Using agent as tool takes a long time for openai to route

API
1

Hey everyone,
So I’m using agent as a tool setup in my application to distribute task for each section.

My older setup was:

  • Master Agent (responsible only for conversational logic and routing to sub agents)
  • Twitter Agent (sub-agent-1): Has prompt to write tweet
  • Linkedin Agent (sub-agent-2): Has prompt to write linkedin post
  • Marketing Agent (sub-agent-3): Has prompt to write brand-positioning doc, blog and landing-page info

Now this setup was fine, starting to respond within the first 30s (first token stream). But it had issues that the marketing agent was outputing very poor quality brand-positioning doc, blog, landing-page info etc.

I figured that since the prompt for these were so detailed (in marketing agent), and all stuffed into one big prompt, let me try to re-structure it like this:

My newer setup:

  • Master Agent (responsible only for conversational logic and routing to sub agents)
  • Twitter Agent (sub-agent-1): Has prompt to write tweet
  • Linkedin Agent (sub-agent-2): Has prompt to write linkedin post
  • Marketing Agent (sub-agent-3): THIS IS NOW A ROUTER TO 3 FURTHER SUB AGENTS
    -     Blog Agent (sub-agent-3-1): Prompt to write blog
    • Brand positioning agent (sub-agent-3-2): Prompt to write doc on brand
    • Landing page agent (sub-agent-3-3): Prompt to write doc on landing-page

Now the outputs from this is drastically better, but the time has shot up to around 2 minutes for a response.

My questions are:

  1. is this the correct way to setup agent-as-tools?
  2. why has the time gone up drastically? it gives a poor UX.. what are my other options for this?

Thanks!

2 Likes
2

I’m using the openai js sdk

3

Thanks for taking the time to flag. I’ve raised this.

4

Thanks! Is there a workaround for this?

5

Hey, Thank you for flagging the issue. Can you please share the latest request id (<3 days) so we can review in the backend what is causing this issue for you. Thank you!

1 Like