Best-practice pattern for sequencing multi-stage conversational Agents

[sivan-spiritt]

I’ve got a 4-stage conversational flow—each stage handled by its own Agent. Initially, I structured it as follows:

root_agent
   ├─ stage1_agent
   ├─ stage2_agent
   ├─ stage3_agent
   └─ stage4_agent

However, in practice, the root_agent only forwards the first user input directly to the stage1_agent, creating redundancy.

To address this, I collapsed the structure, making the initial agent itself the root:

stage1_agent (acting as root)
   ├─ stage2_agent
   ├─ stage3_agent
   └─ stage4_agent

While this simplifies execution, it feels somewhat odd, as the first stage is no longer a sibling to the others but now acts as their parent.

Any guidance or best practices here would be greatly appreciated!

[suhasygh]

Spitballing here but would the following approach work? Each stage agent is defined as a subagent for the previous stage

• Root Agent
  • Stage 1 Agent
    • Stage 2 Agent
      • Stage 3 Agent
        • Stage 4 Agent

[sivan-spiritt]

Hey, thank you for answering!
It might be actually better, I'd give it a try.
I wonder if they will still be able to call each other in case of needing to go back to prior stages (in the implementations I shared, all agent could transfer to all other agents)