AI Agent JSON Mode Output 2026

Natural language output is difficult for programs to parse. "Help me parse the user's intent" — have the Agent return JSON, and the program can use it directly without writing regex.

💡 What You Will Learn

Natural language output is difficult for programs to parse. "Help me parse the user's intent" — have the Agent return JSON, and the program can use it directly without writing regex.

JSON
{"intent": "", "entities": {"key": "value"}, "confidence": 0.0-1.0}

from pydantic import BaseModel

class UserIntent(BaseModel):
    intent: str
    entities: dict
    confidence: float

# schemaAgent
schema = UserIntent.model_json_schema()

{
  "intent": "book_flight",
  "entities": {
    "destination": "",
    "date": "2026-07-17",
    "time_period": ""
  },
  "confidence": 0.92
}

Summary