Aula de cursos

Goal: build the first step of our adapted method: specifying what the user is really trying to decide.

A vague request hides conflicting aims

“Help me save money” sounds straightforward. Add “My income is irregular, and I cannot cut food, essential transport or legal obligations,” and the problem changes. The model now needs a sustainable strategy rather than an assortment of generic tips. Your restrictions are not noise; they help define the problem.

Prepare a problem sheet containing decision, unit of analysis, time horizon, hard constraints, success criteria and unknowns. Be explicit about whether the system has access to records, can perform calculations and has permission to act. Never infer that it has inspected information merely because it describes it confidently.

A good initial prompt

Act as an analytical partner, not a decision-maker. The situation is [context]. My goal is [result]. My constraints are [resources and obligations]. My time horizon is [period]. First restate the operational problem, separate established facts from assumptions and ask up to three essential clarification questions. Do not propose solutions yet.

2026 laboratory

Ask an assistant “Where am I losing customers?” without data. Record whether it recognizes its inability to observe actual customers. Repeat using a clearly fictional table of losses by channel and month. Check which conclusions change because evidence was added rather than because the prose became longer. Inspect what tool was really used; a described calculation is not necessarily an executed calculation.

Deliverable: a problem definition and a map showing inputs, operations, findings, unknowns and review responsibility. Avoid pasting identifiable customer information.