diff --git a/architecture/agents.md b/architecture/agents.md
index 5f17dd37d3..625d4ccf59 100644
--- a/architecture/agents.md
+++ b/architecture/agents.md
@@ -25,6 +25,8 @@ Preferably all in a single place.
 class BaseAgent:
     def agent_primitives(self) -> list[BaseAgent]:
         # Returns a list of Agents that are utilized by this agent to generate inputs
+        # We use agent primitives here instead of subagents because these are going to be part
+        # of the message graph, not a subagent tool call.
         raise NotImplementedError
     
     def tools(self) -> list[BaseTool]:
@@ -37,7 +39,7 @@ class BaseAgent:
         tools = self.tools()
         for agent in self.agent_primitives():
             tools.extend(agent.tools())
-        tools = set(tools)
+        tools = remove_duplicates(tools)
         tools = initialize_tools(tools, config)
         return self(llm, tools, config, *args, **kwargs)
     
diff --git a/architecture/message_graph.md b/architecture/message_graph.md
index d81b054d4e..251a3a416a 100644
--- a/architecture/message_graph.md
+++ b/architecture/message_graph.md
@@ -65,6 +65,7 @@ and how this ends up parsing into on policy training data, if requested.
 Edges are the connections between nodes, and there are two types we are concerned with:
 - **Sequential edges**: These represent the flow of conversation, connecting messages in the order they were sent. For example, a user message followed by an assistant response.
 - **Parallel edges**: These represent versioning, e.g. edit history, context squishing, etc.
+We, however, are only concerned about parallel edges when we break the prefix, and ignore any other parallel edges.
 
 ## So what does this look like in practice?