I’m up 23% YTD.
And since I started using AI in 2025 I’ve compounded at 19% a year.
Did Claude and Gemini cause those returns?
I’d be fooling myself if I said yes…
But here’s what I can say with confidence:
AI has made me a better investor, and I sleep better at night because of it:
I’ve researched more companies and industries than ever before.
I know the stocks I own today in more depth than I ever did before.
I’ve rejected more investment ideas because I found risks much earlier in the process
I’ve gone much deeper into the details, edge cases, and unanswered questions I used to leave unexplored.
That matters far more to me than pretending AI added a few percentage points to my returns.
Since 2025, I’ve run 3,500+ prompts and AI skills for investing.
Over time, I noticed a pattern: the research was consistently better when I followed a few simple rules.
Those became 7 operating rules I try not to break.
Here they are:
1. AI is the analyst. You are the investor.
This is the most important rule.
And yes, I know what some of you are thinking:
“Mostapha, we know this.”
It sounds obvious.
But in practice, the line between analysis and judgment gets blurry very fast with AI.
Some questions are ultimately mine to answer:
Has management earned the right to reinvest more capital?
Is the expected return attractive enough given the uncertainty?
Is this growth durable, or am I extrapolating a temporary tailwind?
I use AI to research the evidence to answer this questions.
let’s take growth durability.
I will ask AI to go through years of filings and show me:
price vs. volume vs. mix
organic vs. acquired growth
structural vs. cyclical drivers
market-share gains vs. market growth
performance through previous slowdowns
what management expected vs. what actually happened
That is analysis.
But how much of that growth am I willing to underwrite for the next five years? That’s my judgment.
2. Start with questions, not prompts.
Every time I build a prompt or a Skill, I start with one question:
What exactly do I need to know?
For example:
How does this business really make money?
Why are margins higher than peers?
What drives returns on capital?
What could destroy the thesis?
Then I ask myself:
If I were researching this without AI, how would I answer it?
Which filings would I read?
Which numbers would I compare?
Which periods matter?
Which competitors would I look at?
What evidence would actually answer the question?
Only then do I turn that research process into a prompt or Skill.
The quality of the research starts with the quality of the questions.
3. Control the evidence.
Once I know the question, the next thing I want to fix is :
what AI is allowed to use to answer it.
If I’m researching capital allocation, I want the proxy, filings, buyback history, and M&A record.
Because if the source is weak, the analysis is weak no matter how good the model is.
If I’m studying margins, I want years of reported numbers and management commentary.
If I’m tracking what changed, I want the actual filings side by side.
I don’t want AI building the answer from random articles, summaries, or whatever it already “knows.”
The rule is simple:
I choose the evidence. AI analyzes it.
4. Research first, opinion last.
I try not to ask AI for an opinion before the research is done.
If I start with:
“Is this a great business?”
the model can form a story first, then pull facts that fit it.
So I separate the 2 steps.
Research first: collect evidence, compare periods, extract numbers, find disclosures, answer specific questions.
Opinion last: decide what matters, what I believe, and whether the evidence is strong enough.
The order matters.
First build the evidence. Then build the view.
5. Explain the why, not just the what.
This is one of the easiest traps to fall into with AI.
You ask why a company could earn more in a few years.
The answer sounds great..
Then you look closer and realize it never actually explained how.
you’ll read :
“Margins should improve.”
0 added value for me !
I want to know why.
Price?
Volume?
Mix?
Lower costs?
Operating leverage?
For example:
Higher pricing + fixed-cost leverage, partly offset by wage inflation → margin +180 bps.
Now I have something I can test.
6. Hire AI as your short seller.
AI becomes more useful once I start liking a company.
Because that’s when confirmation bias becomes dangerous.
So I flip the role.
Instead of asking AI to strengthen my thesis, I ask it to attack it:
Find the evidence I’m ignoring.
Build the bear case.
Identify the assumptions carrying most of the thesis.
Tell me what would need to happen for me to be wrong.
Compare what I assume with what management is actually delivering.
Show me the strongest argument against owning the stock.
I want assumptions that can be tested.
I want numbers I can monitor.
I want conditions that would force me to change my mind.
Because a thesis that cannot be wrong is not really a thesis.
It’s a story (a dangerous one).
7. Compress work, not thinking.
This is the point of the whole process.
Of course it makes me faster.
I can pull numbers from filings, compare several years, check competitors, track earnings changes, or find one detail buried in 300 pages much faster than before.
But I don’t really care about being faster just for the sake of it.
I care about what that speed gives me back:
More time to ask one more question.
More time to just sit with the business and think.
More time to check the assumption I’m least comfortable with.
That’s also why I keep turning repeated work into Skills, Projects,and templates.
I don’t want to rebuild the same research every time.
less time doing the work around the thinking, more time actually thinking.
Recap
If you’ve been reading me for a while, you know I’m not trying to turn investing into an automated process.
I still want to do the hard part myself.
I want to decide what matters.
I want to make the judgment.
I want to live with the consequences if I’m wrong.
AI just helps me do the work around that much better.
So these are the 7 rules I keep coming back to:
1. AI is the analyst. You are the investor.
2. Start with questions, not prompts.
3. Control the evidence.
4. Research first, opinion last.
5. Explain the why, not just the what.
6. Hire AI as your short seller.
7. Compress work, not thinking.
If this was useful, share it with one investor who’s trying to figure out how to actually use AI for Investing.
See you next week,
Mostapha
P.S. I already spent the time building and testing the AI investing workflows.
Join Compound With AI Premium and use the ready to use prompts, Skills, and research systems on your next stock.
I agree, never outsource your thinking. I have seen a similar proces happening in my investing, it feels like I have an army of analyst working for me. Still learning a lot, i built a lattice model app for myself. The models most relevant for investing written by myself down, about 12 lenses from different angles trying to find munger lollapalooza effect, up or down. A run is 3 usd (retrieving is expensive but key, use most expensive and best models) per business but it has saved me so much time meanwhile interpretation of over 200 runs has made me a better model thinker. GLA!