Can AI Teach? What Building Proglly Taught Us

Can artificial intelligence teach? The question saturates the EdTech industry. Every major platform is building an answer to it. The implicit frame: AI as the teacher, student as the receiver. Better explanation, delivered at the right level, personalised to the individual.

It is a compelling frame. It is also, we believe, the wrong one.

In 2024, Proglly was an AI video generation platform for students. Type any topic, receive a personalised explainer video. Narrated. Animated. Calibrated to the student's level, not a generic curriculum.

The product worked. Educators called it a step change. In the Base44 Back-to-School App Challenge, Proglly won the $10,000 Grand Prize.

The judges were right. It was genuinely impressive. Nobody else was building this. And then we looked at what students were actually doing after they watched the video. They were closing the app.

The problem was not the video quality. The problem was the format. Watching a video is passive. Even a perfectly crafted, perfectly personalised video is something that happens to a student, not something they do.

Cognitive science is clear on what produces durable learning: retrieval practice, elaborative interrogation, and active encoding — generating understanding rather than receiving it. A video produces recognition. It does not produce retrieval. The ceiling of video-based learning is a format problem, not a quality problem.

See the difference directly — toggle between what we built first and what we built instead.

If the insight is that active beats passive, the design question becomes: how do we make learning something students do, not something they watch?

The answer: five formats in parallel. Visual cards, real diagrams, interactive diagrams (where students tap any component to interrogate it), mindmaps, and smart quizzes — each activating a different learning mechanism. One topic in. Five interactive formats out. Thirty seconds.

Can AI teach? Our reframe: the question assumes the bottleneck in learning is the quality of explanation available. But the bottleneck for most students is format and engagement — the gap between passive reception and active understanding.

AI does not need to be a better teacher. It needs to create the conditions for better learning. Those are different design briefs, with significantly different product architectures.