I recently found out that the creator of OpenClaw, an open-source platform for personal AI agents, is from Europe. When I found his blog, I really liked it, so I decided to create my own one.
In his blog, he writes that he rarely reads code anymore, but rather outsources coding to Claude Code and Codex. So I was not surprised when hearing about safety issues with OpenClaw. At the same time, I was fascinated by how far you can get using coding agents in CLI.
Since I did not want to spend money and already had a Google AI Pro subscription, I tried Gemini CLI. The tool is cool, but I could not use the latest model. After letting it run for ~1h and occasionally giving feedback, I took a deeper look at @steipete’s blog. I found out that the code is open-source, and the template is referenced (and also open-source).
Instead of using coding agents, I just used this. It’s better to think first, and then let coding agents work for you (if needed).