By Juan Herbozo Echevarría

In PISA 2025, 54% of students in Spain reported using AI chatbots to help them learn at least once a week, compared with an OECD average of 46%. That frequency raises a classroom question: what should students learn once they know how to ask AI for an answer? OECD, PISA 2025 country note on Spain.

Imagine a class wants to help other students find answers about the school library. The first question seems simple: “How many times can I renew a loan?” A chatbot says twice and adds that renewals can be done online. Both details sound plausible. Where did they come from?

Use the tool with a purpose

The students start by asking the question and providing their own library’s rules. They can ask the chatbot to explain the policy in plain language and point to the passage that supports its answer.

Learning to write a prompt now serves a concrete purpose. “Can I renew a book?” may produce a generic answer. If students mention that the book is due tomorrow and someone else has reserved it, they can see whether those details change the answer and whether the library rules cover the situation.

Understand what supports the answer

Next, the class checks the answer against the library rules. Students also test what happens when the section on renewals is missing or when two versions give different deadlines. If the chatbot cites a rule nobody can find, students have to explain what they checked and what question remains open.

The exercise shows how sources and context influence what the system generates. It also teaches students to distinguish a well-written answer from one supported by evidence. That distinction matters whether they are studying with a chatbot or reviewing a prototype they built themselves.

Build a system you can test

The next step is to design a small library assistant. The class decides which documents it can consult, what it should cite in each answer, and when it should refer a question to the librarian. To test it, students can write questions that the rules answer, cases the rules leave open, and questions that depend on information the assistant lacks, such as whether a book is currently reserved.

The first prototype can be a sketch on paper. A class with programming experience can later turn it into a simple application. Either way, students can show what they have learned by explaining why the assistant responded as it did, which source it used, and how they would improve its behavior after a mistake.

What I want to explore through Khipuwasi

Through Khipuwasi, I’m interested in bringing systems architecture into problems people can relate to. The library exercise exposes decisions that often sit behind the interface: what information the AI receives, how it uses that information, and how someone can verify the result.

I would start by teaching every student how to check an answer. Then I would introduce small projects where students can decide how a tool should work and test whether it does what they expect. Asking good questions, evaluating answers, and building solutions are skills students can practice in the same exercise.

If a student already knows how to ask AI for an answer, what should they learn to do next?