The Course is Dead (But L&D Hasn’t Gotten the Memo)

If we can capture what the very, very best tutors do in a software agent, then education will have changed in a super deep way.

Lecture is the transfer of the notes of the lecturer to the notes of the student without passing through either.

The course is dead.

Not dying. Not transforming. Dead.

It just doesn’t know it yet — and neither do most learning professionals.

Here’s the blunt truth: Within five years, the entire infrastructure we’ve built around courses — the LMS, the instructional design models, the completion certificates, the learning paths — will seem as quaint as a rotary phone.

AI tutors are about to make the fundamental unit of learning — the course — as obsolete as that erstwhile fundamental unit of long-distance communication, the telegram.

If you design courses for a living, this should either terrify or excite you, depending on whether you’re ready to let go of the past.

§1. How we got here: the course as knowledge-delivery system

To understand why courses are doomed, let’s think about why they exist.

In 1763, Frederick the Great of Prussia mandated compulsory schooling. What emerged was breathtakingly efficient: Put 30–50 students in a room with one teacher. Have them sit in rows. Ring bells to change subjects. Test periodically. Sort by age, not ability. It was an assembly line for producing standardized citizens.

By the early 20th century, virtually every developed nation had adopted this model. As educational historian David Tyack notes in “The One Best System,” the grammar of schooling — the student-teacher ratio, the batch processing, the lecture-centric model — became so entrenched that we forgot these were design choices, not natural laws.

This model succeeded brilliantly for its intended purpose. It created literate citizens who could function in industrial society. We should appreciate it for what it was: a brute-force solution that was appropriate given the constraints of its era.

The problem? We’re not living in that era anymore.

§2. The scarcity assumption

The course model assumes teachers and knowledge are scarce, students plentiful. This equation shaped everything:

• Courses exist because one teacher must deliver content to many students efficiently.
• Lectures exist because it’s the most economical way to reach a cohort at once.
• Textbooks exist to standardize content and get individual teachers up to speed on it quickly.
• Multiple-choice tests exist because individual skill assessment doesn’t scale.

Larry Cuban’s research in “Teachers and Machines” documented how each new technology of the 20th century — film, television, computers — was heralded as revolutionary but ended up, at best, delivering the same learning model more efficiently. E-learning courses are usually just recorded lectures over slides.

Seymour Papert called this “technocentric thinking” — applying new technology in service of existing activities rather than reimagining them. Roger Schank noted that the first movies were recordings of stage plays. There’s always an imagination lag.

But the course model’s stability goes deeper. None of the previous revolutions solved the fundamental problem: teacher scarcity. You still needed a human for real interaction, and everything followed inexorably from there.

Until now.

§3. Back to the future: the learning model we left behind

Before mass education, humans learned through apprenticeship. A novice worked alongside a master, learning by doing, receiving immediate feedback, progressing at their own pace. It was personalized, practical, and profoundly effective.

We didn’t abandon apprenticeship because it didn’t work — it worked brilliantly. We abandoned it because it didn’t scale.

§4. Why apprenticeship worked: the cognitive science

Modern cognitive science has validated what medieval masters knew intuitively. Apprenticeship succeeded because it aligned perfectly with how humans actually learn. Three principles in particular made it work.

1. We learn by doing — there is no other route to mastery

Aristotle distinguished between theoretical knowledge (episteme) and practical knowledge (phronesis) and noted that the latter comes only through practice.

John Dewey refined this: “We learn from reflecting on experience.” But for reflection to occur there must first be experience. Roger Schank, the AI pioneer turned learning scientist, put it more bluntly: “Humans learn by doing. Passive learning is an oxymoron.”

This isn’t just philosophy — it’s backed up by psychology and neuroscience. You can’t learn to ride a bike by studying angular momentum or become a surgeon by memorizing anatomy. The apprentice working actual metal in the forge learns what the student taking notes about metallurgy never will.

2. The selection and sequencing of practice is everything

Medieval masters possessed a striking ability to assign exactly the right task to the novice at the right time. This wasn’t mere intuition — it was due to many generations of accumulated wisdom about skill progression. Modern research confirms what these early mentors knew intuitively. Vygotsky’s zone of proximal development shows learning happens at the edge of current ability; Ericsson’s deliberate practice research reveals that random practice is worthless, and only targeted practice at specific weaknesses improves performance; cognitive load theory shows that overwhelming novices destroys learning.

The master’s curatorial genius — selecting precisely the right challenge, not too easy, not too hard, targeting exactly what needs work — made apprenticeship transformative. It’s an ingredient that most modern training desperately lacks.

3. Immediate feedback is the oxygen of learning

In the workshop, every hammer stroke instantly revealed success or failure. But it took a master’s eye to provide insight — not just “that’s wrong” but “here’s why and how to fix it.” Without this feedback loop, practice doesn’t make perfect — it may just lead to confusion or giving up.

As Ericsson’s research shows, it’s not 10,000 hours that create mastery — it’s 10,000 hours with immediate, focused feedback. Patricia Cross captured it perfectly: masters provided “minute-by-minute and day-by-day feedback about performance.” Not quarterly reviews. Not end-of-course evaluations. Minute-by-minute. Because in-the-moment correction is what drives real learning.

The medieval workshop worked as a training ground because it was built on how humans actually learn. We have the science now to prove what those masters knew intuitively. The question is whether we’ll use it.

We didn’t have a choice to scale this. Until now.

§5. Enter the universal expert tutor

Here’s what changes everything: AI can now provide personalized tutoring to every learner, playing the expert role in the expert-novice relationship. The scarcity equation that created courses has shattered.

Benjamin Bloom’s 1984 research showed that one-on-one tutoring improves performance by two standard deviations — moving average students to 98th percentile performance. Bloom called this the 2 Sigma “problem” — rather than, say, the 2 Sigma “solution” — because he couldn’t scale it.

What Bloom couldn’t foresee was Generative AI. Gen-AI based tutors have already shown great promise in studies at Harvard and Stanford — and they are improving rapidly. As Koedinger demonstrated, the key tutoring factors — like immediate feedback, personalized pacing, and targeted practice — are algorithmically codifiable.

Let us paint you a picture. It’s 2027. Sarah, a sales professional, just lost a deal. In the old world, she’d find a course: “Advanced Objection Handling.” She’d get generic content created for anonymous learners, with no knowledge of her situation.

In the new world:

Sarah tells her AI companion during her commute: “I choked when the client said ‘your price is 30% higher than competitors.’”

Her AI responds:

“Let’s work on that. Tell me the context. What industry? What product? What was the competitor offering?”

Over 20 minutes, the AI:

• Creates a simulation of that exact scenario.
• Plays her prospect with uncanny accuracy.
• Lets Sarah practice different approaches.
• Provides immediate feedback on tone, pacing, arguments.
• Adjusts difficulty based on performance.
• Draws on thousands of successful sales interactions.
• Remembers everything for future sessions.

No course. No modules. No certificates. Just continuous, personalized practice with expert feedback exactly when needed.

By 2030, this won’t seem revolutionary. It will seem obvious.

§6. The resistance

When we present this vision, we see a predictable pattern.

First, eyes light up. Learning professionals get the possibilities.

But then implications hit. You see the shift: “If we don’t need courses… what happens to instructional design? To our LMS? To my ADDIE slides?”

Then comes the motivated reasoning: “AI will enhance what we do. Help create courses faster. But people will always need structured paths.”

As Upton Sinclair observed: “It is difficult to get a man to understand something when his salary depends on his not understanding it.”

§7. The counter-revolutionary illusion

The need to see AI as additive rather than transformative is well-documented. Kahneman and Tversky’s work shows we’re wired to see change as incremental. “Anchoring bias” makes us adjust from what we know rather than imagine something different.

Christensen’s research predicts exactly what we’re seeing. Organizations will first use AI to:

• Create courses faster.
• Generate assessments automatically.
• Provide chatbot support.
• Personalize existing paths.

This “sustaining innovation” phase always precedes “disruptive innovation,” where technology enables entirely new models that obsolete the old. We saw this with:

• Newspapers using internet for faster delivery (before Google destroyed their model).
• Taxis using apps for dispatch (before Uber reimagined everything).
• Blockbuster adding websites (before Netflix reimagined delivery).

§8. What abundance-based learning looks like

A master’s degree in 2035

Jennifer wants to transition from marketing to data science. Old world: Apply to universities, wait for acceptance, study nights or quit, follow fixed curriculum, emerge with degree and $80,000 debt.

In 2035: She tells her AI her goal. It assesses her skills through real problems, not tests. It identifies her analytical strengths from marketing but need for programming, statistics, ML expertise.

Her “curriculum” is increasingly complex real-world projects. She analyzes her company’s marketing data using Python. The AI provides just-in-time instruction. When she struggles with regression, it doesn’t assign Chapter 12 — it creates visual, interactive explanations tailored to her background.

Six months: contributing to open-source projects. Eight months: freelance data science. Ten months: hired as junior data scientist.

Time: 10 months while working. Cost: Far less than $80,000. Relevance: 100%.

§9. Corporate training in 2032

Global Corp needs to upskill 10,000 employees in negotiation. In the old world, we’d commission a course — probably an e-learning lecture course — mandate completion, and hope for the best.

In 2032, each employee’s AI will understand their specific role and creates relevant scenarios:

Procurement: “Your supplier announced a 15% increase. Practice negotiating while maintaining the relationship.”

Sales: “You’re at 95% quota with one day left. That client wants a 30% discount. Handle this.”

HR: “A top performer wants a 40% raise to match a competing offer. Your budget maxes out at 15%.”

Learning happens in the workflow. There is no course to complete — just continuous skill building integrated organically into daily activities.

§10. The learning science that survives

While course-focused instructional design dies, learning science becomes MORE relevant:

Zone of Proximal Development (Vygotsky). AI precisely identifies each learner’s ZPD, adjusting in real-time.

Deliberate Practice (Ericsson). AI structures practice with immediate feedback, repetition, progressive difficulty — personalized to current skill.

Cognitive Apprenticeship (Collins, Brown & Newman). AI makes expert thinking visible, provides coaching, scaffolds tasks, fades support.

Desirable Difficulties (Bjork & Bjork). AI introduces precisely calibrated challenges for long-term retention.

Spaced Repetition (Ebbinghaus). AI optimizes review for each learner’s unique forgetting curve.

Flow Theory (Csikszentmihalyi). AI continuously calibrates challenge-skill balance.

These principles will be implemented by AI in real-time, personalized for each learner, not baked into static courses.

§11. The new role of learning professionals

What happens to learning professionals when courses disappear? The truly human parts become MORE important:

Learning Experience Architects design parameters for AI-created experiences. You’re designing possibility spaces, not courses.

Performance Consultants identify what actually improves performance versus what people think they need.

Human Connection Facilitators create communities and mentorships AI cannot.

Ethical Guards ensure AI teaches responsibly.

The shift isn’t human-to-AI. It’s human-doing-repetitive-tasks to human-doing-uniquely-human-tasks.

§12. A personal note

We’ve spent careers trying to create better learning within harsh pragmatic constraints. Every innovation felt like making the horse-drawn carriage faster.

Now we see the metaphorical equivalent of the automobile. Not a faster carriage — fundamentally different transportation.

Will it be disruptive? Absolutely.

Will some be left behind?

Unfortunately, yes.

But will it create a world where every human can learn anything they need, when they need it, perfectly adapted to their unique mind?

Yes. And that’s a future worth building.

The course is dead.

Long live learning.

§13. References

Barsalou, L. W. (2008). Grounded cognition. Annual Review of Psychology, 59, 617–645.

Bjork, R. A., & Bjork, E. L. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In M. A. Gernsbacher, R. W. Pew, L. M. Hough, & J. R. Pomerantz (Eds.), Psychology and the real world: Essays illustrating fundamental contributions to society (pp. 56–64). Worth Publishers.

Bloom, B. S. (1984). The 2 sigma problem: The search for methods of group instruction as effective as one-to-one tutoring. Educational Researcher, 13(6), 4–16.

Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132(3), 354–380.

Chi, M. T., Bassok, M., Lewis, M. W., Reimann, P., & Glaser, R. (1989). Self-explanations: How students study and use examples in learning to solve problems. Cognitive Science, 13(2), 145–182.

Christensen, C. M., & Raynor, M. E. (2003). The innovator’s solution: Creating and sustaining successful growth. Harvard Business School Press.

Clark, R. C., & Mayer, R. E. (2023). e-Learning and the science of instruction: Proven guidelines for consumers and designers of multimedia learning (5th ed.). Wiley.

Collins, A., Brown, J. S., & Newman, S. E. (1989). Cognitive apprenticeship: Teaching the crafts of reading, writing, and mathematics. In L. B. Resnick (Ed.), Knowing, learning, and instruction: Essays in honor of Robert Glaser (pp. 453–494). Lawrence Erlbaum Associates.

Cross, K. P. (1999). Learning is about making connections. The Cross Papers, 3. League for Innovation in the Community College.

Csikszentmihalyi, M. (1990). Flow: The psychology of optimal experience. Harper & Row.

Cuban, L. (1986). Teachers and machines: The classroom use of technology since 1920. Teachers College Press.

Dewey, J. (1916). Democracy and education: An introduction to the philosophy of education. Macmillan.

Dunlosky, J., & Metcalfe, J. (2009). Metacognition. Sage Publications.

Ebbinghaus, H. (1885/2013). Memory: A contribution to experimental psychology. Annals of Neurosciences, 20(4), 155–156.

Ericsson, K. A., Krampe, R. T., & Tesch-Römer, C. (1993). The role of deliberate practice in the acquisition of expert performance. Psychological Review, 100(3), 363–406.

Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive-developmental inquiry. American Psychologist, 34(10), 906–911.

Gates, B. (2023, April). Remarks at the ASU+GSV Summit, San Diego, CA.

Henderson, R. M., & Clark, K. B. (1990). Architectural innovation: The reconfiguration of existing product technologies and the failure of established firms. Administrative Science Quarterly, 35(1), 9–30.

Immordino-Yang, M. H., & Damasio, A. (2007). We feel, therefore we learn: The relevance of affective and social neuroscience to education. Mind, Brain, and Education, 1(1), 3–10.

Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263–291.

Kalyuga, S. (2007). Expertise reversal effect and its implications for learner-tailored instruction. Educational Psychology Review, 19(4), 509–539.

Keith, N., & Frese, M. (2008). Effectiveness of error management training: A meta-analysis. Journal of Applied Psychology, 93(1), 59–69.

Kestin, G., et al. (2025). AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15, article 17458. https://doi.org/10.1038/s41598-025-97652-6

Koedinger, K. R., Corbett, A. C., & Perfetti, C. (2012). The Knowledge-Learning-Instruction framework: Bridging the science-practice chasm to enhance robust student learning. Cognitive Science, 36(5), 757–798.

Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. Cambridge University Press.

Marton, F., & Tsui, A. B. (2004). Classroom discourse and the space of learning. Lawrence Erlbaum Associates.

Mazur, E. (2009). Farewell, lecture? Science, 323(5910), 50–51.

Papert, S. (1987). Computer criticism vs. technocentric thinking. Educational Researcher, 16(1), 22–30.

Pekrun, R. (2006). The control-value theory of achievement emotions: Assumptions, corollaries, and implications for educational research and practice. Educational Psychology Review, 18(4), 315–341.

Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.

Schank, R. C. (1995). Tell me a story: Narrative and intelligence. Northwestern University Press.

Sinclair, U. (1935). I, candidate for governor: And how I got licked. End Poverty League, Inc.

Sweller, J., Ayres, P., & Kalyuga, S. (2011). Cognitive load theory. Springer.

Tyack, D. B. (1974). The one best system: A history of American urban education. Harvard University Press.

VanLehn, K. (2011). The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systems. Educational Psychologist, 46(4), 197–221.

Vygotsky, L. S. (1980). Mind in society: The development of higher psychological processes. Harvard University Press.

Willingham, D. T. (2021). Why don’t students like school? A cognitive scientist answers questions about how the mind works and what it means for the classroom (2nd ed.). Jossey-Bass.