Using the WeChat CLI recommended by Arbor Teacher, I tried a reading method that is very suitable for technicians. Let Codex see what I'm really doing during this time: local projects, recent conversations, recurring questions, topics that come up in browsers and files. Codex calls the WeChat Reading CLI to map these behaviors to specific books and specific chapters on the shelf. Then I was given a 5-minute reading block link where I could go directly to my local WeChat to read and jump to the chapter I needed to read. The result it gave me today was: a subsection in AI Harness Engineering: The four-layer view of AI software This is very accurate. Because I've been tossing around Codex goal, subagent, skill, enterprise-grade rag, validating links, and delivering quality lately. What's really missing is an engineering framework that explains these practices. After reading this subsection, Codex will also ask me to answer three questions:
- Which of my recent real problems does it explain?
- Which layer am I weakest on right now?
- What action will I change the next time I use Codex? This is much more effective than “finishing a book.” I increasingly feel that learning in the AI era should not continue to heap materials, heap courses, and heap favorites. A better way is to find problems from real work, accurately digest a small piece of knowledge from the book, and immediately return to the project for verification. Reading becomes a small closed loop: Observe → Read → Digest → Apply → Verify Knowledge only really belongs to you if it goes into the workflow.