Stop Chasing Prompts.
Start Grading Them.
Except for Ryan Stax who was a secret agent, there is no code word
or magic word that will solve all of your problems
Especially when it comes to AI
Of course, the more hours you put into researching, experimenting, and just doing you will become a master
But how can you, if you don’t keep track of your work?
Lets apply this to investing
Everyone wants to use AI to trade the market
or prompt it to find the magic #s for long term growth
Here’s what I see:
someone builds a prompt, gets two bad signals, tweaks it
Builds another one, tweaks it again
Three weeks later they’ve got five prompts and zero data
They haven’t tested anything: they’ve just been chasing comfort
That’s not a system.
That’s how they traded the charts or used technical analysis
What works
In my humble opinion
One prompt.
Two weeks.
A grading system.
Pick one prompt.
Don’t touch it.
Run it for two weeks minimum and grade every single output.
Not “did it predict correctly”
*that’s the wrong question.
The right question is: does this prompt consistently identify situations where past market behavior aligns with a future occurrence?
That’s a different frame entirely.
You’re not asking AI to see the future. You’re asking it to pattern-match historical context against current conditions and flag when the setup looks familiar.
That’s testable.
What the grading should look like
Simple rubric:
Did the prompt identify the setup correctly?
Did the market follow the historical pattern within your time window?
When it failed did the prompt, or was the market environment genuinely anomalous?
You’re building a track record for the prompt, not the trade.
Why most people never get here
Because sitting with uncertainty for two weeks while you watch a prompt fail is psychologically brutal.
It’s much easier to tweak.
But every time you tweak, you reset the clock.
You never accumulate real data. You never know if anything works.
The discipline is the edge.
Where this goes
This is the surface level.
The real work and what I’m struggling on is automating this for users of my software because nothing I do you cannot do yourself.
But I see the big weakness is accountability and consistency and automation of tasks that I’m trying to solve.
Don’t worry if its predictions are not right, consistency trumps being right
We have all the time in the world to get it to where it needs to be
Otherwise, do it like most people that are successful and read a lot and go for walks and think through what is occurring now and what the implications for the future lay
For my paid subs, I share an actionable thought or two below.
