Why It Works

The Method

Every Agentic Socratic AI course rests on one finding and one answer. The finding is forty years old. The answer is the Lighthouse Coach.

The Problem

Bloom’s Two Sigma

In 1984, educational psychologist Benjamin Bloom reported a striking result: students who learned one-on-one with a tutor, under a mastery-learning discipline, performed about two standard deviations above students taught conventionally. The average tutored student outperformed roughly 98 percent of the classroom group.

Bloom called the challenge of reproducing that result the “two sigma problem”: the method worked, but one expert tutor for every learner was too costly and impractical to scale. For four decades, the best-known way to learn stayed the least available — reserved for the few who could put a tutor in the room.

That problem is why Cognosco Learning exists in its present form. Since 2002 we have taught in the room, one group at a time. The question Bloom left behind is the one we set out to answer: how does every learner get the one-to-one discipline — without the one-to-one cost?

The Problem — and the answer
The Answer

One Coach, One Learner at a Time

The Lighthouse Coach is our answer to Bloom’s problem: an agentic AI coach that works the way Bloom’s tutors worked — Socratically, one learner at a time. It asks the question, listens to the answer, names the misconception it hears, affirms what is right, and asks the next question. It does not advance a learner merely because time has passed; the learner must explain, apply, and prove each idea. Mastery is the standard.

That discipline now runs across the whole Signature Series — over 300 Agentic Socratic AI Learning courses live today at $195 each, alongside Cognosco’s over 300 Live Learning courses: over 3,000 hours of high-impact Socratic learning, with every Agentic Socratic AI course held to that same method.

Bloom’s figures belong to Bloom’s research, not to our catalog. We cite them because they name the standard honest learning should aim for — not because Cognosco claims its courses have been measured at two sigma.

The Proof — Harvard, 2025

In 2025, researchers at Harvard ran the modern version of Bloom’s experiment. In a randomized controlled trial in Harvard’s largest physics course, students learned the same lessons two ways: in class, through the best active learning experienced instructors could deliver — or with an AI tutor the research team had designed for the course.

The AI-tutored students’ learning gains were more than double those of the in-class group. They got there faster, too — a median of 49 minutes against a full class session — and they reported significantly higher engagement and motivation. The study was published in Scientific Reports in 2025.

How that tutor taught matters as much as the result. It did not lecture, and it did not lead with answers. It guided students one step at a time and asked them to attempt each step first — the guided-questioning discipline at the heart of Socratic learning.

The researchers add a caution, and we repeat it because it is honest: this was one course at one university, on first-exposure material, and long-term retention was not tested. Evidence, not magic. But the question is no longer whether one-to-one AI tutoring can outperform the classroom. Harvard measured it.

Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, S. (2025). “AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting.” Scientific Reports, 15, 17458.

See It Work

Hear the method work

One full concept, coached end to end — free. Meet Sutton, take the Preview, and judge the method the way every learner does: by working with it.