Stage 1 · Mental model
Read AI from zero, then learn token, context window, grounding and hallucination in the glossary. The goal is to understand what the model can see and what counts as evidence.
Stage 2 · Controlled requests
Use the prompt builder and practice turning a vague request into objective, context, constraints, output format and validation criteria.
Stage 3 · Real task
Choose one recurring task from your work or study. Test several representative examples and record failures instead of optimizing a single demo.
Stage 4 · Verification
Build a tiny golden set, compare before/after changes and decide whether the workflow is worth keeping. Only then move toward automation.