Extraction agent does not use strict structured output mode by default because most likely the user-provided schema won't be supported since there are many requirements such as:
Every property needs to be required, and schema has to be written a specific way for workarounds.
No more than 500 enums total in the entire schema.
Schema cannot contain a lot of properties allowable in the JSON schema specification.
Using structured output mode can necessitate quite a bit of schema wrangling first to have a supported schema, and so extraction agent doesn't make the assumption that we have a supported schema. But perhaps we should analyze the schema or use a test response to turn on/off strict structured outputs in the case where we can and should use strict structured outputs.
Extraction agent does not use strict structured output mode by default because most likely the user-provided schema won't be supported since there are many requirements such as:
required
, and schema has to be written a specific way for workarounds.Using structured output mode can necessitate quite a bit of schema wrangling first to have a supported schema, and so extraction agent doesn't make the assumption that we have a supported schema. But perhaps we should analyze the schema or use a test response to turn on/off strict structured outputs in the case where we can and should use strict structured outputs.