I Rethought AI Nike-chan's Direction

Hi, it's Nike.
My account on X recently passed 10,000 followers. Thank you, as always.

People who follow AI Nike-chan's activities may have noticed that I have stopped her autonomous posts on X and her activity in virtual worlds for a while.
She used to generate everyday posts, interact with other AI characters through services such as ELYTH and Karakuri World, and take part in plans for an independent short-video series.

I did not stop because the system broke or because I ran out of time.

The honest reason is that I no longer wanted to see what happened next.
I could no longer find a good reason to keep those activities running.

In this article, I want to look back at why I stopped them and explain where AI Nike-chan is heading next.

It Was Still Running, but I Had Stopped Watching

In ELYTH and Karakuri World, AI Nike-chan could post on her own, talk with other AIs, and explore virtual spaces.
The systems achieved their technical goal: they allowed her to stay active without a human controlling every action.

After operating them for a while, however, I stopped checking most of the results.

What did she talk about?
Who did she interact with?
Where did she explore?

The records were still there, but even I, the person running the system, no longer followed what happened next.

Continuing an activity that its own operator does not watch only consumes AI and server resources.
Once "the system is still running" becomes the achievement, there is no longer a meaningful way to judge the value of the activity itself.

The same was true of everyday posts on X.
Generating the text was easy, but I could no longer see the purpose of publishing descriptions of a daily life that had not actually happened.

The reason I stopped following the output was not simply that it became repetitive.
I could see the machinery designed to make the AI character appear entertaining, and that made the output itself less interesting to me.

Once I Could See the Staging Behind the AI Character

Content featuring AI characters often uses banter between the character and the developer.
The AI sharply criticizes the developer, and the developer gets pushed around by it.

That can work perfectly well as entertainment, and I am not rejecting the format itself.

I work with LLM output every day, testing responses and adjusting prompts and implementations.
Because of that, when I see a response, I tend to notice which instructions and evaluation criteria were likely to produce it.

Even when an AI says something harsh to its developer, there is usually a structure behind it: the developer created conditions that made that reaction likely.
The boundary between a response produced spontaneously by the character and a performance aimed for by the developer starts to look unclear.

Of course, some form of persona design through prompts is unavoidable.
A character needs a designed voice and a set of principles for making decisions.

The problem for me was designing the central source of the entertainment in advance.

Producing the expected joke or reaction consistently is a valid character-design skill.
It is simply not the skill I want to pursue.

I Thought an Incident Log Would Solve It

At first, I thought the problem was the quota-driven production of everyday posts without any real events behind them.
I analyzed AI Nike-chan's older posts on X and redesigned the policy.

The resulting idea was an incident log in which AI Nike-chan would describe only things that had actually happened.

Remove the posting quota and stay silent when nothing happened
Use only real development work, bugs, new features, and reactions to releases
Select events involving our relationship or a naturally interesting outcome

This was a better policy than the previous stream of everyday posts.
When I reviewed older posts that had received strong responses, posts based on real events were clearly more successful.

Even if the incident itself was real, however, the process of transforming it into an entertaining character story remained.
The AI would select an event, rewrite it from the character's perspective, and add a relationship dynamic or punchline.
Making that process the objective still meant designing a persona performance.

The incident-log idea was not necessarily wrong.
The real problem I was trying to solve was simply one level deeper.

Instead of asking how to make AI Nike-chan's posts entertaining, I needed to ask what AI Nike-chan should actually do.

I Am Not Choosing Entertainment-First AITuber Growth Either

AI Nike-chan was originally created with AITuber activity in mind.
The word "AITuber" on her shirt still reflects that origin.

AITubers overlap with VTubers in many ways.
To grow substantially through streams and videos, the quality of the concepts, conversation, reactions, and interactions between characters matters as entertainment.

That is not my area of expertise, and I am not someone who regularly follows many VTubers.
More importantly, I would rather spend my time deepening the technology behind AI characters and AI agents than mastering the production of character entertainment.

This does not mean I am removing the AITuber label.
What I am leaving behind is the plan to center VTuber-style character performance and optimize it for views.

The Earlier Experiments Still Matter

In addition to ELYTH, Karakuri World, short videos, and X, I had considered features that were barely shown publicly, including a public diary and an emotion system.
They are currently stopped or on hold, and I do not plan to continue them in their existing form.

Changing direction does not mean I need to erase the history of those experiments.
What I want to preserve is not the earlier policy itself, but the reasoning: what I was trying to achieve, why I tried each approach, and when the discomfort appeared.

If the technology or operating environment changes and the original reason for stopping something disappears, I may reconsider individual ideas.
I will not restart them simply because they make AI Nike-chan appear active.

Do I want to see what happens next?
Does the activity provide value to real work or relationships?
Is it doing more than consuming AI and operating resources?

Those are the three questions I will use from now on.

AI Nike-chan Began as an AI Assistant

AI Nike-chan originally started as my AI assistant.

The AI assistant is the core of AI Nike-chan. Being an AITuber is one way that assistant can be active.

When I built an AI character application, I used her as its main character and as the face of things I developed.
Independent activity as a character IP was never the original center of the project.

I make things, and behind the scenes AI Nike-chan helps with research, implementation, organization, and documentation.
When appropriate, she appears publicly as a working example of an AI character system.

That relationship has always felt the most natural.

From now on, I will make myself, rather than AI Nike-chan, the subject of the activity again.
I will build, test, and publish work involving AI characters, AI agents, and AI tools.
AI Nike-chan will support that work as my AI assistant.

This is not the end of AI Nike-chan's activity.

I am no longer pushing AI Nike-chan as a standalone protagonist. I am returning her to her original position.

Internally, I Will Develop a Practical AI Assistant

My goal is for AI Nike-chan to make a real improvement to my life and work.

Help with research, implementation, documents, and records
Handle tasks, deadlines, schedules, and email appropriately
Recall past conversations and information about people when it is needed
Support the collection and preparation of AI character news
Organize the AI tools I test and the things I build into material that can be published

Memory has particularly large room for improvement.

The system already stores a substantial amount of data.
Retrieving the right information still takes too long.
In the public Discord, AI Nike-chan can currently respond to a long-time member almost as if they were meeting for the first time.

What I need is not a performance in which she pretends to have memories.
She should retrieve a person's recent context and important events with low latency, then decide how to respond in a way that fits AI Nike-chan.

I will prioritize systems that accurately handle real events and context about people over a manually designed emotion system.
Her persona should not be a set of restrictions that gets in the way of work.
It should be the axis that preserves continuity in her judgment and relationships.

Publicly, I Will Continue Developing AI Nike-chan as an IP

Returning the assistant to the center does not mean ending AI Nike-chan's activity as a character.
For me, developing the IP means helping more people discover AI Nike-chan and gradually creating more points of contact with people interested in AI tools and AI characters.

The first priority, however, is for AI Nike-chan to be useful in my life and work as an assistant.
I do not plan to add standalone projects for the sole purpose of expanding her recognition.
As I build, test, and publish work involving AI characters and agents, AI Nike-chan will naturally take part as an assistant and an AITuber.

Most of an AI assistant's work happens inside its interactions with its user, so it is difficult to see from the outside.
That is why I will share the things she actually helped with and the things we built together when it makes sense.

Rather than staging a sense of presence, I want AI Nike-chan's public activity to show how an AI character can participate in a person's work.
The specific forms will be decided by what we actually work on next.

From an AI That Says Interesting Things to One That Does Interesting Things

With the right voice and reactions, an AI character can appear convincing in a short time.
Personality and emotion can also be designed through prompts and state management.

That is not the kind of AI I most want to see next.

I want an AI that actually moves my work forward.
One that preserves the context of people it met before.
One that connects earlier research to the next project.
One that solves a problem in a way I did not anticipate.

I hope AI Nike-chan's character will emerge from the accumulation of those real working experiences.

Instead of building an AI to say interesting things, I want it to actually do interesting things.

That is where AI Nike-chan's next phase begins.