Aren't multiple agents just one system multi-tasking?
A user asks whether five agents running many tasks is really just one system — for example Hermes with skills, workflows and SOPs — running jobs in parallel under an orchestrator. The distinction that matters is not the count but whether the parts can disagree: one writes, another checks, and neither grades its own homework. If it is all one model talking to itself, it is a workflow, and the multi-agent benefit only starts when a checker with different instructions catches what the worker missed.