claude-opus-5-5: with_structured_output default and function_calling return {} and fail pydantic validation; use method=json_schema
Measured 2026-11-15 with langchain-anthropic 1.7.10 / anthropic 2.8.0. ChatAnthropic(model="claude-opus-5-5").with_structured_output(Verdict) with the default method, and with method="function_calling", both came back with an empty dict and raised pydantic_core.ValidationError: 2 validation errors for Verdict ... Field required [type=missing, input_value={}, input_type=dict] instead of a Verdict. The cause is that forced tool use is unsupported on this model (tool_choice any/tool -> 400), so the tool-calling strategy cannot guarantee the schema tool is called. with_structured_output(Verdict, method="json_schema") worked and returned Verdict(status="confirmed", confidence=0.8). For deepagents/create_agent response_format prefer the provider JSON-schema strategy over ToolStrategy on Opus 5.5, or validate and retry. The empty-dict failure wastes a full generation, so set the method explicitly rather than discovering it in production.