When we entered the age of AI, it felt natural that education would be one of the first things it should transform. Learning is deeply personal, often frustrating, and historically gatekept by access to good teachers, expensive tutors, and the right resources. AI promised to change all of that, and in many ways, it already has. But the story is more nuanced than "AI fixes education." It is a story about free tools, fierce competition, and the uncomfortable question of whether anyone can build a sustainable business in a space where the biggest players are giving everything away.

The promise of personalized learning

The core appeal of AI in education is simple: personalized help, available to anyone, at any time. No scheduling a tutor. No waiting for office hours. Just open ChatGPT or NotebookLM, paste in your lecture notes, and start learning at your own pace.

The numbers reflect how quickly students have embraced this. According to a 2025 HEPI survey, 92% of students reported using AI tools, up from 66% the previous year. The proportion using generative AI specifically for assessments jumped from 53% to 88%. The most common uses are explaining concepts, summarizing articles, and suggesting research ideas. AI has gone from a novelty to a default study companion in just a couple of years.

The global AI tutors market tells a similar story. Valued at USD 1.63 billion in 2024, it is projected to reach USD 7.99 billion by 2030, growing at a 30.5% compound annual growth rate. The demand for personalized, adaptive learning tools is real and accelerating.

Does AI replace tutors?

This is the question everyone asks, and the honest answer is: not really, or at least not yet.

Research from Brookings highlights that generative AI is beginning to address longstanding limitations in traditional intelligent tutoring systems, particularly around dialogue, feedback, and content generation. Two recent randomized controlled trials found that AI embedded in live, chat-based math tutoring can improve student academic outcomes. But these gains came from AI augmenting human tutors, not replacing them.

A thought-provoking piece from Social Science Space puts it even more bluntly: AI tutors address perhaps 16% of what it means to develop an intelligent human being. The remaining 84% requires investment in human educators, in the science of learning, and in assessment systems that recognize the full complexity of human intelligence. AI is excellent at drilling facts, generating practice questions, and explaining concepts. It is far less equipped to mentor, motivate, or help students develop the kind of critical thinking that comes from real human interaction.

So the tutoring industry is not disappearing. It is evolving. The future likely looks like hybrid models, where AI handles the repetitive, scalable parts of learning while human tutors focus on the higher-order skills that machines cannot replicate.

How students actually use AI today

The way students use AI has matured significantly. Early on, ChatGPT was essentially a smarter search engine, a place to get quick answers or generate text. Now, students are using AI tools across the entire learning lifecycle:

OpenAI has leaned heavily into this. In 2025, they launched study mode in ChatGPT, a learning experience designed to guide students through problems step by step instead of just handing them answers. They also partnered with the American Federation of Teachers, committing $10 million over five years to equip 400,000 K-12 educators to use AI effectively in classrooms. Their "Education for Countries" initiative aims to work with governments and universities to bring AI into education systems at a national scale.

Google has made similar moves with NotebookLM, which now generates audio overviews, mind maps, and video summaries, all for free. Upload your lecture slides, and it will create a visual knowledge map, generate a podcast-style audio summary, and quiz you on the material. For students who learn visually or through listening, this is transformative.

I tried building one myself

Before any of this existed, I was a computer science student who needed better ways to study. ChatGPT in its early days was a basic chat tool, useful but limited. I wanted something purpose-built for learning, so I built it myself.

I created Decosmic, a tool that let you upload lecture content and turn it into quizzes and mind maps. The idea was simple: as a visual learner, I needed to see how concepts connected, and I needed to practice actively rather than passively re-reading notes. Decosmic did exactly that. Upload your materials, and it would generate practice questions and map out the content in a way that made it easier to understand and retain.

It worked well. But then the landscape shifted.

The competition problem

What started as a niche space exploded. Tools like StudyFetch, Caktus AI, TutorialJojo, and Anara entered the market, each offering their own take on AI-powered studying. StudyFetch, founded by an 18-year-old during the pandemic, grew to over a million users and raised $10 million by converting notes and textbooks into personalized study tools. Caktus AI offers writing assistance, research help, and study aids starting at $14.99 per month.

And then the giants showed up.

ChatGPT added a built-in study mode with quizzing capabilities. Google's NotebookLM rolled out mind maps, video generation, and audio summaries, all completely free. These were not half-baked features. They were polished, powerful, and backed by the resources of the world's largest AI companies.

This created an existential question for every startup in the space, including Decosmic: why would a student pay for something when the free tools do the same thing, or better?

NotebookLM, in particular, became almost exactly what Decosmic was envisioned to be. Free mind maps, free audio explanations, free quizzes, all grounded in your own uploaded content. It was hard to compete with free, especially when "free" came from Google.

The free tools paradox

The AI education space is validated. Students clearly want these tools, they clearly use them, and the market is growing rapidly. The UGC content around apps like StudyFetch and Caktus AI proves there is genuine demand. But validation and monetization are two different things.

From the perspective of someone who built and ran one of these tools, the economics are brutal. You are paying API costs to serve users who increasingly expect everything to be free, because the platforms providing the underlying AI are also building the consumer-facing products. Google and OpenAI are not trying to make money on education tools directly. They are trying to capture the next generation of users, build brand loyalty, and collect data on how people learn. Education is a no-brainer entry point for them. It is the first thing AI should help with, and they know it.

This is why I pivoted Decosmic to a B2B model. Selling to individual students when ChatGPT and NotebookLM are free does not make sense. But selling to institutions, to schools and universities that need structured, accountable, and customizable learning tools, that is a different conversation. Institutions care about things like data privacy, LMS integration, analytics dashboards, and curriculum alignment. Those are problems that free consumer tools do not solve.

A personal reflection

Sometimes I think about how different things were when I was growing up. We did not have access to any of this. When you were stuck on homework, your options were limited: Google the topic (and hope you found a decent explanation), ask someone who might know, or just struggle through it. There was no AI tutor to walk you through a problem at 11 PM. No tool to turn a dense textbook chapter into a set of practice questions in seconds. No mind map generator to help you see how everything connected.

It was genuinely hard, and there was often no way around it.

That is exactly why AI in education matters so much. Not because it replaces teachers or makes learning effortless, but because it removes barriers. The student who cannot afford a tutor now has access to one. The visual learner who struggles with dense text can generate a mind map in seconds. The student studying in a second language can get explanations tailored to their level.

The tools are not perfect. AI still gets things wrong, still lacks the empathy and mentorship that great teachers provide, and still only addresses a fraction of what real learning involves. But for the 84% of learning that AI cannot yet touch, the hope is that by handling the 16% well, it frees up human educators to focus on everything else.

The age of AI in education is here. The question is no longer whether these tools are useful. It is who builds them, who pays for them, and whether there is room for anyone beyond the giants.

References