AI Fluency - Now what
You don't need code to master AI. This guide applies the AI Fluency framework to financial analysis, teaching you how to "Redline" 10-K reports to find market edges using AI.
Please be aware that all the information provided here is for informational and educational purposes only. It is not financial advice, investment advice, or a recommendation to buy, sell, or hold any security.
TL;DR
“But Fernando, I’m not a coder. I can’t leverage AI!”
This is the single biggest misconception holding professionals back. The future of work isn’t about writing code; it is about Architectural Thinking. If you can think, you can win.
In this deep dive, we shatter the “coder fallacy” by applying the AI Fluency Framework to a high-stakes financial task: finding “Alpha” in complex 10-K reports. You don’t need to be a programmer to outsmart the market; you just need to know how to orchestrate the machine.
I will walk you through a step-by-step “Redlining” workflow. You will learn how to manually Curate high-signal context to avoid noise, how to command the AI as an Expert auditor to spot hidden risks, how to Navigate around hallucinations when the model fails, and how to Collaborate to synthesize the “story behind the numbers.”
This isn’t theoretical. It is a practical, copy-paste blueprint for analyzing financial documents faster than any human analyst. The edge isn’t in the code—it’s in the narrative, and you are the architect.
Now that the AI Fluency series has been finalized, I’m getting questions:
But Fernando, I’m not a coder. I have no intent to learn to code. I can’t leverage this!
And my answer is:
Can´t you?
Let’s get busy. Have you ever heard about 10-K reports? A 10-K is a comprehensive annual report required by the SEC, filed by public companies. It details the company’s financial performance, business operations, and major risks for investors.
The headline numbers on a 10-K (revenue, EPS) are the places everyone will be looking for information; that data is already priced in by the time you see it. The edge isn’t in the numbers themselves, but in the narrative and the nuance that explains them.
THE PROBLEM: What to look at?
As a trader, your goal is to find information that the market has missed, misunderstood, or under-appreciated. The 10-K is a legal document, and management must disclose its risks and (to a large extent) its reality.
Here is where I would hunt for alpha in a 10-K.
1. Item 1A: Risk Factors
This is, in my opinion, the most valuable section for a trader. It’s not just a legal formality.
The “Boilerplate” vs. “Specific” Test: Every company lists “cybersecurity,” “economic downturns,” and “competition.” Ignore those. You are looking for new, highly specific risks.
No Edge: “We face intense competition, which could harm our business.”
Edge: “We rely on a single, non-diversified supplier for the new micro-compressor used in our flagship product, which will account for 60% of next year’s projected revenue.”
Year-Over-Year Changes: This is the real secret. Use a document comparison tool (or just your eyes) to see what new risks were added this year and, just as importantly, what risks were removed. A newly added risk about supplier concentration, a specific lawsuit, or a regulatory change is a direct signal from the legal team.
Ordering: Management is supposed to list risks in order of importance. If a risk about “customer concentration” (losing a big client) suddenly jumps from #8 on the list to #2, that’s a massive red flag.
2. Item 7: Management’s Discussion & Analysis (MD&A)
This is where management gets to “tell their story” in their own words. You’re looking for the story behind the numbers.
Non-GAAP vs. GAAP Reconciliation: This is a goldmine. The company gives you its “adjusted” or “pro-forma” earnings. Look at the reconciliation table that bridges this to the official (GAAP) number. What are they “adjusting” out?
The Tell: Are “one-time” restructuring charges or “extraordinary” legal fees happening every single year? If so, management is trying to paint a rosier picture. These “one-time” costs are just the real cost of doing business.
Liquidity and Capital Resources: This tells you if the company is about to run out of money.
Cash Burn: How much cash did they burn in operations?
Debt Covenants: Look for any mention of “covenants” on their loans. Are they close to tripping a covenant (e.g., their debt-to-EBITDA ratio)? If they breach it, the bank could call the loan, forcing a desperate and dilutive equity raise. This is a powerful signal for a short position.
Contractual Obligations: This table shows you how much cash is already spoken for in the coming years (leases, debt, supplier contracts). If this number is massive relative to their cash on hand, they have zero flexibility.
3. Item 8: Financial Statements & Footnotes
The numbers in the statements are the “what.” The footnotes are the “how.” Analysts are often too lazy to read them.
Footnote 1: Accounting Policies (Revenue Recognition): This tells you how they actually book revenue. Are they aggressive? For a software company, do they book a 5-year contract’s entire value upfront or recognize it over the 5 years? A change in this policy can create a “sugar high” in revenue that isn’t real.
Segment Data: This footnote breaks down the company’s performance by business unit or geography. It’s common for a high-growth, sexy “new” business to be masking a rapid decline in the “legacy” cash-cow business. This note tells you if the core is rotting.
Legal Proceedings (Item 3): This is where you find the lawsuits. The company will say, “we do not believe this will have a material impact.” Your job is to find the case number, go to the court dockets, and see if that’s actually true.
Acquisitions: When they buy another company, they have to estimate the value of “goodwill.” If they overpaid, they will eventually have to write down that goodwill, which will crush their earnings in a future quarter. This note gives you a hint of how much “fluff” is on the balance sheet.
4. Item 1: Business (The Competition Section)
Skip the fluffy description of what they do. Go straight to the “Competition” sub-section.
The Edge: Compare this section to last year’s. Who are they newly listing as a competitor? This is management’s first official acknowledgment of a new threat. If a small, disruptive startup suddenly appears in the 10-K of an industry giant, that startup might be the real long-term winner (and the giant might have a problem).
THE SOLUTION: A trading technique, “Redlining”, combined with AI Fluency
The single best way to get an edge is to compare the current 10-K against the prior-year 10-K.
The Goal: A Manual, AI-Assisted “Triage”
Our objective is to manually execute your 3-step “Trader’s Triage” by applying the AI Fluency framework. We will use AI as a “Co-Intelligence” partner to find the “edge” in under an hour.
The Workflow: Your 3 Steps Mapped to the AI Fluency Roles
Your Step 1: The “Redline” This is a Curator + Expert task.
Your Step 2: The “Deep Dive” This is a Curator + Expert task.
Your Step 3: Synthesize “The Edge” This is a Collaborator task.
(Error Handling at any step This is the Navigator role.)
Part 1: The Curator (Manual “Redline” Curation)
Before you can run your “Redline,” you must manually prepare the context. As a Curator, your job is to build a “Hybrid Contextual Cockpit” for the AI.
This is a manual “Parent-Child” curation strategy:
Open the “Parent” Documents: You have two files open:
Current_10-K.txtandLast_Year_10-K.txt.Manually Create “Child” Chunks: Your WBS (your “Step 1” instructions) tells you exactly which “child” chunks you need. You manually copy/paste them:
Context_1A_Current= [All text from “Item 1A. Risk Factors” from the current 10-K]Context_1A_LastYear= [All text from “Item 1A. Risk Factors” from the prior 10-K]
Feed the AI: You will feed the AI only these two high-signal, low-noise packets. This manual curation avoids the “Garbage In, Garbage Out” problem of “Lost in the Middle” by not forcing the AI to read the entire 100-page documents.
Part 2: The Expert (Executing the Triage with Grounding Prompts)
Now you act as an Expert, assuming the AI is a “plausibility engine” that will “fabricate” if not constrained. You will use precise “grounding prompts” to force the AI to act as an auditor.
Task 1: The “Redline” (Your Step 1)
Start a new, clean chat.
[YOUR PROMPT]
You are a meticulous financial editor. Your task is to act as a “document comparison” tool. You will compare the [CURRENT] text and the [LAST YEAR] text and identify only the material changes.
You must base your answers ONLY on the two texts provided. Do not infer, add, or use outside knowledge.
[LAST YEAR: Item 1A. Risk Factors]
[You manually paste Context_1A_LastYear here]
[CURRENT: Item 1A. Risk Factors]
[You manually paste Context_1A_Current here]
[TASK & FORMAT]
Provide a report only on the differences, formatted as follows:
1. New Risks (Additions):
[List any new risk factor sentences that appear in [CURRENT] but not [LAST YEAR]]
2. Re-ordering:
[State if a major risk (e.g., ‘supply chain’, ‘talent’) has moved up in the list]
3. Word Changes (Modifications):
[Identify any subtle language shifts. Example: “we are confident” “we believe”]
You then manually copy this “Redline” output into your draft document.
Task 2: The “Deep Dive” (Your Step 2)
This is a separate, more complex task. You start a new, clean chat to avoid “context pollution”.
First, you manually curate the “child” chunks for this task (e.g., the A/R and Revenue numbers from the footnotes).
[YOUR PROMPT]
You are a forensic accountant. Your task is to analyze the following financial figures and compare their growth rates.
You must base your answers ONLY on the data provided. Do not use outside knowledge.
[FINANCIAL DATA]
Last Year:
Revenue: $100M
Accounts Receivable: $10M
Current Year:
Revenue: $110M
Accounts Receivable: $20M
[TASK & FORMAT]
Calculate the year-over-year growth rate for Revenue.
Calculate the year-over-year growth rate for Accounts Receivable.
Answer this question: Is Accounts Receivable growing faster than Revenue?
You manually copy this “Deep Dive” output into your draft document.
Part 3: The Navigator (What to Do When It Fails)
What happens when the AI fails Task 1? You ask for a “Redline,” but it just gives you a summary of both sections (Strike 1). You correct it: “No, I need a comparison“ (Strike 2, it does it again, but badly).
As a Navigator, you recognize the “context is polluted”. You abandon that chat.
You apply the “Restart & Refine” strategy. You open a new, clean chat window and craft a superior “Genesis Prompt”. This time, you manually add “few-shot examples” to pre-empt the error.
Refined “Genesis Prompt” (Example)
... [ROLE, TASK, CONTEXT] ...
[EXAMPLE (Few-Shot Prompting)]
Here is an example of the output I need:
BAD OUTPUT (Summary): “Last year’s risks were about the supply chain. This year’s risks are about cybersecurity.”
GOOD OUTPUT (Redline):
ADDITION: “We are now dependent on a single supplier for our new ‘Project Titan’ product.”
WORD CHANGE: Risk #3 changed from “if we fail to innovate” to “given the rapid innovation by our peers.”
... [FORMAT] ...
This refined manual prompt makes your workflow robust against failure.
Part 4: The Collaborator (Synthesizing “The Edge”)
This is Your Step 3. You now act as a Collaborator, using the “AI-Enabled Auditing” model. The AI has done the 80% of data screening; you will provide the final 20% of strategic judgment.
You start a final, clean chat and feed it your manually curated, verified outputs.
[YOUR PROMPT]
You are a hedge fund analyst. Your job is to find the “edge” by identifying where the narrative and the numbers do not match, based only on the verified data provided below.
[Verified Narrative Data]
From MD&A: Management claims “strong, continued demand”
[Verified Numbers Data]
Revenue Growth: +10%
Accounts Receivable Growth: +100%
[TASK]
Based only on this data, state the “Bearish Edge” and explain the discrepancy.
The “Co-Intelligence” Final Report:
The AI generates the 80% (the factual audit):
AI’s 80% (The Audit): “BEARISH EDGE: The narrative claims ‘strong, continued demand,’ but the numbers do not match. Accounts Receivable (+100%) is growing 10x faster than Revenue (+10%), which suggests the company is ‘selling’ products but not collecting the cash.”
Now, you, the Human Collaborator, perform the final 20% “Editor-in-Chief” review. You take the AI’s factual output and add the “irreplaceable... strategic framing”—the “so what?”—that the AI cannot.
Your 20% (The Thesis): You write in your final report: “The story is a lie. They aren’t collecting cash. This is a massive liquidity and demand crisis in the making.”
THE FUTURE: The Real “Edge”: You Are the Architect
So, let’s return to the original question: “I’m not a coder. I can’t leverage this!”
As we’ve just walked through, the entire workflow had zero to do with writing code. The “edge” wasn’t found by running a complex algorithm or building a neural network.
The edge was found because you acted as the AI Architect.
You, the human-in-the-loop, performed the critical, irreplaceable tasks:
As the Curator, you knew what to compare (Item 1A vs. Item 1A). You didn’t ask the AI to “read the 10-K and find the risks.” You fed it a high-signal, manually-curated context.
As the Expert, you knew how to ask. You used precise, “grounding prompts” to force the AI into an auditor’s role, demanding a “redline,” not a “summary.”
As the Navigator, you knew what to do when it fails (the “Restart & Refine” strategy) instead of just giving up and saying “the AI is dumb.”
As the Collaborator, you knew your job was to provide the final 20% of strategic judgment—the “so what?”—that turns the AI’s 80% data audit into a tradable thesis.
This is the future of knowledge work. This is AI Fluency. It’s not about being a coder; it’s about being the architect of your own insight.