Lab 11
Workflow Lab
Prompt, tokens, context, model, agent, tool, data, answer — assembled into the single pipeline that sits behind a working AI feature.
Educational simulation. A scripted end-to-end walkthrough over four rows of made-up sales data. The tokenisation is real, the tool call and query result are fixed sample data, and no model is called at any point.
The request
“Analyse the sales records and tell me which region is falling behind.”
The pipeline
Stage 1 of 9
Stage 1 · Prompt
A person asks for something
One sentence of intent. Everything downstream is shaped by how well this was written — vague in, vague out.
Reading the result
- What just happened
- One sentence of intent. Everything downstream is shaped by how well this was written — vague in, vague out.
- Why it matters
- Each stage is a place where something can go wrong — a badly written prompt, a window that overflowed, a tool that returned the wrong rows. Knowing the stages is how you debug an AI feature instead of guessing at it.
- Where you meet this
- Any product that answers questions about your own data runs this pipeline, whether or not its documentation describes it this way.
Go deeper
Prompt Playground
The lab that covers this stage on its own.