Worked examples
Each example is a short script a coding agent can follow: the files, the calls in order, the steps only a person
takes, and how to switch everything off. Each one was run on preview; its footer says when and which calls ran.
They are the same pages the Studio MCP serves (studio_guide("example-...")).
| Example | What it builds |
|---|---|
| KPI function and app | a code function on a declared query, and a hand-built app with two tiles |
| Ask AI summary | an llm function that turns a small table into a headline and bullets |
| Classify tickets | a classify function that sorts tickets into four labels |
| Agent with a tool | an agent function that may call one code function |
| Weekly email | a workflow whose weekly digest email a person approves |
| Train, then predict | two functions sharing a model through workflow blobs |
| Backtest | a function step mapped over partitions, reduced by SQL |
| Publish handoff | a token builds an app version; a person publishes it |
| What changed, and why | Detect and Explain on an app's model |
| Chat that answers from your functions | an app's chat calling your KPI function |
What you, the person, do
- Answer the agent's questions first: the decision, and how each number is defined.
- Check one number you already know before trusting any screen.
- Confirm each publish at the link the agent gives you; approve each email in Studio's Approvals.
- Ask the agent for the list of everything it made and how to switch each off.
Build on preview first. On preview, email goes only to the deployment's allowlist.