RESEARCH, WRITING, AND KNOWLEDGE AGENTS

EXAMPLES Example 1: A deep-research agent. A user asks for a briefing on the risks of using agents in hospitals. The agent breaks the question into sub-questions about privacy law, patient safety, error rates, and staff training. For each, it searches, reads many sources, and records notes in a file listing the source and the key finding. It then assembles a structured, multi-section report instead of a single quick answer, and it notes where sources disagree. Example 2: A document generation agent. The research agent hands over a structured list of findings, each with a source. A separate writing agent turns the findings into a polished report with an executive summary, headings, and a clear narrative. A style rule requires plain language and short paragraphs. Because the writer works only from the findings, a reviewer can compare the report against the findings and confirm that nothing was added. Example 3: A citation requirement. Every factual claim in the final report carries an inline citation naming the source and section. A reviewer reading the claim that the average response time fell by a certain amount clicks the citation and sees the exact paragraph. The reviewer can confirm the claim in seconds, rather than having to trust the agent blindly. Example 4: A source-grounding failure. An agent writes a confident sentence claiming that a regulation requires monthly audits, but the source registry shows no source supporting it. An automated check flags the sentence and sends the draft back. The agent searches again, finds that the regulation actually requires annual audits, corrects the sentence, and cites the real source. Without the check, the false statement would have been delivered with full confidence.