The $55,000 Question: What Happens When AI Replaces the Teacher?

This week, CBS Chicago reported that Alpha Schools, a private school built entirely around AI-driven instruction, is enrolling students in the city for fall 2026. The model is striking in its simplicity: children spend one to two hours a day learning core subjects from AI programs. There are no traditional teachers. Instead, each student is assigned a "guide" who leads workshops on public speaking, coding, and outdoor education. According to the school, their students rank in the top 1% on national standardized tests, growing 2.6 times faster than peers on MAP assessments.

The cost: $55,000 per child per year.

It is easy to have a strong reaction to this model. Depending on your vantage point, Alpha Schools is either a glimpse of the future or a cautionary tale about what happens when efficiency becomes the only design principle. But the most interesting thing about Alpha Schools is not the school itself. It is the question buried inside its premise.

The Question Inside the Model

Alpha Schools starts from a radical assumption: the bottleneck in education is the teacher. Remove the teacher from direct instruction, replace them with AI that can personalize pacing and content, and learning accelerates. The data they cite suggests it works, at least by the metrics they have chosen to measure.

But here is the question that does not appear in the enrollment brochure: what are we measuring?

Standardized test performance is one signal. An important one. But anyone who has spent time in a classroom knows that the most consequential moments in a student's education rarely show up on a MAP assessment. The moment a student pauses mid-sentence, reconsiders an assumption, and starts over. The conversation where a teacher notices a pattern in a student's thinking that the student cannot see yet. The revision that happens not because the student was told to revise, but because something shifted in how they understood the problem.

These moments are invisible to standardized metrics. They are also where the deepest learning lives.

What Guides Cannot See

The Alpha Schools model redefines the adult in the room. A "guide" is not a teacher. They do not observe the student working through a difficult math concept. They do not see the essay draft that was abandoned and restarted. They are present for the workshops, the outdoor education, the social development. This is valuable. It is not instruction.

When AI handles the instruction and the human handles the enrichment, something falls into the gap between them: the process of thinking. The AI delivers content and checks answers. The guide facilitates activities. But who watches the student think? Who notices the pattern of avoidance, the moment of breakthrough, the revision that signals genuine understanding?

A Harvard randomized controlled trial published in Nature Scientific Reports found that students using a well-designed AI tutor learned significantly more, in less time, than students in traditional active-learning classrooms. The effect size was between 0.73 and 1.3 standard deviations. But the researchers were careful to note what made their AI tutor effective: it was built around scaffolded questioning and formative feedback. It did not simply deliver content. It engaged the student in the process of reasoning.

The difference between an AI that delivers content and an AI that develops thinking is the difference between a textbook that talks and a teacher who listens.

The Visibility Problem

The OECD's Digital Education Outlook 2026 documented the cost of getting this wrong. Students using general-purpose AI tools saw a 48% performance boost. When the AI was removed, their performance dropped 17% below baseline. The researchers named it "metacognitive laziness." The tool did the thinking. The student's capacity to think independently atrophied.

But purpose-built Socratic AI tools, the kind that ask questions rather than give answers, produced sustained gains. Real learning. The kind that persists after the tool is put away.

This is the design choice that Alpha Schools' model, and every school's model, ultimately comes down to. Not whether AI should be in the classroom. But whether the AI makes student thinking visible or invisible. Whether it captures the process or only the product.

A Different Architecture

At Koan, we start from a different assumption than Alpha Schools. We believe the teacher is not the bottleneck. The teacher is the irreplaceable element. What teachers lack is not a replacement. What they lack is visibility into what happens between the assignment and the submission.

Our AI tutor, Aidan, does not replace instruction. It sits beside the student during the hardest part of learning: the thinking. It asks Socratic questions calibrated to the rubric and adapted to each student's patterns. It does not write for them. It asks them what they are trying to say, and then asks again when the answer is vague.

And every moment of that process is captured. Every revision, every pause, every shift in reasoning. Not as surveillance, but as a timeline of thought. When a teacher opens the WorkHub, they do not see a finished product. They see the journey: the false starts, the three drafts, the five minutes a student sat with an idea before it clicked.

This is what we mean by making learning visible. Not replacing the teacher with AI, but giving the teacher something they have never had: a window into the process of thinking as it happens.

What $55,000 Cannot Buy

Alpha Schools will find families willing to pay $55,000 for its model. The test scores will likely remain strong. For a certain kind of student, self-motivated and disciplined, the approach may work well by certain measures.

But the conversation Alpha Schools has started matters more than the school itself. Because the question is not really about whether AI can teach. The research increasingly says it can. The question is whether we want an education system that optimizes for performance on assessments, or one that develops the capacity to think.

Those are not the same thing. The OECD data proves it. A 48% boost followed by a 17% collapse is the signature of a system that optimized for the wrong metric.

The schools that will define the next decade of education will not be the ones that removed the teacher. They will be the ones that gave teachers the ability to see what was always hidden: the process by which a student learns to think for themselves.

If a student's test score goes up but their ability to think without AI goes down, did they learn? And if we cannot see the process, how would we ever know?