Wall Street PromptBrand
All briefs
Wall Street Prompt · Brand

How To Build An AI Research Agent Without Making It A Stock Picker

Carousel — cover + one idea per slide + caption. This is a post, not a video script.

Carousel
Instagram carousel (+ LinkedIn document)Instagram carousel8 slides
References & validationValidated

Grounded in a real captured creator post (linked below).

Reference creator: AI Automation / Uppit AI (YouTube)

AI Automation / Uppit AI · YouTube postAI Automation / Uppit AI · YouTube postCole Medin · LinkedInAI Automation / Uppit AI · LinkedIn

Slides

One idea per slide

Cover
THE RESEARCH AGENT
THAT DOESN'T PREDICT

Four jobs it owns.
One line it never crosses.

Retrieve. Verify. Challenge. Escalate.
AI builds the file. You make the call.
(swipe)
Slide 1
THE RESEARCH AGENT
THAT DOESN'T PREDICT

Four jobs it owns.
One line it never crosses.

Retrieve. Verify. Challenge. Escalate.
AI builds the file. You make the call.
(swipe)
Slide 2
Everyone builds the same agent:
"Tell me what to buy."

Wrong job.

A research agent is not a forecaster.
It is a librarian with a checklist.

The common mistake:
you ask it to predict.
The failure mode:
it sounds confident on the
part you should doubt most.
Slide 3
DESIGN THE BOUNDARY FIRST

-> The agent gathers and proves
-> The agent never picks a side
-> No price targets, no buy / sell
-> No "this looks like a winner"

The second it outputs a call,
you built the wrong tool.

Human judges. AI builds.
Slide 4
JOB 1. RETRIEVE

-> Pull from the primary source only
-> The filing, the transcript, the comp
-> Not a summary of a summary
-> Log what it pulled and from where

If it cannot point to the source,
it did not retrieve. It guessed.
Slide 5
JOB 2. VERIFY

-> Every figure: source + page + date
-> No citation = the claim gets dropped
-> Cross-check numbers against the face
  of the statements
-> Tag each output: confirmed / unconfirmed

Unconfirmed is not an answer.
It is a flag.
Slide 6
JOB 3. CHALLENGE

-> Make it argue against its own answer
-> "What would have to be true
  for this to be wrong?"
-> Surface the strongest counter-read
-> Separate fact from inference, clearly

An agent that only agrees with itself
is a mirror, not a researcher.
Slide 7
JOB 4. ESCALATE

-> Low confidence -> stop and flag you
-> Sources disagree -> surface both
-> Missing data -> say missing, not zero
-> Anything near a judgment call -> hand up

The agent's best skill is knowing
when to stop and ask a human.
Slide 8
WHY IT MATTERS

An agent that gathers and verifies
makes you faster on the boring 80%.

An agent that predicts
makes you confident on the 20%
you should handle most carefully.

Those are not the same tool.

AI builds the file. You own the thesis.
Save this. Scope your next agent with it.

Educational only. Not advice.

Caption

Paste under the carousel

A finance agent should not predict. That one design choice decides whether you built a research tool or a liability.

Build it around four jobs, and one boundary it never crosses:

1. Retrieve: primary source only. The filing, the transcript, the comp set. Not a summary of a summary. Log where every piece came from. 2. Verify: source, page, and date on every figure, or the claim gets dropped. Tag each output confirmed or unconfirmed. 3. Challenge: make the agent argue against its own first answer. What would have to be true for this to be wrong? Separate fact from inference. 4. Escalate: low confidence, conflicting sources, or anything near a judgment call gets handed up to a human, not guessed.

The line it never crosses: no picks, no price targets, no buy or sell, no "this looks like a winner." The moment it predicts, it stopped being a research agent.

Why it matters. A gathering-and-verifying agent makes you faster on the boring work. A predicting agent makes you overconfident on the exact part that deserves the most care. Build the first one.

Human judges. AI builds the file.

Save this and scope your next agent with it.

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.

Design notes

Instagram 4:5 carousel, also exportable as a LinkedIn document. Consistent near-black background with a single green accent across all slides. Cover: heavy condensed all-caps headline, the four verbs on one line as a rhythm, small green '(swipe)' bottom-right. Body slides: top-left job label in green ('JOB 1. RETRIEVE') in condensed caps, then an arrow list in neutral sans, closing each slide with one sharp bold takeaway line set apart by a thin green hairline rule. Keep one idea per slide, lecture-slide density but never edge-to-edge. Slide 8 carries the implication, the 'AI builds / you own' principle, the save line, and the short disclaimer. Arrows as '->' in green. No em dashes, no gradients, no stock imagery, no charts.

CTA

Save this and scope your next agent with it.

Wall Street Prompt
Wall Street Prompt — internal