Talk: Vibe Signage
Question
Could the orchestrator, at the end of the loop, assess whether the delivered application is in line with the (functional) requirements?
Asked by [email protected] @alkistig5705330171
Answer
Yes and that's the logic of our implementation. The orchestrator is the supervisor that talks to all sub agents, so ofcourse it can check the deliverable since it has access to all tools. At the end is the one that will say the final okay. The last agent on the workflow reports straight up to the orchestrator. We use sub agents because we can use specific models for specific work with the effort that we want. That helps mostly for job scoping and hallucination limitations. Someone could argue that having a model with a big "playbook" is better so you don't context switch between models. But I found the sub agents architecture safer.
Answered by Evryviadis Liapis