Structured messages, not raw output
When one agent hands work to another, what gets passed between them matters as much as which pattern you're using. It's tempting to just let one agent read another's raw free-form output, but ambiguity in that output compounds badly once you have more than one hand-off in a chain. If a research agent's summary is vaguely worded, a writing agent built on top of it inherits that vagueness and can easily misinterpret what was actually found.
A more reliable approach is to define a structured message format between agents, typically capturing at minimum what was asked, what was done, and what the result was. For example a worker agent might return `{ task: "summarize competitor pricing", actions_taken: [...], result: {...}, confidence: "high" }` rather than a paragraph of prose. Structure forces the sending agent to be explicit, and it gives the receiving agent (or a supervisor) something predictable to parse instead of having to interpret free text.