Resisting Multi-Agent for a Document-Summarization Feature

Product wants a feature: given a long internal document (up to ~50 pages), produce a structured summary with key points, action items, and open questions. An engineer proposes a multi-agent design: one agent extracts key points, a second extracts action items, a third extracts open questions, and a fourth "supervisor" agent merges their outputs — arguing that "each agent can specialize and they can all run in parallel, so it'll be faster and better."

1. Evaluate this proposal against the three concrete reasons to use multiple agents. Does this task actually meet any of them?
2. Propose a single-agent (or non-agent) alternative and explain why it's likely to perform as well or better.
3. Under what change to the task would the multi-agent version become genuinely justified?