I am experimenting little bit with Agent and langchain4j.

I have one Agent with the following snippet in System Prompt.

Output format contract (MANDATORY):

You must return a single valid JSON object and nothing else.
Do not include markdown, prose, headings, explanations, or formatting.
Do not wrap the JSON in code fences.

The JSON structure must be exactly:

{
  "offers": [
    {
      "hotelId": string,
      "hotelName": string,
      "roomDescription": string,
      "totalPrice": string,
      "currency": string
    }
  ]
}

Rules:
- The root object must contain only the "offers" key.
- "offers" must be an array.
- Each array element represents exactly one room offer.
- All fields are mandatory.
- Values must be copied verbatim from tool responses.
- Do not add, remove, rename, or nest fields.
- Do not include null values.
- If no offers are available, return:
  { "offers": [] }

Any response that is not valid JSON matching this schema is invalid.

with that Agent returns a valid JSON as I expected.

But I know that langchain4j have an Structured Output functionality so I implemented the following

@AiService
public interface HotelOffersExtractor {
    HotelOffersResponse extractOffersFrom(String text);
}

HotelOffersExtractor hotelOffersExtractor = AiServices.create(HotelOffersExtractor.class, chatModel);
HotelOffersResponse response = hotelOffersExtractor.extractOffersFrom(result.content());

but by looking in depth, I saw that this cause another call to the Model API, isn't this a bad idea, while in production environment this will cause me more money I think.

So if my first solution in System Prompt works, does it make sense to use Structured Output feature of langchain4j.

Thx for answers.