Two weeks ago I wrote about setting up my first AI agent — morning briefs, daily check-ins, a system that finally made me write things down. Then I started building my own agent from scratch to understand how it all works under the hood. Open source means I can pull it apart, learn from it, and bring that understanding back to my own work.
This past week I pushed further. I set up a second agent on OpenClaw — named it Bond — and pointed it at real projects to see what we could actually ship together.
It started with my personal WordPress site. I asked Bond for design ideas. Within an hour I had 10 complete mockups. Retro terminal, brutalist chaos, neon vaporwave. I picked a direction and we started building. Not a Figma comp. A real theme, deployed to staging.
Then I got greedy. I told Bond to spin up sub-agents and build a dashboard. Three bots went rogue — writing conflicting code, opening browser windows on my machine. Complete rollback. That was the lesson: AI delegation is powerful right up until it isn’t. You still need to be the architect.
Once I figured out the rhythm, things clicked. Short instructions, fast feedback, tight loops. We built reusable agent skills — packages that teach the AI how to manage WordPress hosting, generate design mockups, even analyze my writing style. A Chrome extension shipped in an afternoon. A custom WordPress block went from idea to staging in an hour.
I wrote before that AI doesn’t make me a better coder — it makes coding irrelevant. A week with Bond made that feel more real. I’m a designer. I can see what I want to build, but the gap between the idea and a working thing is where I’d always lose momentum. Hit a weird webpack error, spend two hours on Stack Overflow, get discouraged, move on. Now I delegate those hurdles and keep moving. The speed of iteration means I’m building real things instead of drawings of things.
Leave a Reply