diff --git a/pyproject.toml b/pyproject.toml
index a1eeee12..eaf8b48c 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -43,6 +43,8 @@ dependencies = [
     "duckdb>=1.4.3",
     "pandas>=2",
     "async-lru>=2.0.5",
+    "apscheduler>=3.10.0",
+    "fastapi-sso>=0.10.0",
 ]
 
 
diff --git a/src/agentics/core/agentics.py b/src/agentics/core/agentics.py
index 87b6bc9b..6514c826 100644
--- a/src/agentics/core/agentics.py
+++ b/src/agentics/core/agentics.py
@@ -93,7 +93,7 @@ class AG(BaseModel, Generic[T]):
         None,
         description="""Python code for the used type""",
     )
-    states: List[BaseModel] = []
+    states: List[BaseModel] = Field(default_factory=list)
     tools: Optional[List[Any]] = Field(None, exclude=True)
     transduce_fields: Optional[List[str]] = Field(
         None,
@@ -138,11 +138,11 @@ class AG(BaseModel, Generic[T]):
         20,
         description="The size of the bathes to be used when transduction type is amap",
     )
-    areduce_batches: List[BaseModel] = []
+    areduce_batches: List[BaseModel] = Field(default_factory=list)
     save_amap_batches_to_path: Optional[str] = None
 
     crew_prompt_params: Optional[Dict[str, str]] = Field(
-        {
+        default_factory=lambda: {
             "role": "Task Executor",
             "goal": "You execute tasks",
             "backstory": "You are always faithful and provide only fact based answers.",
@@ -1063,21 +1063,26 @@ def product(self, other: AG) -> AG:
 
     def merge_states(self, other: AG) -> AG:
         """
-        Merge states of two AGs pairwise
-
+        Merge states of two AGs pairwise.
+        
+        The 'other' AG's fields take precedence over 'self' fields when there are conflicts.
+        This ensures that newly generated data (other) overwrites existing data (self).
         """
         if len(self) == len(other):
             merged = self.clone()
             merged.states = []
             merged.explanations = []
+            # Put self.atype first so its fields appear first in the merged schema,
+            # but other.atype values will overwrite in the instance creation below
             merged.atype = merge_pydantic_models(
                 self.atype,
                 other.atype,
                 name=f"Merged{self.atype.__name__}#{other.atype.__name__}",
             )
+            # Create instances with self first, then other overwrites
             for self_state, other_state in zip(self, other):
                 merged.states.append(
-                    merged.atype(**other_state.model_dump(), **self_state.model_dump())
+                    merged.atype(**self_state.model_dump(), **other_state.model_dump())
                 )
             return merged
         else:
diff --git a/src/agentics/core/mellea_pydantic_transducer.py b/src/agentics/core/mellea_pydantic_transducer.py
index 95774861..260539b0 100644
--- a/src/agentics/core/mellea_pydantic_transducer.py
+++ b/src/agentics/core/mellea_pydantic_transducer.py
@@ -65,13 +65,21 @@ async def structured_decoding_using_mellea(
     # At this point we're outside the `with`: session should be closed
 
     if mellea_output is None or mellea_output.content is None:
-        return targetAtype()
+        logging.warning(
+            f"Received None or empty response from LLM call. "
+            f"LLM: {llm}, Input length: {len(input)} chars. "
+            f"Check LLM configuration, API credentials, and model availability."
+        )
+        raise ValueError(
+            "Invalid response from LLM call - None or empty. "
+            "Please check your LLM configuration and API credentials."
+        )
 
     raw = mellea_output.content
     # If Mellea gave us a dict-like object:
     if isinstance(raw, dict):
         return targetAtype.model_validate(raw)
-    # If it’s a JSON string:
+    # If it's a JSON string:
     if isinstance(raw, str):
         return targetAtype.model_validate_json(raw)