Structured Generation: teaching AI agents to color inside the lines

In the previous article, we explored generating free-form text in a workflow, as well as dividing responsibility for different parts of a task among agents. This time, let's look into generating machine-readable structured data. 💡 tip Skip to the action if you're already familiar with structured data and schemas. Motivation Why would we want data to be structured? First, it is easier to filter, transform and combine documents with automated tools when we know ahead of time the shape of responses and what properties they can contain. For instance, if we had to sort and organize thousands of profiles in unstructured text: "John was born twenty five years ago and programs Python" "Alice is a cryptography expert born in 1998" etc. Using traditional text-based tools there are an uncountable number of permutations, phrasings, exceptions and edge cases to consider. Instead, by using a language model to transform texts into structured data, we could use simple operations to fill in missing dat