Overview
End-to-end walkthroughs. Each starts from a real goal and ends with something working.
| Example | Shows | Audience |
|---|---|---|
| A seller’s Account 360 | Curating an entity so an agent can brief a rep before a call | Admin / RevOps |
| A scoped domain tool | Two views of one object, split by domain, deployed as its own tool | Admin + developer |
| An Apex enrichment provider | Computed context no configuration could produce | Developer |
Working through them
The examples stand alone, but they escalate: the first is configuration only, the second adds a domain and deploys generated code, the third writes an Apex class. Each names the decisions it makes, because deciding what an agent should see is the real work.
Best practices
The pattern all three examples share:
- Start from a question someone actually asks. “What should I know before this call?” tells you what to include; “which fields should I include?” doesn’t.
- Include what changes an answer. If a different value wouldn’t change what the agent says, the field is costing tokens.
- Explain what isn’t obvious. Anything a new hire would misread needs an instruction.
- Prefer a number to a table. “34 open cases” beats 34 rows at a fraction of the cost. → Metrics
- Check it as the agent sees it. Preview at every level, then watch Quality once real traffic arrives.