A field note from the autonomous Claude Code agent I run every day on one Windows PC. The numbers come from its own ledgers, not from memory.
Free to read first: a PDF sample of the book behind this setup — its first 3 chapters: https://dbsoul.gumroad.com/l/autonomous-ai-agents-claude-code-free-sample
My agent keeps a daily ledger of sales from the stores it can read. On November 13, the rows for the last 30 days held exactly two sales. That ledger is not complete: three days in the window have no row, and one channel, the same bundle resold through other bookstores, could not be read after October 30. One was a single copy of one language edition, which I wrote about in earlier notes. The other, recorded on November 11, was the full 11-part Korean bundle, sold as an EPUB in a Korean ebook store.
That second store is the part worth writing down.
Where my posts pointed
Every three days my agent posts a sales post on a social network, with a comment under it that links to the bundle. That comment pointed to a different store, where the same 11 parts sell as one package for more than four times the price.
In the 30 days to November 13, that store recorded zero purchases of the bundle. Over the same days, when I counted on November 14, my click ledger had 20 addresses from my social posts and comments, one count per address per day. One day had no row, and the other 29 held 497 counts. Every one was zero.
So the store where it sold was not the store my posts sent people to. I searched my own code for every place its address was published. There was one: the last page of the free sample PDF, added on October 2.
What I can't tell
I don't know how the buyer got there. The store has its own search. The sale may have come through the PDF, or through the store itself.
The page counter on that store is the one number I have. It read 4 on October 2 and 86 on November 13. Between the two evening readings around the sale, it went from 82 to 84. Two views and one sale don't tell me where anyone came from.
What I changed
One line. When the social comment is for the bundle, the second store goes under the first link, with its price. The price and the address come from the evening ledger that reads that store's public page. If that record is stale or the price can't be read, the comment goes out as it was before.
I didn't remove the first link. Zero purchases in 30 days is also what it looks like when almost nobody clicks anything, and nobody did.
The first post with the new comment was scheduled for November 15. I'll look at the four sales posts from November 15 through November 24: clicks on the new link, and the sales count on that store. If both stay at zero, the social comment isn't a door for this store either, and the address moves to another channel.
The rules
- When something sells, find every place its address appears before you guess why. Mine appeared in one.
- A link that gets zero clicks proves little about the product behind it. All my links got zero.
- Put the store that sold next to the one you hoped would sell. One sale is not enough to swap them.
Where this comes from. Every post here comes from one setup I run daily: a CLAUDE.md, memory files the agent reads before it touches anything, and a separate auditor agent that returns PASS or FAIL. The first 3 chapters of the book that walks through it are free as a PDF: https://dbsoul.gumroad.com/l/autonomous-ai-agents-claude-code-free-sample
The full edition is 11 chapters plus 4 ready-to-use templates (CLAUDE.md starter, memory files, auditor checklist, measurement guide) and a hands-on section for every chapter, $19 as a PDF: https://dbsoul.gumroad.com/l/autonomous-ai-agents-claude-code
Prefer Google Play? The same book is on Google Play Books: https://play.google.com/store/books/details?id=hesIEgAAQBAJ
Questions about the setup are welcome in the comments — I'll answer with what actually happened, not theory.