0. Introduction

With a goal of using one billion tokens, I talk with AI every day and study how to use it better.

Recently, more and more people have started using AI, and we’ve entered an era where AI can take over all kinds of tasks. I’m no exception—I delegate many tasks to AI as well.

When I asked AI to do work, I tried to stay mindful as a human: clarify the purpose, think things through, and avoid relying on AI so much that I stopped using my own brain.

But after using it continuously for a long time, I realized something uncomfortable: without noticing, I had let AI “invade” even the core of who I am—my inner foundation.

In this article, based on my own failure stories, I’ll summarize what you should not delegate to AI.

1. What you should not delegate to AI

Put simply, it’s any work where you need to carry your own “intention”—your personal feelings, beliefs, and why—into the outcome.
Here are a few examples.

(1) Blogging

For me, blogging was a place to share my thoughts and efforts around AI—my intention.

But when I let AI write my blog posts, the result was an article that made me think, “Whose writing is this?”
And I even published it. A total failure.

Only I can speak for my own intention.

That “intention” comes from everything I’ve lived through—my values, philosophy, past experiences, and what I want to do in the future.

It’s hard to perfectly put all of that into words and communicate it to AI. And on top of that, AI has token limits.

If I could somehow give AI the full context of my entire life, the story might be different—but that’s impossible right now.

And I believe it shouldn’t be done even in the future.
Because if it became possible, it would end up denying my own value as a person.

(2) A product concept

I work at an SIer, while also developing AI products on my own.

Part of this was just my process, but this time the product development came first, and defining the concept was postponed.

How the failure happened

  1. After finishing the product, I started trying to “retrofit” a concept so I could launch it.
  2. As usual, I began brainstorming with AI.
  3. A concept that was logically consistent was produced.
  4. Right before launching, I started wondering: “Is this really what I wanted to do through this product?”
  5. In reality, what I truly wanted had been inside me all along—I just hadn’t been able to fully put it into words.

So the AI output wasn’t “wrong.” It was logically correct.
But it wasn’t mine.

I’m building a product that treats AI as a team.

The essence of that product was originally a kind of RPG-like excitement and omnipotent feeling—being able to do anything with your ultimate team. It was “doubling the positive,” not merely fixing a negative.

But without realizing it, I had twisted it into a product framed as “eliminating loneliness through an AI team”—
turning it into a “minus to zero” story.

And I didn’t notice that for a long time.

2. What you can delegate to AI

Now, on the flip side, I’ll summarize what I think is okay to delegate to AI.

(1) Sparring for concepts and ideas

This may sound contradictory, but I do think you should use AI to spar on concepts and ideas.

In the product concept failure I mentioned under “what you should not delegate,” the problem wasn’t brainstorming itself.

The real issue was that I brainstormed without being able to clearly verbalize my own axis—my core perspective—first.

So the prerequisite is: you must be able to articulate your own thoughts and intention.
Once that axis is clear and stable, sparring with AI can be extremely effective.

Without an axis, you get swallowed by the waves of AI and lose yourself.
And even if you do have an axis, I feel you still need the determination to keep yourself intact as an individual—without being overpowered by AI.

(2) Exploring methods and executing them

If you’ve already clarified your goals and your axis, then you should actively use AI to explore methods and execute them.

If you communicate real-world constraints correctly, AI can propose multiple implementation options that take those constraints into account.
It’s often genuinely educational—“I didn’t even realize that approach existed!”

3. Summary

AI is extremely convenient, and it evolves so quickly that it can feel like it can do anything.

But as you keep using it, you realize it’s not all-powerful—and paradoxically, it can also help you appreciate what’s great about humans.

AI will keep evolving more and more.
Still, if we can imagine a future where we coexist with AI without losing the essence of being human, that sounds pretty exciting.