Students leave school judged on what they know and, notably, on how well they can retrieve that knowledge under high-stakes examination conditions: no technology, only pen and paper. In other words, an educator from the late 1800s who travelled through our modern-day school and college systems in a time machine would find great comfort in this familiarity.
Despite the fact that lectures and seminars are less effective at producing learning outcomes than flipped classrooms, active learning, and project-based learning, chalk-and-talk remains the dominant form of knowledge dissemination through college. There are various reasons for this, but mostly because it’s easier and cheaper. And in the context of this dominant mode, AI literacy connotes knowing how to use AI to master content and concepts more effectively. In other words, a student who’s used AI to become a better student.
As much as those of us who’ve built careers in education might wish otherwise, the world of work is far different from school. In school, students serve themselves: their intellectual development, and according to AEI, a healthy dose of leisure. At work, we serve others. We produce work for peers, supervisors, and ultimately customers or constituents. And so the great AI debate overlooks the difference between knowing how to use AI to learn and knowing how to use AI to do productive things. To be sure, there’s an intersection e.g., brainstorming, leveraging AI to research and write an original paper. But the intersection is small in the context of the work employers need done and the AI skills they’re seeking.
Employers aren’t going to be satisfied that a high school or college graduate knows how to use AI to write a paper that no one wants to read. That’s better than nothing and a damn sight better than confusing AI with steak sauce. But what they really need are candidates who know how to use AI to do a better job, or a more efficient job, or – best of all – redefine a job. And using AI to become a better learner or produce a paper is scant evidence of that. So saying that a new grad – who’s used AI to successfully complete 120 credit hours of lectures and seminars – is AI literate is like the Subway guy saying “I’ll do the tuna.” He hasn’t used AI to actually do anything.
***
The solution has to be – finally, urgently – a shift to more effective learning modalities. Project-based learning becomes a necessity. Faculty must rethink assignments and assessment. In STEM this means asking students to build things. For the social sciences: campaigns, mapping projects, cobbling together budgets and audits. For arts and humanities: productions (live or digital) or simulations. The most charming education article I’ve read in a while was last week’s New York Times piece on University of Chicago’s simulation of the papal conclave of 1492. This year’s resulted in the election of a Jewish pope, and at the time, the only pope, leading to a mic drop of a closing line: “As far as I know, I’m the only person in the world claiming to be the pope right now,” Mr. Kind said. “I think that technically makes me pope.” Modeling scenarios with AI to optimally manage complex negotiations and internecine feuds is a skill that clearly transfers from the Renaissance.
But none of these involve learning to use AI to do work for an actual employer. Which will be an order of magnitude more complex than the most involved school project. Because when today’s high schoolers enter the workforce, they won’t be asked to prompt a chatbot. They’ll be human copilots to intelligent agents. Employers will want candidates who’ve worked with agents, understand business processes, and know how to prompt, tune, and challenge. It’s a role that will require analytical skills, orchestration skills, and AI product experience, very little of which students will get unless they’ve already been exposed to agentic AI through work-based learning.
So perhaps schools and colleges have been looking at the AI problem all wrong. Focusing on how AI can or should be used to help students learn is a trivial problem in comparison to solving for how to give students experience working with AI to actually do things.
The latter challenge is easier to understand but harder to solve. Because schools aren’t set up to engage employers and connect students with them. Witness the many limitations of career services and the growing internship gap. And without a willing employer on the other side – an employer that has invested in agentic AI to power its business processes, and that’s now willing to employ unproductive workers (aka students) at some level (contradictory impulses, at least on the surface) – students stand little chance of becoming AI literate in the way that matters.
***
As work has been dumbed down – as “I’d like” has been replaced by “I’ll do” – young Americans are working less in and out of school. High school students aren’t scooping ice cream or clearing tables as often as they did a generation ago; they see one pathway to economic opportunity (college) and mistakenly believe admissions offices place less value on paid work than expensive summer programs on the campuses of selective universities and service learning trips to exotic locales. And while about 40% of full-time college students (and nearly 80% of part-time students) support themselves through degree programs by working, only a small percentage have career-relevant jobs.
At a time when students need more exposure to work than ever – when exposure to AI to do things will become a prerequisite for career launch – there’s already less work than there used to be. So if we don’t want the college graduate underemployment rate to skyrocket from the already unacceptable level of 52% – and the college graduate unemployment rate to continue to creep up from a three-decade high – our education system must take three concrete steps:
1) Accelerate learning in high school
Effective project- and work-based learning require prior content and concept mastery. In order for postsecondary education to effect this tectonic shift, middle and high schools need to accelerate learning so students are ready to get to work by the time they reach college. The good news is we’ve got just the technology to do it: the AI transforming the labor market and necessitating acceleration in the first place. This will be a major change and students will probably need to specialize earlier in order to gain sufficient capabilities in their field of study. They’ll also need to become better writers by the time they enter college: writing (and with immediate feedback) rewriting much more. But the most innovative high schools will deploy AI flashcards, podcasts, and a thousand other tools to lead the way.
2) Faculty shift to project-based learning
Colleges need to reorganize programs of study and courses around project-based learning. Replace tests, exams, and purposeless papers with projects where students will need to use AI to produce meaningful work product. More hack-a-thons, datathons, design-a-thons – any a-thon will do as long as it yields a tangible result.
3) Colleges become work-based learning institutions
While faculty and department heads are busy effecting a project-based learning revolution, administrators should be tasked with work-based learning transformation i.e., ensuring every student is employed in at least one career-relevant job before graduating into a full-time job search. The simplest path is to become a co-op school. But because Bank of America isn’t going to launch co-op programs with 300 different universities, it won’t happen through the efforts of university leaders alone. For most schools, this many-to-many problem will only be solved by the emergence of an ecosystem of intermediaries –
co-op or internship service providers – aggregating both supply of and demand for work-based learning.
At the same time, colleges can get the ball rolling by employing as many students as possible themselves. Not solely as the custodians and dining hall workers funded by the Bizarro incentives of the Federal Work-Study program. But in new jobs that improve the institution and surrounding community. Universities are partnering with companies like Saxbys to open coffee shops and bookstores managed entirely by students for the purpose of providing work-based learning opportunities. Education at Work operates help desks on university campuses. Last year I wrote about University of Iowa’s School of Journalism buying community newspapers for students to work on. The hottest colleges of the next decade will be those that think creatively about dramatically increasing the supply of work-based learning.
Thanks to AI, priority A1 for the entire education sector should be an all-of-the-above strategy to ensure that the first decent jobs our children get – e.g., serving coffee and capellini – won’t be their only jobs. With the salutary side of understanding what real work is so they’ll avoid saying “I’ll do” when ordering lunch.