Do you run multiple coding agent? How do you pick who gets which task?

I'm trying to understand how people that use multiple coding agent (Claude, Codex, Cursor, Gemini CLI, Copilot, OpenCode, Aider, etc) actually split the work between them: Which agents do you run, and on what kind of work? How did you decide which one got that task? Have you caught an agent saying a task was done when it wasn't? How did you find out? Do you track which agent handles which kind of task well? Have you stopped using certain agent for certain work? What happened to make you stop? Has an agent ever sent something somewhere it shouldn't have (a key, an address, private code)? What happened? I'm researching this for an open-source project to help us help ourselves and not get locked into big company harnesses. I'll post a summary of this thread if I get good feedback :D Thanks!!!