“Before I met Gallito, I had no real idea what I wanted to do.” The student who said it attends a public school in Lima, months away from the biggest choice of her life: what to do after secondary school. Gallito (little rooster) is the AI coach behind that sentence, the guide on a platform called Eligiendo Mi Camino (“Choosing my Path”), built to help students figure out what they want to do next and to answer the questions that usually go unanswered. Over the next decade, 1.2 billion young people in developing countries will reach working age. How well they move from school to work depends, in part, on whether anyone answers those questions.
Generic information is not enough
Ask a Lima public school classroom who knows what they want to study, and the answer is sobering: one teacher in the program counted five or six students out of 27. Many would be the first in their families to reach higher education, so the people they trust most cannot describe the paths ahead. The school rarely can either: in Peru, career guidance is folded into a weekly tutoring hour led by classroom teachers rather than dedicated counselors, part of what the OECD describes as “patchy” school career guidance across Latin America.
Economists have tried to fill the gap with information. Students often misjudge the returns to education, and correcting those beliefs can shift behavior (Jensen 2010; Hastings, Neilson and Zimmerman 2015). But information alone has had mixed results, likely because information arrives generic while the questions are personal: which careers fit her interests, what they pay and cost, and what her first step should be. What moves decisions is personal: intensive mentoring works where information and even cash incentives do not. However personal guidance is costly to scale. It reaches families who can pay for it. In a public school, it barely exists.
The intervention: personalization at near-zero marginal cost
The AI career coach is one of two components of Eligiendo Mi Camino, built with Lima’s regional education authority (DRELM), the Peruvian startups uDocz and Aidea, and AI lab Anthropic; the other component is an AI math tutor. Since April and running through July, teachers and the AI career coach have taken final-year secondary students through an eight-step journey built around three questions: who am I, what are my real options, and what is my concrete plan. It draws on salary data for more than 130 occupations across twelve sectors. In addition, the AI coach interviews each student about their interests, and doubts, corrects the myths they hold, and, because family expectations can be the hardest obstacle, lets them prepare for that conversation before having it at home. One student arrived wanting something that combined medicine and technology and left having discovered that biomedical engineering existed. The platform not only provides students with information but also answers their questions.
What was expensive about counseling was never the information; it was the professional’s time to personalize it. That is the part AI now covers, at about 0.24 dollars per student per month, roughly one dollar per student for the four-month program beyond fixed setup costs. In Lima the coach ran inside the school day, in computer labs schools already have, led by teachers who received a two-day training using structured material and continuous support to conduct the program. No new staff, no new devices, no extra instructional time.
Preliminary results
We evaluate the program in a pre-registered randomized trial across more than 100 public schools; to our knowledge, it is among the first randomized evaluations of an LLM-based career coach in a developing country. Treated schools received the coach, control schools carried on as usual. The results are preliminary as the program is still ongoing, covering over 4,500 students, and measure what students know, believe, and plan.
Take-up is high: Nearly nine in ten treated students are using the platform. Treated students score 0.12 standard deviations higher on a career-readiness index, an average of five measures: how certain they are about which career to pursue, how informed they feel about their options, how feasible they judge their plans, how well they know the jobs in their preferred sector, and how clearly they see their own strengths and interests.
The clearest movement is in beliefs about the technical track. Quality short-cycle programs deliver solid labor market returns in the region, yet students undervalue them: control students estimate a two-year technical degree pays about S/1,300 a month, against the roughly S/1,922 that workers with technical education actually earn, according to INEI. The coach moves treated students’ estimates up by roughly 21 percent, toward the truth. Plans follow beliefs: intentions to enroll in a technical institute rise by 14 percent. These are beliefs and plans, not yet enrollment decisions.
The binding constraint for scale is connectivity At a dollar per student, the program fits any ministry budget, and because its guidance is data-based, it travels: load local salaries and options and it works in any labor market. Peru is waiting on this evaluation to decide on scale; Ecuador is adapting the platform. But the coach runs on a cloud-based model, and connectivity is far from universal. So we are exploring a small language model that runs locally, without internet.
That is what the World Bank’s jobs agenda asks of education systems: better transitions from school to work, not just better test scores, especially for the young people the labor market currently skips. In the students’ own testimonies, the transition already sounds different. As Jeffrey put it: “I see my future with hope.”
read more: https://blogs.worldbank.org/en/latinamerica/from-learning-to-earning-an-ai-career-coach-for-students-choosing-what-comes-next
