We are observing significant performance degradation when executing complex tasks within the Google ADK framework. In several scenarios involving multi-step orchestration and cross-agent communication, the total execution time reaches 6-7 minutes. This duration is far beyond acceptable limits for interactive or automated workflows and negatively impacts the user experience and system reliability.

Observed Behavior

Complex tasks requiring multiple agent hand-offs or sequential tool calls experience high latency.

Total processing time consistently peaks between 6 and 7 minutes for high-complexity requests.

The system remains in a "processing" state for an extended period, which can lead to timeouts in upstream services.

Additional Context

This issue has been identified during internal testing and discussions regarding the orchestration of complex task sequences. While the individual agents perform their specific functions, the cumulative overhead of the current execution flow for complex tasks is the primary concern.

• Agents: 31
• Arch: both sub-agents and agent-tools
• Specialized agents: both sequential and parallel agents used

Question

Did you face with such problem? What information or recommendations do you have to optimize the performance?