Module 2 · Making agents dependable · Lesson 4

Structured outputs and grounding

Make an agent return typed, schema-validated JSON in which every finding carries an exact quote that your code can check against the source document.

75 min 1 video 3 exercises 7-question quiz Answers

Setup for the exercises:Install Claude Code Node.js, pnpm and TypeScript Get an OpenRouter key OpenRouter or a local model?Local models with Ollama

By the end of this lesson you can

Design a Zod schema for a title-review report that uses enums, required evidence and explicit 'not found' values
Get schema-validated JSON from Claude Code headless mode, and explain how the Agent SDK's outputFormat does the same
Implement a validate-and-retry loop and explain why schema validity does not imply correctness
Ground findings in evidence quotes and verify each quote against the source in code

An agent that reviews title registers is only useful if its findings can go somewhere: into the case-management system, onto a screen where a conveyancer ticks them off, into the report on title, or into an eval that checks them automatically. None of that works with a friendly paragraph that says "I noticed a few things worth mentioning". It needs structured output: data in a shape your code defines and checks. And because the reader is a solicitor who must rely on it, each finding also needs grounding: evidence from the document, precise enough that a machine can confirm it is really there.

Module 2 · Making agents dependable · Lesson 4

Structured outputs and grounding

Make an agent return typed, schema-validated JSON in which every finding carries an exact quote that your code can check against the source document.

75 min 1 video 3 exercises 7-question quiz Answers

Setup for the exercises:Install Claude Code Node.js, pnpm and TypeScript Get an OpenRouter key OpenRouter or a local model?Local models with Ollama

By the end of this lesson you can

Design a Zod schema for a title-review report that uses enums, required evidence and explicit 'not found' values
Get schema-validated JSON from Claude Code headless mode, and explain how the Agent SDK's outputFormat does the same
Implement a validate-and-retry loop and explain why schema validity does not imply correctness
Ground findings in evidence quotes and verify each quote against the source in code

Free · sign in to start

Sign in to keep reading.

Free, no card. We use your account to save progress, quiz scores, bookmarks and occasionally send updates from our team.

10 more sections
3 hands-on exercises
A 7-question quiz
Progress, quiz scores and bookmarks