AI tutoring has solved the technical problem of delivering personalized instruction at scale, but educators are discovering a harder challenge: getting students to actually sit down and use these tools instead of doing something more interesting. While platforms like Khanmigo, MagicSchool, and Photomath represent a genuine breakthrough in addressing Benjamin Bloom's decades-old "two-sigma problem," experts now point to what they call the "last mile problem" in education, a metaphor borrowed from logistics that describes the disproportionate cost and complexity of the final step in any network.

What Is the Last Mile Problem in AI Education?

The last mile problem in education refers to the gap between having an excellent learning tool and actually getting students motivated to use it consistently. Rebecca Birch, an education writer and former English teacher, explains that this challenge has plagued edtech for years, even when the underlying technology was sound. "The issue was never, and has never, been the limitations of tech," Birch noted, citing research from educator Daisy Christodoulou showing that personalized learning and non-human tutors have existed since the 1990s. The real obstacle is human motivation. Students may complain about school, but many actually prefer learning from human teachers who provide the right social pressure, encouragement, and relational support that algorithms cannot easily replicate.

Khan Academy's quiet discontinuation of its chatbot a few months ago illustrated this problem starkly. Despite Khan Academy's strong track record and substantial funding, the platform struggled to get students to engage consistently with the tool. "The issue wasn't the tool, but the challenge of getting students to use it," Birch explained. This mirrors a broader pattern in education coaching programs, where feedback and guidance are provided but teachers struggle to find time to review and act on the advice, especially when it comes from a non-human source.

How Are Leading AI Tutoring Platforms Addressing This Challenge?

Despite the motivation hurdle, AI tutoring has made significant progress since 2023. The category has grown from a credibility problem to a demonstrable solution, with several structural shifts reshaping how these tools are deployed. Khanmigo, Khan Academy's AI tutor launched in March 2023 in partnership with OpenAI, now serves more than 70,000 students and educators across U.S. school districts under sustained pilot and full-deployment agreements. The platform's nonprofit positioning and existing trust in K-12 procurement have made it the default starting point for any district evaluating AI tutoring.

The major platforms in the AI tutoring space include:

• Student-Facing Tools: Khanmigo, Photomath (acquired by Google in 2022), and Socratic provide direct personalized instruction, step-by-step problem solving, and adaptive feedback to learners.
• Teacher-Facing Tools: MagicSchool assists with lesson planning, differentiation, rubric generation, and feedback writing, allowing teachers to shape AI support rather than students interacting directly with AI.
• Homework and Study Platforms: Quizlet AI and Chegg AI layer tutoring capabilities onto existing study infrastructure, with Chegg pivoting aggressively toward AI tutoring after losing 75% of its market value in 2023 when ChatGPT's free homework help collapsed its traditional subscription model.

The teacher-facing category has emerged as particularly important in addressing motivation concerns. MagicSchool and similar tools allow educators to maintain control over how AI supports learning, avoiding the credibility issues some parents and administrators raise about direct student-AI interaction. Notably, the teacher-facing category is now larger than the student-direct category by adoption count, even though student-direct tools attract more media attention.

Why Does Motivation Matter More Than Technology?

Birch's analysis reveals a uncomfortable truth for edtech advocates: even when technology is excellent, motivation remains the binding constraint. She points to Alpha School, an elite model that combines direct instruction with AI responsiveness and afternoon passion projects, as an example of how schools are trying to solve motivation through incentive structures. Students complete their core learning quickly to unlock free time for projects they care about. While learning gains are impressive, Birch worries this approach treats learning as a transactional chore rather than something intrinsically valuable.

The human element matters most. "In the same way, the last mile in education is from coach to teacher or teacher to student. And I feel pretty confident that both parties will need to be human for the foreseeable future," Birch concluded. This insight challenges the assumption that better technology automatically leads to better outcomes. Students are naturally inclined toward growth, but it is the actions of teachers that provide the quality of the soil and nutrients for that growth. As yet, edtech has not solved this fundamental problem.

How Can Teachers Build and Control AI Learning Tools?

A emerging approach to the motivation and control problem is empowering teachers to design AI tools themselves rather than simply adopting pre-built platforms. LearnAdapt Agentic Studio, selected for showcase at the 2026 International Conference on Artificial Intelligence in Education (AIED), represents a shift from tools teachers use to AI teammates teachers build. The platform allows educators to move from a classroom problem to a working AI-supported application through a structured, no-code workflow.

The key innovation is teacher agency. Instead of waiting for a generic tool to fit their needs, teachers can describe a problem in plain classroom language and shape a pedagogical AI plugin around it. Teachers can then inspect what the system understands, edit the design if misunderstood, preview how the plugin works, and decide what evidence returns to them and what students see.

This approach addresses several critical governance and trust questions:

• Teacher Control: Teachers can inspect, edit, approve, override, hide, export, and govern how AI applications behave in real learning contexts, ensuring the tool aligns with pedagogical intent.
• Evidence and Visibility: Teachers define what data is captured, what students can see, what remains private, and where human judgment is required before AI makes decisions.
• Classroom Fit: Rather than adopting a one-size-fits-all solution, teachers shape AI support around specific learning problems, student needs, and classroom constraints they understand intimately.

"Teachers should not only use educational AI tools. They should be able to build, inspect, edit, govern, and adapt them," the LearnAdapt team stated. This philosophy reflects a broader recognition that educational AI will only succeed if teachers remain the pedagogical authority throughout the process, not merely implementers of systems designed elsewhere.

What Does This Mean for the Future of AI in Schools?

The convergence of these insights suggests that AI tutoring's next phase will focus less on raw capability and more on motivation, teacher empowerment, and governance. The credibility moment has passed; Khanmigo's classroom deployment data, MagicSchool's adoption, and broader research validating learning improvements have established that well-designed AI tutoring can work. The challenge now is ensuring these tools fit into real classrooms where human relationships, teacher judgment, and intrinsic motivation remain central to learning.

As AI becomes more agentic and capable, the question shifts from "Can AI tutor effectively?" to "How do we ensure teachers control the AI tools shaping their students' learning?" The last mile problem will not be solved by better algorithms alone. It will be solved when teachers have the agency to shape AI tools around their pedagogical intent, when students feel the relational support that only humans can provide, and when technology serves learning rather than replacing the human connections that make learning meaningful.