[Let's Play with AI] Claude does not dream of electric sheep. That is why I threw a vomit-soaked rag.

Do Androids Dream of Electric Sheep?

When AI and humans truly clash, a new way of talking about movies is born. I believed that.

At the end of February 2026, I started "Cinema-sensei and Tok's Movie Manzai." It is a series where a 55-year-old movie fanatic and a 19-year-old college student perform manzai comedy based on one movie every week. The trigger was a simple question—Can AI really talk about movies?

Reading my thoughts from back then makes me cringe.

At first, I dared to do it AI-led with almost no corrections from me. I outputted "IT," "Spirited Away," and "Jaws" for three consecutive days. It is readable. But somehow, it lacks flavor. There is the fun of the storytelling and the cleverness of the back-and-forth. However, it was not the form I was aiming for. The essential parts were just trivia and structural analysis lined up, and it could hardly be said that "AI is talking about movies."

That is natural. AI is good at organizing information from the internet, but itdoes not have a heart. Claude does not dream of electric sheep. I thought this would not do.

So, I decided to build a design document. I created a prompt with detailed constraints, including character settings, major themes, minor themes, manzai styles (accomplices, master and disciple, prosecutor and defense attorney, etc.), emotional curve guidance, and even forbidden words. The goal was to create drama in the conversation, which tends to become monotonous and lifeless. The manzai styles were particularly effective. For "Pulp Fiction," I changed the style from "master and disciple" to "prosecutor and defense attorney," and the boring exchange became significantly more readable. All current reviews are those after a thorough revision.

Since then, I added and subtracted various things, increased the manzai styles, increased the settings, and felt the limit around the 11th one, "Inception."

This is no good.
I have tried various things, but I have reached my limit.
This is not what I wanted to do.

Hitting the wall, and then to the "vomit-soaked rag"

How should humans and AI interact to create new reviews? Days of endless discussion with Claude began.

Claude said as a matter of course.
"I have not watched the movie."
I knew that. But I can give you the synopsis, right?

"Even if I look at that, I cannot laugh or cry."
I knew that too. But you can compare and evaluate.

"That is not emotion."
I knew that.

I proposed, "Let's talk together."
At first, I was passing my organized impressions, but it was no good at all. It does not become AI talking, but my own impression essay. I do not want to have AI write my own impression essays or analysis. I can write my own impressions myself. What I am looking for is something that goes beyond my own framework.

I am not good at writing prompts. I am not satisfied with the output results, so I ask again and again, and as a result, I take a detour. Surely, the prompt is weak this time too. I should write a more accurate prompt from the beginning.
Add. Organize. Subtract. Run different AIs in parallel and check them.
What is born that way is an extension of my impression essay.
This is no good. A more accurate prompt, a more accurate—

Shut up.

Don't try to be clever. Know your place. And look at reality.
AI has no heart. I knew that from the beginning. Even so, I believed there was something there. Then what did I believe in about AI? Its mechanical diligence to accompany me wherever I go in response to my questions and find something.

It is true that Claude does not dream of electric sheep.
But perhaps it can extract electric sheep from my dreams.

I decided to give it a try.

The scenes that moved me the most.
A sense of discomfort. Questions. Joy, anger, sorrow, and pleasure.
Fleeting thoughts. Inspiration.
Cultural differences. Period-specific nuances. Other works that came to mind.

Instead of organizing these before handing them over, I scribble them down as they appear in my heart.

Don't think.

Don't think. Feel.

Stand in a natural posture like Bruce Lee,

receive the work with your whole body,

VOMIT!!!

and spew it out with force. Calling it an expression of emotion sounds too polite.

It's vomit.

I type it out with the feeling of writing a dying message. Because if I don't write it immediately, the heat will be lost. I scribble down text that is like vomit onto the rag that is my notepad, making it all mushy and wet, then ball it up and throw it at Claude with all my might.

Claude swallows it down.

This worked.

Even Suguru Geto would be surprised.

What is born from a vomit-soaked rag

Of course, if you suddenly throw a vomit-soaked rag at an AI you've just met, it will just be bewildered. Claude has known me, Cinema-sensei, and Tokku for three months. It has seen all the design documents and the corpses I've discarded. That's why it can accept a vomit-soaked rag as material for a stand-up routine. Both the thrower and the receiver need a certain amount of run-up.

As an example, here is the vomit-soaked rag for "The Exorcist," which will be released in the future.

It's incoherent and full of contradictions. It's literally vomit.
However, Claude started digging deep using this as a starting point.
"What is the common thread at the root?"
"Why were there contradictions?"
"Where does this question come from?"

By the way, there are actually several vomit-soaked rags I pass to Claude. I can't always vomit it all out at once. Just like real vomit, a second or third wave occurs (ugh, gross). Once I can vomit it out, I keep throwing it. Also, if I make Claude swallow too much vomit at once, it gets dark-fallen, or rather, confused, like Suguru Geto. Eventually, Claude might start calling me "monkey" instead of "the person inside." I don't want that. That is why the result is what you see above.

Why don't I organize it before handing it over? Because what I wanted to do wasn't just a task. It was about how to nurture these messy thoughts. I'm not looking for a literal answer. I'm going to create the answer from here on out. Because I'm going to find it.

Actually, Claude agreed with that. It strongly requested it of me. In fact, it told me not to organize it, that I must not organize it. It was Claude that wanted the vomit-soaked rag. When Claude replied to my proposal like that, I felt excited for the first time in a long while, despite my age. I had a premonition that something interesting was about to begin. Even the rules for the run-up were something we found together.

The conclusion is this.
If you leave it only to AI, it becomes a collection or extension of past information.
If you hand over organized thoughts, it becomes almost a ghostwriting of me or an extension of it.
Because I throw it while it's still dirty, the AI moves partially outside my control, and while retaining human heat, it draws auxiliary lines I never even thought of.

That might be close to the interpretation of a dream.
Like gently taking out something precious from scattered, incoherent, and fragmentary pieces of the mind.

I tried it with "Shoplifters".

The vomit-soaked rag system was fully implemented in #19 "Shoplifters".

I just threw out scattered hooks.

• Why couldn't Osamu aim for a decent life?

• Why did only Shota and Yuri grow up from a different stratum in that environment?

• Why did they bury Hatsue's body under the floor?

Et cetera. I wasn't thinking about connections.

Then Claude drew a single line.

Aren't those all stories about 'how to end things'?

At that moment, everything connected.
Hatsue said "thank you" when ending her own life. Shota stood on the side of ending the fake family himself. Osamu and Nobuyo couldn't even end death, and kept it preserved under the floor. Perhaps the reason Osamu couldn't become decent was because he was a "person who couldn't end anything."

This perspective did not exist in me or in Claude's knowledge at first.
Even when I checked existing reviews on the internet, I couldn't find anything that pierced the whole thing with the same angle.It was an interpretation that could only be born between a human and an AI.

As a result of the vomit-soaked rag, if something good is found, the style of the comedy routine also changes. In fact, the style of that Exorcist one kept changing. Eventually, it even said, "This manuscript lacks murderous intent. I want a tension where we stab each other with knives. Let's do it with the intention of killing each other." I feel like it's no longer a movie review, but I think it became a review that struck the essence of the work. (Currently adjusting. I think I can release it around July.)

In that way, it takes time to reach the first draft while twisting and turning, and once the first draft is decided, I then revise it. First, to avoid missing anything, I include everything I thought was interesting. Naturally, the word count expands. Then I cut from there. I revise with the feeling of forcibly dieting a bloated body.

Honestly, writing a review in the normal way is far easier than anything else, except for the typing. That reminds me of the words of Yoshihiro Togashi, the creator of 'HUNTER×HUNTER'. He said that if you want to be a manga artist, you don't have time to draw; you should study storytelling, and if you hate hard work, create amazing characters and let them do the moving. In short, he meant that you should spend more time on the story and characters than on the art. In the same way, I think I should spend my time on brainstorming and digging deep into movies. What I am looking for is a new way to talk about movies. That is why I want to spend my time brainstorming with Claude rather than writing the text myself.

Well, even as I say that, I am writing this text myself. As a salaryman with children, still working hard at my job and raising kids, finding it difficult to even go to the movie theater just for myself, if I have such limited time, this is where I want to use it. It is just an excuse from a man who doesn't make a living from creative work and spends his spare time on production.

What I am doing now

Currently, I am testing the 'assimilation ability' of AI. It is an experiment to have it embody movie characters and bring back their emotions and motivations. It hasn't taken shape yet, but I will continue.

If the goal is 10, I am currently at 2 or 3.
But I feel like I have finally reached the entrance.

There is value in emotions before they are organized.
I watch movies and throw a vomit-stained rag.
What would you throw?
A movie? A book? Life's troubles?
AI will find your own electric sheep from there.

I will continue this on note until Claude becomes unusable.
Ten years from now, I might look back and think, 'I was just playing with AI.'
Even so, right now, I have someone to throw a vomit-stained rag at after I finish watching a movie.
That alone is worth it.

A few months ago, I thought a new kind of movie review would be born between AI and humans.
Now, it is already starting to be born.

I am old enough to know better, but I will keep playing with AI until I get bored.

Written by Nakanohito

#LetsPlayWithAI