Meet the Lobster. 🦞
No, not dinner. I’m talking about Clawdbot, an AI agent that lives in my server.
Most AI tools are just chat apps. You ask, they answer. Boring. Clawdbot is different. It has shell access. It can run commands, edit files, and deploy my code. It’s like having a junior dev who types really fast but sometimes breaks things.
I used it to move this blog from Zola to Astro. Here’s what happened.
The “Intern” Vibe
It changes how you code. You stop typing and start managing. I say: “Split the blog into notes and posts.” It executes: Writes the config, moves the files.
But it makes mistakes. Once, it broke all my image links with a bad find-and-replace command. I told it: “You broke it. Fix it.” And it did. It even apologized.
I once made it run 100 laps as punishment. It actually counted to 100.
Why It’s Cool
1. It Runs Commands
When a build fails, I don’t copy-paste errors. Clawdbot sees the error because it ran the build. It fixes the config and tries again. Fast.
2. It Remembers
I don’t have to keep saying “I use Bun” or “My timezone is Jakarta.” It remembers.
3. Infinite Toolbelt
It’s not just a coder. I gave it the bird skill, and now it runs a cron job to summarize my Twitter/X timeline. It uses gog to check my calendar, and intomd to parse documentation. It even has a frontend-design skill to help me build UI components.
4. It Listens (Literally)
I don’t always want to type commands. Sometimes I just send a voice note on Telegram: “Deploy to staging.” Clawdbot transcribes the audio, understands the intent, and executes the code. It feels less like a terminal and more like talking to a teammate.
5. It Debugs Itself (Sometimes)
When a deployment failed because of a permission error, Clawdbot didn’t just crash. It suggested a fix: “Revert the commit and try again.” It reads its own error logs. That’s better than most interns.
The Scary Part
Giving AI shell access is scary. I watched it delete folders and rewrite configs. But the speed? Crazy.
- Built a custom CLI content generator.
- Set up CI/CD pipelines.
- Added Mastodon-powered comments. All in one weekend.
Conclusion
Is it perfect? No. But for the first time, I felt like I was working with AI, not just using it. It’s messy, fast, and fun. Just double-check its work before you deploy.