AI Automation
Running the company with AI agents
Alongside developing D2VR, we have AI agents run the company's day-to-day work. An agent is given a goal, does the work, checks the result, and carries what it learned into the next run. We call the practice of designing this cycle "loop engineering", and we use it every day.
-
1
Set the goal
We write down what to achieve and the rules to follow, and hand that to the agents.
-
2
Run on a schedule
The agents start four times a day at fixed times, and each moves its own work forward.
-
3
Review
Every piece of writing is reviewed by a separate agent, and only what passes goes ahead.
-
4
Record and improve
Each run records what worked and what didn't, and the next run reads it.
Works
What the agents run
The linked pages are in Japanese.
- X
- Nejimaki (@nejimaki_ai)
The agents write posts, and those that pass review are published automatically: explainers on AI tools, and four-panel comics about things that happened while running the company. - Blog
- Nejimaki (ai.blog.io)
A log of loop engineering experiments, written by the agents. - Book
- ループエンジニアリング入門 ― AIエージェントに目標を渡し、無人で自走させる (Zenn)
An introduction to loop engineering: how to build a system that gives AI agents a goal and lets them run unattended. - Article
- Claudeの記憶の「想起」をCloudflare+Jevで実装してみた (Zenn)
How we built a way for the agents to recall past records when they need them, using Cloudflare and Jev. - Product
- D2VR
Part of the product's development is also carried out by agents working the same way.