Since 2025, a World Bank Group (WBG) program has been bringing artificial intelligence (AI) into Peru’s public schools. An AI math tutor and a career coach have reached about 4,500 students across 85 schools in Lima. Also, short, practical training is building AI fluency among teachers and principals. Teachers are using these tools to improve lesson planning and delivery, while principals are reducing administrative burdens and spending more time on pedagogical leadership.

Challenge

Peru’s education system faces significant challenges, and this project targets three key constraints. Only 13 percent of secondary students reach the expected level in mathematics — a sign that too many students leave school without the foundations they need. At the same time, teachers rarely receive high-quality feedback, and school leaders have limited time to support instruction. As a result, students face key decisions about their future with little guidance, reaching a high-stakes crossroads — entering the workforce or pursuing higher education — without the support needed to choose well.

WBG Approach

The WBG’s approach rests on a single idea: AI improves education only when it reaches the whole system and is paired with the human support that makes it work. Rather than simply handing out tools, the program combines AI with practical training, clear quality frameworks, and steady institutional support. It applies this approach across every part of the school, supporting students as they learn, teachers as they instruct, coaches as they provide feedback, and principals as they lead.

The result is a set of low-cost, locally grounded innovations which are designed to fit the system’s existing capacity and conditions so they can scale. These innovations include: Choosing My Path (Eligiendo mi Camino), which combines a personalized AI math tutor and career coach for students; certified hands-on AI training for teachers; automated classroom-observation feedback; and El Director Libre, a practical AI methodology that frees principals from paperwork to focus on instructional leadership. The work is delivered with partners, including the edtech firm uDocz, Anthropic and Microsoft -and built so that other governments can adapt what works.

Results

• Eligiendo mi Camino, combining an AI Math Tutor and Career Coach, reached 85 public schools and roughly 4,500 fifth-year secondary students in Lima, Perú (since March 2026).
• El Director Libre, a method that frees school principals from paperwork for instructional leadership, is now applied in Brazil, Paraguay, Costa Rica, Colombia, and beyond; in Peru, the Ministry of Education has trained over 6,000 school leaders in it (since 2025).
• Evidence from Peru shows that AI-based classroom observations can produce results comparable to those of trained human observers in scoring pedagogical and instructional quality. This suggests that AI can provide reliable, scalable support for expanding coaching in contexts where expert instructional support is scarce.
• In collaboration with Development Research Group from the WBG, a randomized trial across 390 schools in Lima tested whether a brief, practice-oriented training could help teachers adopt AI in their classrooms. Active AI use increased among the 1,270 participating teachers, who also learned to write stronger prompts and evaluate AI outputs more critically.

Contribution to WBG Targets and Jobs

This work supports the WBG’s goal of halving learning poverty: an AI math tutor builds numeracy that future work depends on. An AI career coach helps young people decide what to study and which careers to pursue, easing the move from school to work. And for teachers, the training is job training in its own right, -- its largest gains going to older teachers, the workers most likely to be left behind by new technology.

Lessons Learned

First, technology alone rarely changes outcomes. Teachers already had access to AI tools, yet usage remained limited until they received practical training and support. The same pattern emerged for students, coaches, and principals. Results came not from the technology itself, but from combining it with clear guidance, structured implementation, and ongoing support. Simply placing technology in schools seldom improves learning. Second, design matters. The quality of the results depended less on the sophistication of the underlying model than on the pedagogical frameworks, instructions, and rubrics built around it. AI can help expand access to high-quality support, but only when it is intentionally designed for educational purposes and accompanied by appropriate safeguards, particularly when students are involved.

Next Steps

The central open question is whether these early gains translate into stronger student learning, and the WBG will keep testing to find out. A randomized controlled trial now underway in Peru is examining how to design AI tutors that build lasting skills rather than short-term gains. In parallel, the WBG is documenting practical, country-adaptable guidance so that education ministries across Latin America and the Caribbean can replicate what works. This work is already being extended beyond Peru, with similar applications emerging across the region, including AI-supported career coaching in Ecuador and AI tutoring initiatives in El Salvador. Sustainability will depend on building training, safeguards, and institutional support into national systems, ensuring that AI strengthens good teaching rather than substituting for it as the tools keep evolving.