Most people build a finance agent and point it at one thing: tell me what to buy. Wrong job.
A research agent is not a forecaster. It is a librarian with a checklist. It gathers the evidence and proves it. It does not get to decide what the evidence is worth.
Four jobs to design into it, and one line you never cross:
-> Retrieve: pull from the primary source only. The 10-K, the transcript, the actual filing. Not a summary of a summary. -> Verify: every figure gets a source, a page, and a date, or it gets dropped. No citation, no claim. -> Challenge: make the agent argue against its own first answer. What would have to be true for this to be wrong? -> Escalate: when confidence is low or sources disagree, it stops and flags you. It does not guess to look helpful.
What the agent never does: pick a side, name a price, or tell you to act. The second it predicts, it stopped being a research tool and became a liability.
Why it matters. An agent that gathers and verifies makes you faster on the boring 80 percent. An agent that predicts makes you confident on the part you should be most careful about. Those are not the same tool.
Human judges. AI builds the file.
Save this and use it the next time you scope an agent.
Educational content only. Not investment advice, and not a recommendation to buy, sell, or hold any security. Wall Street Prompt. Always verify against the primary source.