Hi again, theoretical question. If most use Docker/K8s to run dataflow on (each dataflow has own image, or each flow executes common image), why do we need Agent at all?

The agent is the done that spins up the new job and deploys the container/pod. The flows do not need to be using a common image and in general, the Agent process is the one that looks for Flows to run from Prefect Cloud. The agent and Flows do different things

got it, especially for Cloud part

this may also be useful