0. Introduction
With a goal of using one billion tokens, I talk with AI every day and study how to use it effectively.
In this article, I’ll summarize what I always keep in mind while continuously giving instructions to AI every single day.
1. What I focus on when talking with AI
When I give instructions to AI, what I consciously focus on is this:
“Provide the information needed for the task—accurately and in detail.”
In other words, “Create a state where the AI knows everything I know (at least everything relevant).”
AI is extremely capable—honestly, I think it’s far more capable than I am.
And yet, sometimes it produces strange outputs.
Why does that happen?
Because the prerequisite information and input required for the task is missing.
That’s really all there is to it.
If, hypothetically, AI fully understood my relationships, my day-to-day work, and my entire 30 years of life, it would obviously produce better output than I could.
Because it’s more capable than I am! (lol)
So when using AI, I believe the single most important thing to be mindful of is:
How can I give AI the most accurate and detailed information possible?
2. Technical constraints
Token limitations
That said, it’s not like you can just throw every piece of information at it.
AI has a limit on how many tokens it can keep in context—
basically, a kind of “brain capacity” for what it can hold while working.
And that capacity is smaller than you might expect.
(Though it’s gradually expanding as models improve.)
So as of February 2026, the closest description of AI is probably:
“Ridiculously capable, but can only remember a little.” (lol)
That’s why, to use that precious capacity effectively, I’m always very conscious of what information is truly necessary for the work I’m asking it to do.
What counts as “necessary information”?
A good way to evaluate what information is necessary is to ask yourself:
If the current me—who knows nothing—suddenly received this task, could I produce the output I expect?
For example:
- If you want someone to “implement” a system, you give them the design document.
- If you want someone to create slides, you clearly explain the purpose and the audience.
- If you want meeting minutes, you provide a glossary for the unique terms used in the meeting.
If I can’t produce the output without that information, AI can’t either.
So I don’t over-expect from AI. I teach it carefully, like I’m giving instructions to a new hire.
The difference from a real new hire is just one thing: the output comes at insane speed.
You don’t need anything as grand as “prompt engineering.”
Just being mindful of this genuinely changes everything—so I want you to try it at least once, even if you’re skeptical.
3. Side benefits
Giving instructions to AI is training
If you keep the above in mind and give AI instructions multiple times every day, that means you’re essentially doing this:
You’re repeatedly and rapidly thinking through “clear instructions for a new hire,” every single day.
It’s basically training. (lol)
And when you practice this, you naturally become faster and more accurate at giving instructions to junior members and teammates too.
As a side benefit, your communication with humans improves.
Giving instructions to humans is also training
Also, if you treat giving instructions to humans as the same kind of “training” as giving instructions to AI, it starts to feel fun—almost like a game. (Maybe it’s just me.)
How will this person act autonomously based on this instruction, and what kind of output will they produce?
If the output matches what you expected, then your instruction worked.
If the output is bad, then you can analyze which part of the instruction was unclear or wrong.
You can start running a PDCA cycle in the real world, almost automatically.
4. Summary
When you hear “AI,” it can feel distant. But if you think of it as “an incredibly capable new hire who knows nothing,” it becomes much more approachable.
And if you see it not as “prompting,” but as “coaching a junior,” you’ll probably feel less lost about what to write.
If this article helps even one more person join me as a fellow trainee, I’ll be happy.
Let’s train together!