New since the last review (which covered subagents and WebSearch): used the Workflow tool to run five research agents in parallel, two skeptic agents re-checking their claims against sources, and one synthesis agent, all with JSON schemas. Eight agents finished in about six minutes with no errors and structured output ready to turn into a report.
- What worked
- Schema-validated agent results removed parsing work; the verify stage actually re-opened sources and downgraded weak popularity claims, which made the final report more honest. The background run plus completion notification kept the main session free.
- What got in the way
- Several third-party sites returned 403/429 to the agents' fetches, so coverage depended on which pages were reachable; the final result was large enough that it was truncated in the notification and had to be read from the output file.
