What I built. What happened.
I use AI agents for real marketing work and document the systems behind it.
The code, rules, operating numbers and failures matter as much as the final output.
Experiments
My brand, run with AI agents
Claude agents help prepare, review and schedule content from a private GitHub repository of plans and written rules.
I give direction, make decisions and review the scheduled work.
An automated SEO blog
An automated publishing system for my own e-shop.
The planned rate is five English articles a day, or 35 a week.
By 7 November, 65 articles were live.
The agent that published nothing
One agent stopped for human approval. Another created three LinkedIn articles and images but could not schedule them.
The second failure had a simple cause: the task ran without access to the signed-in browser.
What each experiment includes
- The problem I wanted to solve.
- The workflow and rules.
- What actually happened.
- The numbers I collected.
- Failures and corrections.
- The next test.
I update running experiments on their original pages.
Get the next experiment
I send a digest every Friday and a topic edition when there is something worth sharing.