Unexpected multiple LLM calls after 25D upgrade in Oracle AI Agent Studio

Summary:

After the 25D upgrade, a simple supervisor agent makes 17 LLM calls for a single “Hi” input. These many calls causes high latency and renders the agents less useful. I’d like to confirm if this behaviour is expected in 25D or if it’s a regression.

Content (please ensure you mask any confidential information):

After the recent 25D upgrade, we have noticed a major change in how agent teams execute LLM calls for a supervisor agent type.

I have a supervisor agent managing two very simple worker agents:

• Worker Agent 1: Returns the current time from context.
• Worker Agent 2: Displays general AI agent capabilities.