What could AI in the classroom look like?

AI has the potential to expand access to personalized learning by giving more students the kind of support once limited to one-on-one tutoring.

As alarm bells ring, a new narrative is emerging — one that sees generative AI not as a threat to education but as a tool. Sal Khan, founder of Khan Academy, is one of its most vocal champions. Known for delivering free, high-quality education to learners around the world, Khan sees AI as a chance to personalize and elevate learning for every student.

In his recent book, “Brave New Words: How AI Will Revolutionize Education (and Why That’s a Good Thing),” Khan acknowledged the risks. But he argued they can — and must — be managed. Imperfection or misuse, he contended, are not reasons to sideline such transformative potential.

To Khan, generative AI is on par with the arrival of the internet and personal computers — technologies that reshaped how we communicate, create, and access knowledge. “Used properly, [AI] would positively affect how teachers planned, instructed, and graded,” he wrote. “Soon, students might be able to learn faster and retain more information than ever before, proving AI to be the ultimate learning tool for accelerating human intelligence and potential. AI might hasten learning globally and even get us closer to realizing a world in which every person on earth had access to affordable world-class learning.”

Could generative AI be the catalyst that finally delivers on the promise of truly personalized education?

If you think generative AI in education ends with ChatGPT, think again

ChatGPT was released in November 2022. Several major U.S. school districts — including those in Seattle, Los Angeles, and New York City — quickly moved to ban the tool on school devices, citing concerns over cheating, academic integrity, and the erosion of critical thinking skills.

More than two years later, many educators and institutions remain skeptical, maintaining that generative AI has no place in the classroom. ChatGPT is still banned in many schools, and the overall response is fragmented — marked by hesitation, confusion, and a lack of clear direction. As technology advances rapidly, education is at a crossroads. Khan urged schools to lean away from fear and lean into possibility.

“To take advantage of this technology requires some creativity and bravery, too,” he wrote. “Not blind bravery, but something I have started to call educated bravery, a kind of courage that comes from acknowledging the rational trepidation we all face when presented with sudden advances in technology and then informing ourselves of both the challenges and the potential it presents.”

To fully understand Khan’s argument, it’s helpful to have some background on AI. Here’s a quick overview:

ChatGPT is a generative AI platform designed to respond to prompts in a natural, human-like way. However, it’s not the only tool of its kind. While ChatGPT can be useful for learning, it wasn’t specifically created as a teaching tool. It’s more of a general-purpose AI designed for a wide range of tasks.

Many educators are concerned about its use in education because it can easily provide students with ready-made answers — though not always accurate ones — or generate entire essays at the push of a button.

However, the core technology behind ChatGPT, GPT-4, can power AI tools tailored for specific purposes, including education.

Even before ChatGPT became publicly available, Khan and his team of developers and education experts at Khan Academy set out to imagine what a tool designed specifically for learning would look like. They knew it would need safeguards to protect students from inappropriate content and prevent cheating. As they delved deeper into their research, however, they realized they could unlock far more educational potential than they initially expected.

“The more time we spent testing GPT-4, the more we realized how to mitigate problems with solutions that in many ways made the upsides even better,” Khan noted. “To address concerns around cheating, for instance, we considered what it would take to create an AI tutor that refused to give students answers. Like a good human tutor, it would instead ask leading questions.”

The key insight: GPT-4 technology offers powerful potential for individualized learning when customized for that purpose.

Your personal tutor — at scale

The result of Khan Academy’s work with GPT-4 was Khanmigo — a generative AI tool designed for personalized K-12 learning. It is not the only AI-based tutoring platform available, but Khan uses it to demonstrate what is possible when generative AI is tailored for K-12 learning.

Unlike tools that simply deliver answers, Khanmigo acts more like a thoughtful tutor: It challenges students with questions, encourages critical thinking, and guides them toward deeper understanding. As Khan put it, “Our goal with Khanmigo is to bring the power of one-on-one tutoring to every student, everywhere, helping them learn at their own pace and in their own style.”

The idea behind Khanmigo taps into a long-standing aspiration in education: replicating the power of one-on-one tutoring.

Decades ago, educational psychologist Benjamin Bloom discovered what’s now known as the “2 sigma problem.” He found that students who received individual tutoring performed two standard deviations better than those in traditional classroom settings — the equivalent of jumping from the 50th percentile to the 96th percentile. Tutoring worked. The problem, as Bloom saw it, was scale: It simply wasn’t feasible to provide every student with a personal tutor.

In many traditional classrooms, instruction follows a fixed pace, regardless of whether each student has fully grasped the material. A teacher might introduce fractions, allow a few days for practice, administer a test, and then move on — even if some students score only 80%. That remaining 20% represents a gap in understanding that can snowball and make future topics in algebra and geometry much harder to master.

This approach assumes all students learn at the same rate and in the same way. But in reality, learning is highly individual. Some students need more time and support; others are ready to accelerate. Those who fall behind accumulate unresolved gaps that compound over time, while those ahead are often forced to wait, leading to boredom or disengagement. In this system, the priority becomes covering material, not ensuring mastery.

Khanmigo addresses this challenge head-on by bringing many benefits of one-on-one tutoring to scale. Because it’s designed specifically for education, it doesn’t just deliver answers — it adapts to each student’s pace, asks thoughtful questions, and encourages deeper understanding through Socratic dialogue. It provides real-time feedback, personalized support, and reflection prompts that guide students toward mastery.

In math, Khanmigo breaks down problems step-by-step, asking guiding questions to build true comprehension — and not moving on until the student achieves mastery. In writing, it gives individualized feedback on clarity and structure, helping students revise while keeping their voice intact. In science, it translates complex topics — like evolution or energy transfer — into age-appropriate explanations. In history, it sparks critical thinking through ethical inquiry and role-based dialogue, even speaking as historical figures like Harriet Tubman or Benjamin Franklin to deepen engagement.

What makes Khanmigo even more powerful is its memory. It remembers past conversations, allowing it to learn about a student’s strengths, challenges, and progress over time. This continuity enables more precise, responsive support, like a trusted tutor who knows exactly where a student is and where they need to go next.

But where does this leave teachers?

In an AI-augmented classroom, teachers’ roles don’t diminish — they transform. Teachers shift from being primary content deliverers to becoming orchestrators of learning: guiding, mentoring, and personalizing instruction in ways that weren’t previously possible.

Far from replacing educators, tools like Khanmigo enhance their capacity by automating time-consuming tasks such as grading and providing feedback. This frees teachers to focus on what matters most: building relationships, facilitating deeper discussions, and supporting students through challenges.

“There is no job that is safer in the large-language-model world than teaching,” Khan wrote. “Not only are teachers irreplaceable, but AI is going to support teachers so that they can do more of what they enjoy, from deepening personal connections with their students to developing enriching and creative lessons.”

AI also equips teachers with real-time insights into student progress, struggles, and strengths, enabling more responsive, data-informed decisions. Khanmigo, for instance, tracks not just what students produce, but how they engage. If one student revises an essay through multiple rounds of back-and-forth with Khanmigo, while another submits a polished draft after just a few minutes, that contrast reveals valuable context. It can indicate differing levels of effort and understanding or even the possibility of outside AI-generated content.

This kind of transparency supports not only personalized instruction but also academic integrity, helping teachers intervene meaningfully and guide students toward authentic learning.

With AI as a teaching partner, education becomes less about one-size-fits-all delivery and more about dynamic, individualized support. Teachers are empowered to build trust, spark curiosity, and shape who students become.

It’s a vision that reimagines what learning can look like when every student has access to guidance once reserved for a fortunate few.

“While human coaches and tutors will always be in demand, AI raises the floor for students who have very little access to personalized learning or world-class coursework,” reflected Khan. “Supercharged with memory, incredible content knowledge, a remarkably human and naturalistic voice, an ability to create genuine rapport with students, and the growing ubiquity of access through technology from phones to computers, AI tutors might in time even surpass the results of Bloom’s original findings.”

Khan Academy is supported by Stand Together Trust, which provides funding and strategic capabilities to innovators, scholars, and social entrepreneurs to develop new and better ways to tackle America’s biggest problems.