By ALAN LESGOLD and JOHN RADZILOWICZ*

The workplace transformation is unmistakable: Organizations are rapidly adopting AI systems, with 23% already scaling agentic AI somewhere in their enterprises and 78% of organizations reporting AI use in 2024, up from 55% the year before. Professionals increasingly manage AI systems rather than complete tasks themselves. Our graduates will need to “conduct” AI tools the way a maestro directs an orchestra. But here’s the problem: You can’t conduct an orchestra if you’ve never learned to play an instrument.

This creates a dilemma for universities. Many students arrive on campus without mastering the foundational skills of expository writing, mathematical reasoning, critical reading that they should have developed earlier. We must simultaneously build these foundations while preparing students for AI-integrated careers. The question isn’t whether to use AI in education, but when and how to do so without undermining the cognitive development our students desperately need.

The conductor’s paradox

A symphony conductor must deeply understand music’s structures before effectively directing musicians. Similarly, students must develop core abilities through direct practice before they can orchestrate AI tools effectively.

When students use AI to skip foundational skill-building, they become passive consumers rather than critical thinkers. Without a knowledge base, using AI to solve complex problems is impossible. Practicing a skill isn’t just about producing a product. It develops the mental structures needed for advanced thought. Outsourcing this cognitive work to AI is like sending a proxy to the gym and expecting to get fit.

What students must practice directly

Certain competencies cannot be delegated to AI without serious consequences:

• Writing: The act of writing clarifies thinking. Students who use AI to generate text bypass the cognitive work that develops their ability to organize ideas, construct arguments, and revise thinking. They’ll never gain the judgment to evaluate whether AI-generated arguments are logical or persuasive.

• Mathematical reasoning: Working through quantitative problems builds logical and systematic thinking. Students who let AI solve homework never develop the capacity to recognize when answers are nonsensical or to clearly specify problems they want AI to solve. That capacity is essential for any field requiring quantitative judgment.

• Critical reading: Students must develop the ability to engage complex texts, identify arguments and synthesize sources. Those who always rely on AI summaries never build cognitive stamina for independent scholarship or the ability to recognize when analyses are superficial.

• Research skills: Formulating hypotheses, designing studies, and interpreting data teaches scientific literacy and critical thinking. Students who outsource these processes can’t evaluate research quality or understand the limitations of evidence.

The human advantage

As AI handles complex technical tasks, uniquely human skills become professionally essential—no longer “soft skills” but core competencies.

• Ethical judgment in gray areas where algorithms fail, weighing competing interests, making responsible decisions about AI use and limitations, understanding algorithmic bias and data privacy.

• Empathy and emotional intelligence for building trust, navigating conflict, motivating teams, and understanding nuanced human needs that AI cannot replicate.

• Creativity and visionary thinking to synthesize diverse inputs, imagine novel solutions and ask generative questions beyond existing frameworks.

• Strategic problem-solving to break down challenges, identify root causes, question assumptions and design multi-step solutions — essential for effective AI orchestration.

A developmental approach

Universities should think strategically about progression across the undergraduate experience:

• Foundational courses: For students with skill gaps, early courses must prioritize direct practice with AI use managed to not dilute practice opportunities for basic competences. This builds the cognitive architecture for everything that follows. Students learn when to do basic cognitive work and when to build upon it using AI.

• Intermediate courses: Once students show foundational competence, expand AI use explicitly. Teach AI literacy — how to prompt effectively, validate outputs, recognize limitations. Students learn to think with AI while maintaining intellectual agency.

• Advanced work: In upper-division courses, students with strong foundations use AI as professionals do—to amplify capabilities and tackle greater complexity. This requires the judgment built in earlier coursework.

Key questions for faculty

When deciding whether to integrate AI tools, ask yourself:

• Have my students mastered the foundational skills this assignment exercises? If not, they need direct practice. Using AI to mask deficiencies prevents necessary cognitive development.

• Where does the learning happen — in the process or the product? When learning occurs through the struggle of constructing arguments or working through problems, AI shortcuts undermine the entire pedagogical purpose.

• Can my students critically evaluate AI outputs? Students cannot judge what they don’t understand. If they lack foundational knowledge to critique AI, allowing AI use results in work they cannot defend or understand.

• Am I teaching AI orchestration explicitly? If you allow AI use, teach students how to prompt effectively, validate outputs, recognize limitations, and use AI ethically. AI literacy must be taught deliberately.

• What human capabilities does this develop? Design assignments that create opportunities to exercise judgment in ambiguous situations, navigate ethical dilemmas, collaborate with diverse perspectives, and communicate with real stakeholders.

The bottom line

Students who skip foundational skill-building by using AI helpers will struggle throughout their careers. They’ll be unable to think independently, evaluate information critically, or communicate effectively—regardless of AI sophistication. They’ll be commanded by AI rather than commanding it.

Our job is to decide when students need direct practice to build cognitive capacity, when they’re ready for AI orchestration, and when they should use AI as professionals do. This requires understanding where students are developmentally and being willing to require foundational work even when it’s difficult.

The goal isn’t to teach skills despite AI, but to develop the deep mastery students need to wield it effectively. Students with strong foundations can use AI to amplify capabilities and tackle greater problems. Those who skip developmental work remain dependent on tools they cannot evaluate or control.

The choices we make about when and how to integrate AI will determine which kind of graduates we produce. Let’s choose wisely.

Alan Lesgold is professor emeritus of education, psychology, and intelligent system, and former dean of the School of Education. John Radzilowicz is director of Pedagogy, Practice, & Assessment in the University Center for Teaching and Learning.

* We also used Manus and Claude to refine this article after the original concept was developed, and Gayle Rogers provided helpful critique of an earlier version.