Educational only. Not investment advice.
All briefs
Wall Street Prompt · LinkedIn one-page
BUILD A PORTFOLIO MONITORING AGENT
LinkedIn one-page
A portfolio agent should not tell you what to buy. It should tell you what changed against your thesis, with the source.
LinkedIn1080 × 1350
References & validationExact post missing
Reference creator resolved, but the exact source post was not captured. Capture the specific post (not just the profile) before treating as validated.
Reference creator: Greg Isenberg / Riley Brown
Post content
Dense one-page content — sections, tables and frameworks
THE COMMON MISTAKE
Most people build the wrong agent. They ask it for a verdict instead of a signal.
- Mistake: you prompt the agent for buy, sell, or hold. It obliges. It has no edge and no accountability.
- Failure mode: a confident answer with no source trail. You cannot audit it, so you cannot trust it or correct it.
- Disciplined workflow: the agent watches your existing thesis and flags when a fact underneath it moves. You judge. The agent surfaces.
- Key: a monitoring agent is a research assistant, not a decision maker. Human judges. AI builds.
Why it matters: an agent that gives verdicts hides the reasoning you need to manage risk. An agent that surfaces changes makes your thesis falsifiable, which is the entire point of having one.
WHAT THE AGENT WATCHES
You do not monitor a ticker. You monitor the assumptions that justify the position. Define them first.
| Thesis assumption | Trigger to watch | Primary source |
|---|---|---|
| Revenue growth holds above plan | Reported growth deceleration or guidance cut | 10-Q, earnings call transcript |
| Gross margin is structurally stable | Margin compression two quarters running | Income statement, MD&A |
| Management keeps capital discipline | Buyback pause, dilutive raise, leverage jump | 8-K, cash flow statement |
| Competitive moat is intact | New entrant, pricing war, share loss | Comps, industry filings, call Q&A |
| Key person and governance stable | CFO or CEO departure, insider selling | 8-K, Form 4 filings |
Why it matters: if you cannot write the assumption down, you cannot monitor it. Vague theses produce noisy alerts. Specific assumptions produce signal.
THE SOURCE TRAIL STANDARD
Every alert the agent raises must carry its evidence. No source, no alert. This is the discipline that separates a tool from a rumor mill.
| Alert field | What it must contain | Why it is required |
|---|---|---|
| Claim | One sentence on what changed | Forces a falsifiable statement |
| Assumption hit | Which thesis line this affects | Connects the change to your decision |
| Source | Filing type, date, page or quote | Lets you verify in under a minute |
| Confidence | Reported fact vs inferred read | Separates data from interpretation |
| Materiality | Estimated effect on the thesis | Filters noise from substance |
Heads up: an alert without a source is not an alert, it is a guess. Build the agent to refuse to fire unless every field is populated.
BUILD WORKFLOW: FIVE STEPS
Construct it in layers. Each layer is testable on its own before you trust the chain.
| Step | Action | Output |
|---|---|---|
| 1. Encode thesis | Write assumptions and triggers per holding | A monitored thesis file |
| 2. Wire sources | Connect filings, transcripts, news feeds | A clean, dated input stream |
| 3. Detect change | Diff new data against last known state | Candidate change events |
| 4. Map to thesis | Match each event to an assumption | Scored, relevant alerts only |
| 5. Report with trail | Emit alert plus full source trail | An auditable IC-memo-style note |
Tip: build step 3 to compare against a stored prior state, not against your opinion. The agent detects drift in facts, not agreement with you.
VERDICT AGENT VS MONITORING AGENT
The difference is not cosmetic. It changes what you can defend in an IC meeting.
| Dimension | Verdict agent | Monitoring agent |
|---|---|---|
| Output | Buy, sell, hold | What changed vs thesis |
| Source trail | Usually none | Required on every alert |
| Auditability | Black box | Fully traceable |
| Failure mode | Confident and wrong | Quiet when nothing moves |
| Who decides | The model | You |
| Risk posture | Hides assumptions | Surfaces broken ones |
Why it matters: position sizing and risk management depend on knowing why you hold something. A verdict agent erases the why. A monitoring agent protects it.
GUARDRAILS BEFORE YOU SHIP
An agent that touches your portfolio process needs hard limits written in before you trust it.
- No recommendations. The agent reports change, never a trade action.
- Cite or stay silent. Unsourced output is suppressed by design.
- Separate fact from inference, and label every line as one or the other.
- Log everything. Every alert is timestamped and reproducible for review.
- Human sign-off. No alert changes a position without your judgment in the loop.
Key: the agent earns trust by being boring. It is silent when the thesis holds and loud, with receipts, when it breaks.
Caption
LinkedIn post copy
Visual design notes
- Background near-black forest green (#0B0F0D base) with a single teal-green accent (#1FB68A) used only for the keyword, the subtitle banner, and table header rules. No second accent color.
- Headline in heavy condensed sans (Anton or compressed Helvetica), all caps, left-aligned. Emphasize the word MONITORING in the teal accent while the rest of the title stays off-white.
- Subtitle sits in a thin teal banner bar directly under the headline, full content width, small caps or medium weight, single line wrapping to two max.
- All six sections left-aligned in a single column. Section headings in accent-underlined small caps; generous vertical rhythm between blocks so density reads as structured, not cramped.
- Tables are the visual backbone: dense 2 and 3 column layouts with thin hairline row dividers in muted green-gray, header row in accent, alternating row tint at 4 percent opacity for scanability.
- Include one small inline diagram beside the Build Workflow table: five stacked nodes (thesis to sources to detect to map to report) connected by a single vertical accent line, mirroring the five steps.
- Why it matters notes set in italic or a left accent-bar callout so they read as the teacher voice, visually distinct from body bullets.
- Footer locked to the bottom margin as a thin single line; WSP wordmark or small mark at top-right of the header. Keep 64px outer margins on 1080x1350 so it breathes on a phone feed.
Production checklist
- ☐Design the 1080x1350 one-pager in the WSP template: near-black forest background, single teal accent, heavy condensed headline with MONITORING emphasized, teal subtitle banner.
- ☐Build all five tables (What The Agent Watches, Source Trail Standard, Build Workflow, Verdict vs Monitoring, plus the assumption rows) with hairline dividers, accent header rows, and consistent column widths.
- ☐Add the five-node vertical workflow diagram beside the Build Workflow table, connected by a single accent line.
- ☐Set the Why it matters and Heads up notes as left accent-bar callouts distinct from body bullets.
- ☐Add header WSP mark top-right and the locked footer line at the bottom margin; verify 64px outer margins.
- ☐Proofread for Dave voice compliance: no buy/sell/hold, no targets, no promised returns, no em dashes, no hype.
- ☐Export final PNG at 1080x1350 and a matching PDF version for DM delivery; spot-check legibility at phone scale.
Follow Dave Wang · Wall Street Prompt · For more AI investing workflows · COMMENT & REPOST
CTA
Save this and build your monitoring agent one thesis at a time. Comment THESIS and I will share the source-trail alert template.