You already know: the autonomous software engineer (long agent loops, budgets, caching) and long-form synthesis (notes, citations, consistency). This case adds parallel agents: when it helps, what it costs, how agents share work without duplicating it, and how to stay safe on an open web that is actively trying to manipulate them.
How to read this lesson. No code, on purpose. Numbers marked assumed are assumptions you state out loud; prices are illustrative.
1. The prompt, and the questions that change the design
A mid-level answer: "a planner agent and some worker agents". A staff candidate asks harder questions: when is multi-agent worth its extra tokens, how do agents avoid duplicating each other, how does the lead keep its context small, and how do you trust what agents read on the open web.
Assumptions you state: