1000s of agents with fan-out, fan-in as a workflow as agent that aggregating results to follow up with user or determine which agent can handle the request

[jltnf54]

We have a use case where we have 1000s of agent in an agent registry. Each agent is an intake agent that specializes in self-assessment (these agents are not developed by our team and can be a2a agents, foundry agents, chat agents, etc..) take a user prompt to ask to follow up questions based on differnent products, categories, etc.... We have a workflow as an agent that pulls these thousands of agents and fans-out with what the user has asked. We then fan-in the results and take all agent response to determine what questions to ask the customer to determine which agent can actually handle the request. Do you have any recommendation on how to approach this challenge so we can scale?