Pick the right CrewAI Process for a fan-out research pipeline and defend the choice

**CrewAI** ships two main Processes. Sequential runs Tasks in order, like a checklist. Hierarchical spins up a manager LLM that reads the goal and picks which agent does what next. They are different tools. For a pipeline with a fixed shape, four researchers feeding one synthesizer, you already know the assignment graph. There is nothing for a manager to figure out. Sequential is the right pick, and it is cheaper because there is no extra LLM call per routing decision. The catch is that Sequential, by default, really is sequential. Task 1, then Task 2, then Task 3. To make the four researchers run at the same time, you set `async_execution=True` on each of them. CrewAI dispatches all four in parallel. The synthesizer Task must stay synchronous so it waits for the four to finish before running, and you wire its inputs with `context=[web_task, arxiv_task, docs_task, news_task]` so its prompt actually receives all four outputs, not just the previous Task's output.

**Pick Sequential**, not Hierarchical. The graph is fixed in your code (four researchers, then one synthesizer), so a manager LLM would only add cost and a routing-failure mode without any decision to make. The trick is that vanilla Sequential is serial. To get true parallel fan-out, set `async_execution=True` on each of the four researcher Tasks. CrewAI then dispatches them concurrently. The synthesizer Task must stay synchronous (`async_execution=False`) so it blocks until the four return, and you must wire the dependency explicitly with `context=[web_task, arxiv_task, docs_task, news_task]` on the synthesizer. Without `context`, the synthesizer only sees the immediately preceding Task's output, not all four. Two practical caveats. CrewAI enforces that any synchronous Task with async predecessors must be the last in the list, so order the synthesizer accordingly. And async does not bypass provider rate limits, all four researchers share the same `llm_config`, so a tight rate cap can serialize them at the provider even though CrewAI dispatched in parallel.