Lesson 6 — Parsing and validating model output

Never trust the shape

Models make mistakes. Even when you ask for JSON, you may get chatty text, a list instead of a dict, or a dict with a key missing. Parsing means turning the text into data; validating means checking the data has the shape you need before using it.

A good parser answers with a clear error instead of crashing:

1. Is it JSON at all? (json.loads raises json.JSONDecodeError if not)
2. Is it a dict? (isinstance(data, dict))
3. Are all required keys there?

Why it matters: an agent that crashes on one bad reply is useless. An agent that says "error: missing args" can tell the model and ask again — you will do exactly that in Lesson 21.