In 2013 Caleb Hicks was teaching nearly 300 students a day and noticed something you’ll find familiar: extremes stand out, but most students get lost in the middle. When ChatGPT arrived in 2022 a practical, not futuristic, question appeared: can AI help spot those students who go unnoticed and give teachers back real time? Hicks answered by founding SchoolAI in 2023, and in two years the platform reached 1 million classrooms in more than 80 countries with hundreds of educational partnerships. (openai.com)

Lesson 1: Keep the teacher at the center

Does AI replace the teacher or amplify them? SchoolAI bets on the latter: every interaction between student and AI is visible to the teacher. In practice this means teachers create learning spaces with Dot, the tool that builds differentiated lessons in seconds, and students work with Sidekick, an adaptive tutor that uses advanced models to adjust pace and support.

"If AI just gives the student the answer, we’ve failed—the point of teaching is to coach and to keep them engaged in the work."

That "teacher-in-the-loop" approach looks for early signs of disengagement so the teacher can step in before a small issue grows. Logging and transparency are built in: everything is observable and auditable by the teaching team. (openai.com)

Lesson 2: Use the right model for the right task

Not every problem needs the same engine. SchoolAI organizes the workflow as agent graphs: each student input passes through specialized nodes that call models, tools and guardrails before returning support. Advantage? Better cost control and higher accuracy where it really matters.

• GPT-4o powers conversation and rapid lesson construction.
• GPT-4.1 is used for deeper reasoning, for example in multi-step math problems.
• Image generation creates custom diagrams or maps to support the lesson.
• Text-to-speech provides spoken feedback in over 60 languages.

Smart routing sends expensive tasks to the most powerful models and light checks to smaller variants, keeping predictability in spend without sacrificing quality. (openai.com)

Lesson 3: Choose a stack and scale with support

Scaling in education demands efficiency and technical backing. When SchoolAI ran a demo for 10,000 teachers, their relationship with OpenAI let them raise usage limits quickly to ensure a smooth experience. Also, lower inference costs helped reduce the cost per student from almost a dollar to a fraction, freeing margin to invest in growth and school support. (openai.com)

What does this change in the real life of a school?

Imagine a newcomer who speaks only Dari and doesn’t participate in class. With real-time translation and adaptive support, that student can start participating, form bonds with classmates and regain confidence within weeks. Concrete examples teachers report: earlier signals, timely interventions and more time for human teaching.

SchoolAI also says teachers have reclaimed more than 10 hours weekly thanks to automation of routine tasks—time that turns into individual attention and noticing subtle shifts in behavior or performance. (openai.com)

If you’re a teacher, school leader or edtech creator, the lesson is clear: AI works when it’s designed to accompany the teacher, is observable and fits the real flow of the classroom. If you want to see the platform, SchoolAI and OpenAI’s developer page are good starting points. (openai.com)

Final reflection

The news isn’t that there’s a new AI, but that someone set rules and tools so AI does what it should in a school: reveal what was unseen and give teachers time back for what only they can do. Can you imagine how much could change with an extra 10 hours a week to teach?