Multiple Crews for tackling different tasks

Crews
1

Hi,

I wanted to know something which might feel a bit trivial but I wanted to reinforce my understanding.
I am working on a project and I have divided my framework in 3 different crews.

  1. Crew with 1 agent and 1 task which mainly works on gathering context
  2. Crew with 2 agent and 9 tasks. One agent generates content and the other basically reviews and re-writes the content if needed. The output of in between tasks is pydantic. Then it stores the data in DB
  3. Last Crew with 1 agent and 1 task mainly focused on selecting best content based on some Quality Assurance Criteria. Loads data from DB pass it to agent as context and the output is generated in pydantic and again stored in DB

My main Question is that if there is a logic or rationale around choosing number of crews or agents. What are advantages of using separate crews and multiple agents within same crew. and what are the disadvantages?

Any guidance will be appreciated. Thanks!

2

Hi
I try to start simple and only add complexity as I see mistakes or hallucinations creep in. So try with a simple crew and if it mis-behaves try different LLMs / crew structures till you find the one that works.

For example right now for me GPT 4.1 behaves the best, but that might change with GPT-5 in August.

3

Thanks! I agree with starting simple.

4

One more question on Async tasks execution, I have an agent which has independent 8 tasks. I tried to execute those tasks in parallel with async = True. It was running fine but then something weird started to happen. The Agent started to miss some tasks and started to duplicate others.
In sequential async = False, it runs fine. But I wanted to improve the speed but async = True causes above mentioned problem.

Do you think the issue might the agent LLM? I am using GPT-4o