We May Be Raising the First Generation That Can Produce Brilliant Work Without Knowing How to Do It
The Work Is Better. Is the Student?
Essays are cleaner. Research is faster. Math solutions arrive instantly. Presentations look professional.
The work is getting better.
But is the student?
Pew found that 54% of US teens have already used AI chatbots for schoolwork, and one in ten say they use them for all or most of it.
Common Sense Media found that 70% of teens use AI for schoolwork. Among users, 63% have used it to obtain answers directly. More unsettlingly, 38% say having AI available leads them to generate fewer ideas of their own, and 39% say they sometimes feel they are missing out on learning when they use it to finish assignments.
The students can already feel the tradeoff.
We may be measuring the quality of the artifact while losing sight of the quality of the mind that produced it.
This Is Not About Cheating
If we frame this solely as a cheating problem, we miss the bigger picture. The real problem exists even when schools explicitly allow AI.
A student can use AI ethically and still outsource: forming the argument, finding the evidence, structuring the reasoning, solving the problem, editing the weaknesses, and generating alternatives.
The real danger is not that students will cheat.
It is that they may complete the education without performing enough of the cognition the education was supposed to develop.
We Are Confusing Output With Learning
This is the central distinction.
Output: the essay, answer, presentation, code.
Learning: the cognitive change that occurred while producing it.
An assignment historically did two jobs: produce something and change the person producing it. AI can separate those.
AI makes it possible to separate the quality of the product from the capability of the producer at unprecedented scale.
Homework Was Never Really About the Homework
Why make a child write an essay when nobody needs another student essay? Why make them solve equations when calculators exist? Why make them research something Google already knows?
Because the artifact was never the real product.
The student was.
The struggle produced: reasoning, patience, retrieval, pattern recognition, argumentation, judgment, and frustration tolerance.
Education has always hidden cognitive training inside apparently inefficient work. AI can remove the inefficiency. It may also remove the training.
The Calculator Argument Is Not Enough
Some will argue, "We said the same thing about calculators." This is true. Calculators can improve learning when introduced appropriately. And current evidence is not uniformly negative: a 2026 meta-analysis found AI systems can improve K-12 cognitive outcomes, including higher-order thinking, depending on how they are used, grade level, subject, and implementation.
The answer is probably not to keep AI out of education. The evidence suggests AI can improve K-12 cognitive outcomes, including higher-order thinking, when implemented well. The problem is deciding where assistance becomes substitution.
Teach Both. Test Both.
New York City has imposed a one-year restriction on AI use for roughly 600,000 elementary and middle-school students, explicitly citing independent problem-solving and human interaction.
Daniel Susskind, who studies the impact of AI, argues for a useful principle: teach both, test both.
Teach with AI. Test without it. Then you can see what actually transferred to the student.
You cannot know whether AI augmented the student unless you occasionally remove the augmentation.
The Autopilot Test
Nobody objects to pilots using autopilot. We object to pilots who never learned to fly.
The same applies to AI. Students should absolutely learn to use it. But before they become dependent on it, they should demonstrate the underlying capability.
AI literacy without independent literacy is not literacy. It is dependency.
The Most Dangerous Student May Look Exceptional
Imagine a student producing university-level writing, polished analysis, sophisticated code, or elegant presentations. They look advanced.
But ask them to: defend the reasoning, reproduce the argument, explain why a source matters, identify an error, or solve a nearby problem without AI, and the capability disappears.
We may create a generation whose work signals competence they have not yet acquired.
Employers Inherit the Problem
Those students become analysts, doctors, lawyers, engineers, executives, and researchers. The employer sees the work product. But what capability exists underneath it?
If AI does the schoolwork, and later does the junior work, where exactly does expertise get created?
The Real Risk: Judgment Without Formation
We talk about knowledge. But senior judgment comes from repeated exposure to mistakes, ambiguity, failure, revision, and consequences. If AI removes those formative experiences, people may inherit authority before they inherit judgment.
The danger is not merely ignorance.
It is confidence without formation.
The Recursum Connection
Most AI interactions optimize for answer delivery.
Recursum's deeper architecture is built around something different: preserving the cognition around the answer.
That means integrating: challenge, contradiction, competing perspectives, historical reasoning, evidence, recursion, and continuity.
The goal should not be to make the answer easier to obtain.
It should be to make the thinker stronger for having obtained it.
AI should not remove the struggle that creates judgment. It should improve the quality of that struggle.
The Parent Test
After your child uses AI, ask them one simple diagnostic question: Could they explain this without the AI?
Not repeat it. Explain it.
Can they: defend it, challenge it, modify it, or recognize when it is wrong?
If not:
The AI completed the assignment. The child may not have completed the learning.
Ending
We should teach children to use AI. They will need it.
We should teach them to collaborate with intelligence more powerful than anything previous generations possessed.
But first, or at least at the same time, we have to protect the process through which they become capable of judging that intelligence.
Because a generation that can produce brilliant work is not necessarily a brilliant generation.
The test is what remains when the AI is turned off.
The purpose of education was never to produce better homework.
It was to produce better humans.
TL;DR: AI allows students to generate advanced work, but evidence shows a significant portion offload core thinking. This isn't about cheating; it's about confusing output with learning. AI makes it possible to separate the quality of the product from the capability of the producer at unprecedented scale. Historically, assignments developed the student, not just the artifact, with "inefficient" cognitive struggle building reasoning and judgment.
While AI has a place in education, the crucial question is what cognitive work must happen before AI enters the loop. Approaches like "teach both, test both" are essential, as AI literacy without independent literacy creates dependency. The danger is a generation whose impressive work masks a lack of acquired competence, leading to employers inheriting students with "judgment without formation." Recursum's architecture aims to preserve cognition by improving the quality of the struggle, not removing it. The ultimate test is what remains when the AI is turned off, as the purpose of education is to produce better humans, not just better homework.
By Ernesto Verdugo, AI Architect, Founder of Verdugo Labs, and Creator of Recursum. He builds systems for what happens when human and artificial intelligence stop working separately.
Links: