How would you build a multi-agent software development pipeline?

Hey everyone, I’m trying to build a software development pipeline using multiple AI agents, but I’m a bit lost with concepts like agent harnesses, loops, graphs, orchestration, and so on. What I basically want is a structure where I can call one main agent , give it a project or goal, and have that agent orchestrate other specialized agents automatically. For example: Main Agent / Orchestrator → Product/PRD Agent → Architecture Agent → Database/Schema Agent → Frontend Agent → Backend Agent → Testing/QA Agent → Security Agent → Documentation/Context Agent Ideally, these agents would share context and continue each other’s work instead of behaving like completely isolated sessions. Is there already a framework, harness, or architecture that works well for this? Would you recommend using something like graphs/workflows, agent loops, subagents, or building a custom harness? Any open-source projects or examples I should look at would be greatly appreciated.