AI can give every student an answer but can it help every student learn?

What 14 years of building math technology taught us is that there’s a fundamental difference between adding AI to a product and building AI around how students learn.

Every math teacher has said: “Show your work.”

The answer matters, but so does the reasoning behind it. Students need to understand each step, recognize where their thinking went wrong and learn how to move forward on their own. 

For teachers, this creates a familiar challenge. Students need timely, individualized support, but one teacher cannot be beside every student at every moment. For superintendents and district leaders, the challenge is broader:

How can schools provide effective math support at scale without lowering the quality of instruction or increasing teacher workload?

AI could become part of the solution. But its educational value depends on a more important question: Does the technology help students think, or does it simply give them answers faster?

That question has shaped Mathleaks360 from the beginning. In the first episode of The Education Equation podcast, CEO and co-founder Jesper Mårtensson shared how his own frustration as a student ultimately led to Mathleaks360, and years later, to the development of Educational AI built specifically for math education. 

His central argument is straightforward: effective AI in K–12 education must begin with how students learn, not with the technology a company can most quickly bring to market. The idea is simple: effective AI in K-12 education should start with how students learn, not with how quickly a company can add AI to its product.

The Problem: Students need to understand reasoning, not just get the answer

Jesper’s journey began with his own experience as a student in Sweden. He enjoyed math, but like many students, there were times when he needed another explanation or one more step to understand a difficult concept. His teachers couldn’t always provide that support at the exact moment he needed it. At university, Jesper experienced something different. Before exams, students received worked examples that showed not only the final answers but the complete reasoning behind them.

The difference was significant. Instead of merely checking whether an answer was correct, students could study the process, identify where their own reasoning had broken down and apply the method to similar problems. It raised an obvious question: Why was this kind of support not widely available to high school students?

The idea: Make step-by-step math support available to every student

The first version of the platform was simple: Students could find a problem from their textbook and see a complete, step-by-step solution. These weren’t  answer keys or generic hints. They were carefully written explanations designed to help students understand the method and work through similar problems on their own. The idea quickly gained traction. According to Mathleaks360’ company data, around 40 percent of Swedish high school students were using the platform within a few years. Some students described the experience as a “brain massage”: that feeling of relief that comes when a difficult concept finally clicks. Students don’t always need someone to give them the answer. They need the right support to see the next step.

For school leaders evaluating AI tutoring or digital math tools, this distinction matters. A tool that completes the work  for a student isn’t necessarily helping that student learn how to do it on their own.

14 years of building around how students learn math

For 14 years, Mathleaks360 has relied on experts in math and education to write review and refine step-by-step solutions. The result is more than a collection of correct answers. It reflects years of understanding which steps matter, where students are likely to struggle, how misconceptions develop and how  difficult concepts can be explained clearly.

That foundation became especially important when generative AI entered the mainstream in 2022. The EdTech market changed almost overnight. Companies could suddenly integrate powerful general-purpose AI models into existing products and quickly launch AI-Powered chatbots, tutors and teaching assistants. These tools can be useful. But access to a powerful model does not automatically create an effective educational product.

A general-purpose AI model is designed to respond to a wide range of requests. It is not necessarily designed to teach math, identify student misconceptions or support a particular instructional approach. It may provide a strong explanation in one interaction and an unclear or incorrect explanation in another.

In a K–12 setting, that inconsistency is not a minor technical issue. It can reinforce misconceptions, reduce trust and make it harder for districts to determine whether a tool is genuinely supporting learning.

For Jesper, the rapid rise of generative AI reinforced the value of what Mathleaks360 had already spent years building: a consistent foundation shaped and validated by human experts. The technology: Educational AI designed specifically for math learning

Mathleaks360 decided to use that foundation to develop specialized AI models for math education. The distinction isn’t about who owns the technology. It’s about purpose, control and accountability. What was the system designed to do? What information shaped its responses? Can the company address weaknesses in the model rather than wait for an outside provider to change it?

When an educational product relies entirely on a general-purpose model, it also relies on that model’s behavior, updates and limitations. Purpose-built educational AI allows developers to focus on a defined learning problem: helping students understand how to solve math problems rather than simply generating a final answer. Visual reasoning provides a useful example.

Math understanding is not expressed through words and numbers alone. A graph, diagram or geometric figure can make the difference between confusion and clarity. The challenge for AI is not simply producing an image. The system must recognize when a visual explanation is needed and determine which visual will best support the reasoning. That requires more than calculation. It requires an understanding of how mathematical ideas are taught and communicated.

Mathleaks360 has trained its technology using years of examples created and reviewed by human experts. These examples demonstrate both how to explain a solution and when a diagram can make the reasoning clearer.

The work is still developing. According to Jesper, the model currently achieves approximately 72 percent accuracy on problems requiring this type of creative visual reasoning. That figure is not presented as proof that the challenge has been solved. It illustrates both the difficulty of the problem and what becomes possible when AI is developed for a specific educational purpose.

The future: AI that strengthens teaching and learning

The future of AI in K–12 education should not be a race to place the newest model in every classroom. It should be a disciplined effort to determine where AI can improve teaching and learning, and to build evidence that it does.

For Mathleaks360, that means developing technology that supports independent reasoning. An effective AI math tutor should recognize where a student is struggling, provide an appropriate level of support and know when to step back. It should help the student move forward without removing the productive struggle that learning requires.

This vision keeps teachers at the center. AI can provide consistent explanations, demonstrate procedures and support students between lessons. Teachers can then spend more time on feedback, discussion, motivation and the relationships that make learning possible.

Technology may help answer, “How do I solve this?” But students also ask, “Why am I learning this?” That question still requires human judgment, context and connection.

For superintendents and district leaders, the larger lesson is that not every AI-powered education product is built in the same way or for the same purpose. The presence of a chatbot says little about the quality of the learning experience behind it.

School systems should look beneath the interface. They should ask whether the technology was designed for the educational problem it claims to solve, what knowledge and data shaped it, how it supports student reasoning and what evidence demonstrates its impact.

After 14 years, the idea behind Mathleaks360 remains the same: students need more than answers. They need support that helps them understand the reasoning and continue on their own.

AI creates new possibilities for providing that support at scale. But meaningful educational technology cannot be built by starting with AI and searching for a use. It must begin with a real learning problem, draw on a deep understanding of how students learn and be developed with a clear educational purpose.

The future of educational AI will not be shaped by who adds AI the fastest. It will be shaped by who builds it around what students, teachers and school systems actually need.