What this case covers
In a WeChat Channels short-video sales project, Senlin spent half a month connecting Obsidian, VS Code, Claude Code, Codex, and data tools into an AI secondary-creation copywriting system. From June 9 to 15, it ran a total of 954 orders, with sales of 55,929.53 yuan, estimated commission of 20,245.53 yuan, and used data to continuously update copywriting and account rules.
Questions it answers
- After 10 accounts were throttled, why must you shift from revising individual posts to building a system?
- Why should AI rewriting be split into a three-step pipeline?
- How does Codex participate in account allocation and publishing duplication?
- How is a high-converting piece of copy broken back down into a rules library?
- How should a beginner start running for 7 days from the smallest system?
Our observations
- The system's value lies not only in copy generation, but also in connecting accounts, publishing, transactions, and review into a feedback loop.
- The 954 orders and estimated commission come from the combined effect of the existing account matrix, product selection, and operations, and cannot be attributed solely to AI tools.
- Having humans confirm copy and allocation first, then gradually automating, can reduce the risk of homogenization and errors being amplified at scale.
From the author
Don't just save chat logs anymore. Create a folder. Put today's raw text, final drafts, rules, and reviews into it. The system started growing from this moment.