Hello, I'm seeking suggestions for workflows. I've developed a customer support workflow that functions effectively. I'm currently considering whether it's more advantageous to utilize a single primary agent and invoke other workflows, or to branch out to multiple agents using a text classifier. I've experimented with a single main agent, but I consistently encounter difficulties with the agent calling the correct tools. Consequently, I've opted to consolidate all agents within the same workflow and implement explicit guidelines for the text classifier. Does anyone have any feedback on this approach? I'm eager to hear your thoughts and opinions.
Hello everyone,
It seems like this approach could be effective. Please review the following:
This appeared to function, but it's crucial to precisely define the agent's behavior and its available tools; it shouldn't be left to guess. Therefore, utilize system prompts and include all available tools. In my setup, the primary AI agent is aware that it can invoke sub-agents, and these sub-agents have workflows or other tools associated with them.
Best regards,
Samuel
Thanks for the response. I've set up workflows structured similarly, but I consistently face challenges getting the primary agent to invoke tools. Often, I utilize prompt generators to create highly precise prompts, yet the agent still struggles to call the appropriate tools. Do you have any recommendations? I'd prefer to proceed this way but frequently encounter this problem.
Which AI model are you utilizing?
Best regards,
Samuel
Just an AI Agent node utilizing an OpenAI LLM.