'Students who use AI as a crutch don't learn anything'

Framing: “Crutch” vs. “Co‑intelligence”

• Several commenters say the headline overemphasizes “AI as a crutch”; the interview itself stresses using AI as “co‑intelligence” that keeps humans in the loop.
• Others think the title accurately reflects the danger: when AI replaces thinking end‑to‑end, students skip learning altogether.

Comparisons to Calculators and Earlier Tech

• Repeated analogy: “people said the same about calculators / Google / writing.”
• Counter‑points: calculator outputs are verifiable, operate at the level of single steps, and don’t usually replace conceptual understanding; LLMs can produce whole solutions with non‑obvious errors.
• Some argue every new abstraction shifts what’s considered “basic” and frees time for more advanced topics; others say this erodes essential numeracy and mental models.

Learning, Practice, and Education

• Distinction between “busy work” and “practice”: repetitive tasks can be essential for building fluency (e.g., arithmetic drills).
• Concern: students using AI to do homework or code assignments may “skip the struggle” that forms mental models, leading to shallow understanding and weaker future performance.
• Example of constructive use: a child writes essays herself, then uses AI for feedback and rewrites without direct copying—seen as a powerful complement.
• Debate over which skills truly need deep mastery (e.g., long division, trig) vs. which can be offloaded.

Experiences Using AI for Coding and Projects

• Some describe AI as transformative: quickly shipping small apps or scripts they would otherwise never finish; it boosts confidence to explore new stacks.
• Others admit they “learned almost nothing” in the process—AI handled nearly all code and debugging.
• Senior developers worry juniors already struggle with fundamentals and now can hide behind AI output they cannot explain.

Code Quality and Productivity

• One cited study claims AI‑assisted code had more bugs and no net efficiency gain; others question methodology and suggest benefits depend on skill and use‑case.
• Many report AI works well for boilerplate, small scripts, or well‑trodden APIs, but breaks down on novel, complex, or niche tasks and can thrash when fixing its own bugs.

Cognitive and Societal Effects

• Some foresee atrophy of reasoning, dependence, impatience, and reduced ability to validate results—analogous to over‑reliance on GPS or card payments.
• Others argue humans continually “move up the stack”; the real risk is not using powerful tools at all, but using them uncritically without preserving core understanding.