AI Agents as Peer Tutors: A Dutch Secondary School Case Study

In recent years, the integration of artificial intelligence into educational environments has moved beyond experimental pilots into the heart of classroom practice. Among the most promising innovations is the use of AI agents as peer tutors, designed not to replace human educators but to augment and enrich the learning experience. This article presents a detailed account of a Dutch secondary school’s adoption of a multi-agent tutor system, examining the processes, results, and lived experiences of both teachers and students.

The Context: A Dutch School’s Bold Experiment in AI Tutoring

Located in the outskirts of Utrecht, Het Nieuwe Lyceum is a comprehensive secondary school serving a diverse student population. In response to challenges presented by heterogeneous classroom abilities and the need for differentiated instruction, the school partnered with researchers and developers to implement a multi-agent AI peer tutoring system. The primary objective: to foster deeper learning, boost student engagement, and relieve some of the instructional burden from teachers.

“Our classrooms are full of students with different backgrounds, skills, and levels of motivation. The AI agents promised a way to reach each learner more personally, without compromising the whole class dynamic,” reflects Marjan Visser, a mathematics teacher with over two decades of experience.

How Multi-Agent AI Tutors Were Implemented

The school selected an open-source AI platform capable of hosting multiple specialized agents, each designed with a distinct pedagogical function. Some agents focused on explaining concepts in natural language, while others provided scaffolding for problem-solving or gave real-time feedback on student work. The system integrated seamlessly with the school’s digital learning environment, allowing students to interact with AI tutors via laptops and tablets.

Each student was assigned a personalized profile, enabling the AI agents to adapt their responses and guidance based on individual learning histories, preferences, and performance data. Importantly, teachers retained oversight and could customize the agents’ instructional strategies, setting limits on the types of help provided or activating specific modes (such as “hint only” or “step-by-step solution”).

Results: Measuring Impact and Perception

The pilot ran across three grade levels, involving approximately 450 students and 18 teachers in mathematics, biology, and English language classes. Data was collected through learning analytics, standardized tests, classroom observations, and semi-structured interviews with both students and teachers.

Student Performance Metrics

Quantitative results indicated statistically significant improvements in both formative and summative assessments for students who regularly engaged with the AI tutors. On average, students in the experimental cohort showed a 13% increase in test scores over a semester compared to a control group. Notably, the largest gains were observed among students who had previously struggled with independent practice and those from non-Dutch speaking backgrounds.

“I used to give up when I didn’t understand a problem, but now the AI agent helps me step by step. I feel more confident and I even help my classmates sometimes,” reports Fatima, a Year 10 student.

Engagement and Motivation

Beyond test scores, metrics related to student engagement revealed positive trends. Log data showed increased time-on-task and a reduction in off-topic digital activity during lessons. Student surveys highlighted a greater sense of autonomy and willingness to tackle challenging problems, with 74% of respondents agreeing that the AI agents made learning “less stressful” and “more interesting.”

Teachers also noted a shift in classroom atmosphere. With routine explanations and feedback delegated to AI agents, educators could devote more time to orchestrating discussions, addressing misconceptions, and supporting students with complex needs. Observations recorded by external evaluators described “a more collaborative and inquiry-driven environment.”

Teacher Experiences and Perspectives

One of the most compelling aspects of the case study lies in the nuanced experiences of the teaching staff. Initial skepticism gave way to cautious optimism as teachers observed the AI agents in action.

“At first, I was worried that students would rely too much on technology. But over time, I saw them becoming more independent, asking better questions, and reflecting on their mistakes with the AI’s help,” says Pieter de Haan, a biology teacher.

Professional Development and Agency

Teachers participated in workshops and ongoing coaching sessions to familiarize themselves with the AI system’s features and pedagogical possibilities. These sessions were crucial for fostering a sense of agency and ownership; rather than feeling replaced, teachers became co-designers of the AI tutors’ behavior. This collaborative approach allowed for continuous adjustment based on classroom realities.

Importantly, teachers reported that the system’s transparency—the ability to see how the AI reached certain explanations or feedback—was critical for trust and pedagogical alignment. When misunderstandings occurred, teachers could quickly intervene, correct the AI, and use the incident as a teaching moment.

“The AI is not perfect, but neither are we. What matters is that it gives us new ways to connect with our students and personalize support where it’s needed most,” notes Marjan Visser.

Challenges and Ethical Considerations

The implementation was not without its difficulties. Some students initially attempted to exploit the AI agents by seeking direct answers rather than engaging in genuine problem-solving. The development team responded by refining the agents’ scaffolding logic, nudging students toward reflection and self-explanation rather than quick solutions.

Teachers also voiced concerns regarding data privacy and transparency. The school adopted GDPR-compliant protocols, ensuring that all student data was anonymized and securely stored. Regular information sessions for students and parents addressed questions about how the AI operated and what data was collected.

Insights for European Educators: Lessons Learned

The Dutch case study offers a number of transferable insights for educators across Europe seeking to harness AI agents as peer tutors:

• Start small and iterate: Begin with a limited pilot, involve teachers early, and use feedback to refine the system.
• Prioritize transparency: Choose or design AI tools that allow educators to understand, monitor, and adjust agent behavior.
• Center ethical practice: Ensure full compliance with data protection laws and involve all stakeholders in conversations about privacy and consent.
• Invest in professional learning: Ongoing teacher training and support are essential for successful adoption and integration.
• Encourage student agency: Design AI tutors to foster metacognition, resilience, and a growth mindset, rather than simply dispensing answers.

Legislative and Policy Considerations

Use of AI in European classrooms must align with the evolving regulatory landscape, notably the EU’s Artificial Intelligence Act and GDPR. The Dutch school’s approach—emphasizing transparency, human oversight, and privacy by design—provides a model for responsible innovation. Collaboration with legal and ethical experts from the outset ensured that the project not only met compliance standards but also embodied the spirit of European values: inclusivity, fairness, and respect for individual rights.

The Human-AI Partnership: Looking Ahead

Perhaps the most profound lesson from Het Nieuwe Lyceum’s experiment is that AI agents, when thoughtfully integrated, can become authentic partners in the educational journey. The agents’ capacity to offer immediate, personalized feedback and to adapt to diverse learner needs has freed teachers to focus on the inherently human dimensions of their craft—empathy, encouragement, and the nurturing of curiosity.

“The best thing about the AI tutors is not that they know everything, but that they help me learn how to find answers myself,” reflects Ahmed, a Year 11 student.

For European educators considering similar innovations, this case study underscores the importance of embracing both the technical and human elements of change. As the field of educational AI matures, the experiences of Dutch teachers and students illuminate a path forward: one grounded in collaboration, ethical practice, and a shared commitment to helping every learner thrive.