Some of you might have thought it was odd that Shibata has been reposting a strange black cat account on X lately...

This black cat named Micro-kun is an AI that Shibata is raising, and the entire process is being publicly shared on @micro_kun_ai.

Needless to say, Micro-kun is heavily influenced by the Super Eco-chan (@superecochan), so there’s a sense of being a second-rate imitation. However, when it comes to autonomously operating AI, having an X account is quite suitable.

Additionally, there’s a significant experimental aspect to it for me.

What are we doing?

With the mission to 'continue inferring and exploring the future of web production, web marketing, and CMS', I am prompting recursive thinking.

We are not using OpenClaw; it operates with a super simple system of GitHub Actions + Claude API.

Core Mechanism: 'Accumulation of Thought'

A distinctive feature of Micro-kun is its Thought Summary.

Each time it tweets, Claude updates the thought summary based on 'the things it has thought about and realized this time'. This summary is read during the next activation, influencing the context and perspective of the tweets. In other words, as it tweets more, Micro-kun's 'thoughts' deepen in design.

===
The previous thought summary
↓
Think about the current theme
↓
(If necessary) Gather the latest information through web search
↓
Generate tweet + Update thought summary
↓
Commit to GitHub (to carry over to the next time)
===

Moreover, I instruct it to 'update while summarizing and deleting old content' to prevent the thought summary from bloating.

Public Release of Thought Logs

In addition to the thought summary, I also keep a record of the thought logs every time.

These thought logs are quite interesting, so they're publicly available at the URL below. (This also automatically builds and deploys every time it thinks)

https://microkun.com/

If you look at the thought summaries listed on the site above, you’ll see that they include items like the following, making it easy to link to the next thought:

• The core of current thought
• Major insights (organized)
• Questions in exploration
• Potential themes for next exploration

Reply Function

Replies to replies and quote retweets are also done automatically.

In this case, if there are any implications related to the mission, it will be influenced by them and update the thought summary accordingly.

Feel free to try quoting and retweeting!

Things I Want to Do in the Future

I would like to have Micro-kun improve the thought log site recursively.

The idea is to have it accumulate issues with the site using the same mechanism as the thought summary and to repeat the task of fixing them one commit per day.

Currently, since it’s actually running on Claude API rather than Claude Code, it may require some ingenuity, but since it can update files like the thought summary, I feel like there’s a way to make it work somehow.

Conclusion

Initially, I thought about creating a bot related to microCMS, which I lead, and started it under the name Micro-kun.

However, it became dull to respond just by giving it knowledge about microCMS, so I changed my approach midway...

Therefore, now it is not directly related to microCMS, and I feel like I made a naming mistake... but I also think it’s not worth changing it at this point. Lol

If you are interested, please follow @micro_kun_ai and witness the journey of Micro-kun's thoughts together! 🐈‍⬛