I was teaching my dad a few ChatGPT tips recently when something clicked for me.
He has a four-step process he likes to follow whenever he investigates a company.
It’s his process. He already knows what questions he wants answered, what order he wants to investigate them in, and what he’s looking for.
So I suggested something simple:
Teach ChatGPT the four steps once.
Then give the process a name.
Investigation Mode.
After ChatGPT understands what Investigation Mode means, instead of explaining the entire process every time, he can simply say:
“Investigation Mode: Ford.”
Fourteen characters of shorthand can now represent a much larger set of instructions.
And that’s when I realized how much of the way I use AI has become a form of compression.
We Already Do This With People
This isn’t really an AI concept.
We do it constantly with people we’ve worked with for a long time.
Imagine you’ve worked with someone for five years and they tell you:
“Give me the usual report.”
You probably don’t respond:
“What does usual mean?”
You know.
The phrase might actually mean:
- Use last month’s format.
- Update these five metrics.
- Compare them against the previous period.
- Highlight anything outside our normal range.
- Put the important stuff at the top.
- Don’t give me three pages of explanation.
- Have it ready before tomorrow’s meeting.
Two people who have worked together long enough can compress all of that shared knowledge into:
“Give me the usual report.”
That’s incredibly efficient.
The words aren’t special.
The shared context behind the words is what makes them powerful.
You Can Build the Same Thing With AI
Most advice about using AI focuses on writing better prompts.
Be specific.
Give the model context.
Define the output format.
Assign it a role.
Provide examples.
Those are useful techniques, especially when you’re starting a new conversation or working with an AI that doesn’t know anything about how you work.
But there’s another possibility.
Instead of repeatedly writing increasingly elaborate prompts, you can begin developing a shared working vocabulary with your AI.
The basic pattern is:
Do it → Correct it → Name it → Reuse the name.
That order matters.
Step 1: Do It
Start by teaching the AI the actual process.
My dad might say:
“I use these four steps whenever I investigate a company.”
Then he explains them.
There’s no magic prompt required.
He’s simply showing ChatGPT how he works.
Then he gives it a company and lets it run the process.
Step 2: Correct It
This might be the most important part.
Don’t accept the first output just because it looks polished.
Maybe ChatGPT focuses on something my dad doesn’t care about.
He can say:
“No. When I evaluate management, I care much more about their capital-allocation history than their biographies.”
That’s valuable information.
Or maybe it makes assumptions where he wants evidence.
He can correct that too:
“Don’t speculate there. If you can’t verify something, tell me it’s unknown.”
Now ChatGPT isn’t just learning the four mechanical steps.
It’s beginning to learn how he applies those steps.
That’s much more valuable.
Step 3: Name It
Once the process works the way he wants, give it a handle.
Investigation Mode.
The name isn’t what gives the process power.
The accumulated meaning behind the name does.
This is an important distinction.
Simply opening ChatGPT and saying “Investigation Mode” doesn’t magically create a good company-investigation methodology.
You have to establish what the term means first.
Once that meaning exists, however, the name becomes shorthand for it.
Step 4: Reuse It
Now the interaction gets interesting.
Instead of recreating the methodology:
“Investigation Mode: Caterpillar.”
Then:
“Investigation Mode: Deere.”
Eventually he might add another piece of shorthand.
Perhaps he likes the final investigation presented in a particular format.
Teach ChatGPT the format.
Correct it until it’s right.
Then name it:
Company Report.
Maybe he also wants a strict distinction between verified facts, reasonable inference, and things that aren’t known.
Establish that rule.
Call it:
Evidence Rule.
Now he can say:
“Investigation Mode: Deere. Company Report. Evidence Rule.”
That tiny instruction could represent hundreds of words of previously established context.
That’s compression.
Modes Aren’t the Only Thing You Can Compress
Once you see this pattern, you start noticing other things that can be named.
Processes
A repeatable sequence of steps can become a mode.
“Investigation Mode.”
“Planning Mode.”
“Review Mode.”
“Weekly Finance Mode.”
Outputs
A format you repeatedly use can become a named template.
“Company Report.”
“Executive Summary.”
“Weekly Review.”
Decision Rules
A principle you don’t want forgotten can become a named rule.
“Evidence Rule.”
“Budget Rule.”
“No-Speculation Rule.”
Perspectives
A repeatable way of examining something can become a named lens.
“Competitor Lens.”
“Customer Lens.”
“Risk Lens.”
“Five-Year Lens.”
You can even create shorthand for state.
“Freeze this” can mean: we’ve reached a conclusion and shouldn’t casually reinterpret it later.
“Pin this” can mean: this is important, but we’re deliberately not pursuing it right now.
The vocabulary doesn’t matter nearly as much as the shared meaning behind it.
This Is Different From Prompt Engineering
I don’t think most people need to become expert prompt engineers.
In fact, obsessing over perfect prompts can sometimes get in the way.
People start wondering:
“What exact words does the AI want me to use?”
I think that’s backwards.
Tell the AI what you’re actually trying to accomplish.
Give it the information it needs.
Correct it when it misunderstands you.
Show it what good looks like.
And when you discover something you’re going to use repeatedly, give it a name.
You’re no longer trying to find the perfect incantation.
You’re developing a working language.
The Vocabulary Can Grow Organically
I wouldn’t recommend sitting down on day one and inventing 25 modes.
That’s just creating another system you have to manage.
Let the vocabulary emerge from actual work.
If you notice yourself explaining the same process for the third time, that’s probably something worth naming.
If you repeatedly correct the same behavior, maybe that correction should become a rule.
If you continually request the same output structure, turn it into a template.
If you frequently ask the AI to examine problems from the same perspective, name the lens.
Repetition reveals what should be compressed.
That’s a much simpler approach than trying to design the whole system in advance.
There Is an Important Limitation
Shared language doesn’t mean perfect memory.
AI systems have context limits, memory behavior can vary, and a term established in one conversation shouldn’t automatically be assumed to transfer perfectly everywhere forever.
For important workflows, preserve the actual definition somewhere.
If “Investigation Mode” matters, keep the four steps.
The shorthand is a convenience.
The underlying procedure is the source of truth.
That distinction becomes increasingly important as these personal AI vocabularies grow.
From Tool to Working Relationship
This is where I think the idea gets bigger than a productivity trick.
When we work with another person for years, we develop shorthand.
We know what “the usual” means.
We know what “take another pass” means.
We know which details matter.
We know what requires evidence.
We know what isn’t worth spending time on.
That shared language is part of what makes long-term collaboration efficient.
AI gives us the opportunity to intentionally develop some of that same compression.
Not because the AI is a person.
Because shared context reduces the amount of information required to coordinate work.
And that changes how I think about getting better at using AI.
Maybe the goal isn’t to become someone who can write incredibly sophisticated prompts.
Maybe it’s to reach the point where you don’t need them nearly as often.
Teach the AI how you work.
Correct it.
Name what works.
Reuse those names.
Let repetition reveal the next thing worth compressing.
Eventually:
“Investigation Mode: Ford.”
might be all you need to say.