What this case covers
The author is an employed programmer who uses AI to assist with Google Ads affiliate campaigns, writing ads for Amazon merchants to earn commission. The article covers campaign basics, pitfalls in automation workflows, replacing browser control with APIs, semi-automated Skills, and token cost control.
Questions it answers
- Why does going fully automatic before the process is run in smoothly cause repeated rework?
- Why avoid concurrency and hard browser control?
- How do you start building a semi-automated workflow from Skills?
- Which rules should be written as scripts, and which judgments should be left to AI?
- How much does burning 60 billion tokens in one month cost?
Our observations
- Running it manually first and then automating is more cost-effective than doing it all at once.
- Browser control and concurrency are the main sources of automation failure.
- Humans making multiple-choice decisions and AI running fixed processes is the more stable division of labor at present.
From the author
AI is the executor, executing linear processes. You are the decision-maker, making multiple-choice decisions.