Business judgment on AI products
An open textbook on how to design and ship an AI assistant, with experiments you can follow.
01
Start inside the user's day · Public facts + observable behavior · 2026-09-23
A developer or learner reads an open textbook and follows its experiments to understand AI agent design principles and engineering practice.
Public materials do not show which documents, courses, or projects users currently rely on to learn AI agent design and deployment.
Public materials give only the textbook topic, not the specific difficulties learners face with existing materials, their frequency, or the consequence of leaving them unsolved.
The trend is that the right to name a new field is moving from papers to textbooks. The entry is not selling empty courses — write a follow-along lab manual for one industry, own the vocabulary, and make money on consulting and hiring.
The repository has 221 stars and 21 forks, which is attention evidence and cannot explain why learners choose it or keep using it; reader feedback and adoption evidence are missing.
① Whether the star count holds at five figures in three months — textbook repos; rise fast and cool fast; ② Whether "Harness engineering" gets cited by papers, job postings, or other; textbooks — a word is only real when other people use it; ③ Whether Pine AI surfaces public funding or partnership s…
Worth dissecting. The repository has 221 stars and 21 forks, which is attention evidence and cannot explain why learners choose it or keep using it; reader feedback and adoption evidence are missing.
in a fragmented, fast-moving technical category, the biggest opportunity is not another tool but "the textbook that threads the fragments together." A coordinate system is scarcer than a component. Whoever writes the book defines the vocabulary; whoever defines the vocabulary captures attention; attention is where all downstream monetization starts.
None. Pure open source, Apache-2.0, no paid content, no enterprise edition.; The author's company does not monetize it directly. ① Whether the star count holds at five figures in three months — textbook repos; rise fast and cool fast; ② Whether "Harness engineering" gets cited by papers, job postings, or other; textbooks — a word is only real when other people use it; ③ Whether Pine AI surfaces public funding or partnership s…
An open textbook on how to design and ship an AI assistant, with experiments you can follow.
The repository has 221 stars and 21 forks, which is attention evidence and cannot explain why learners choose it or keep using it; reader feedback and adoption evidence are missing.
An English validation note will follow from the public evidence.
The product claims to help users complete: “An open textbook on how to design and ship an AI assistant, with experiments you can follow.”. User evidence has not yet verified pain intensity or the cost of doing without it.
The assessment is recorded; an English explanation is pending.
The assessment is recorded; an English explanation is pending.
The assessment is recorded; an English explanation is pending.
02
Market comparison
The Chinese–English market comparison is not complete yet. A conclusion follows only after its coverage and verifiable evidence are recorded.
03
The call and next move come first; the full read retains the evidence and counterevidence.
An open textbook that turns "how AI agents are actually designed and shipped" from folklore into engineering: 10 chapters, 92 runnable experiments, all open source, built around one formula: Agent = LLM + context + tools.
Author Bojie Li (GitHub: bojieli): graduated from the USTC gifted program, was one of Huawei's first "genius youths" (researcher at Noah's Ark Lab), co-founded Logenic AI, and is now Chief Scientist at Pine AI.
The repo analyzed here is chemark/ai-agent-book, a study fork of the upstream bojieli/ai-agent-book. chemark's increment is a Japanese translation (10 chapters, 135 localized figures, PDF/EPUB build).
Read: the author's path is what gives the book credibility — he did frontier research at Huawei and then built startups. This reads like engineering judgment from someone who has been burned, not an academic survey. The fork's Japanese translation is also a tell: the upstream team treats multilingual distribution as a core part of textbook reach.
What it deliberately is not: a video course, a framework's docs, or a wrapper around any model vendor.
Learning agents used to follow three paths, and none of them answered "how does an agent actually work in production":
Framework docs — LangChain, CrewAI and AutoGen each have their own APIs, every tutorial starts at Hello World and teaches you how to call the API, but not how to design memory, how to handle tool-call failures, or how to standardize messages across multiple agents.
Blogs and videos — scattered, fast-stale, mutually contradictory. You finish feeling like you get it, then stall the moment you try to build.
Learning by burning — the most expensive route: pick the wrong framework, design context badly, lose three months and not know which step was the mistake.
The book attacks the shared gap: a systematic, reproducible cognitive framework. It gives the field a common vocabulary (Harness engineering) and a coordinate system for arguing about how to build agents.
None. Pure open source, Apache-2.0, no paid content, no enterprise edition. The author's company does not monetize it directly.
Read: the money is elsewhere. Li is Chief Scientist at Pine AI; the book's distribution is his recruiting channel and ecosystem-positioning statement. When a field moves this fast, whoever writes the textbook defines the vocabulary, and that definitional power flows toward the company he belongs to.
| Dimension | Call |
|---|---|
| Founder-product fit | Exceptional. Genius-youth researcher, startup founder, current AI chief scientist — full chain inside the room |
| Product insight | Caught the "field lacks a systematic textbook" gap and gave readers memory anchors via a formula and Harness engineering |
| Execution quality | 92 runnable experiments plus multilingual build chains plus versioned PDFs; far beyond a typical open book |
| Timing | Precise. The agent boom plus framework fragmentation is exactly when a textbook is most valuable |
This is a textbook case of "whoever writes the textbook defines the vocabulary." The market is not short of tutorials; it is short of a coordinate system. One formula (Agent = LLM + context + tools) and the Harness concept give the whole conversation an anchor. The star velocity (16k in 16 days) says the need is real and widespread.
The book's judgment holds up too. Putting the competitive edge outside the model — context, tools, memory, evaluation — rather than in the model itself is a position largely validated by 2026's agent practice: most people fail at agents not because the model is weak.
But separate the book's influence from its value as a "product." It does not sell, has no follow-on service, and its value is one-time. Do not watch it for revenue; watch whether the vocabulary it invented actually becomes lingua franca — and whether the author's company can absorb the attention.
① Whether the star count holds at five figures in three months — textbook repos rise fast and cool fast ② Whether "Harness engineering" gets cited by papers, job postings, or other textbooks — a word is only real when other people use it ③ Whether Pine AI surfaces public funding or partnership signals enabled by this book's reach
Product logic: in a fragmented, fast-moving technical category, the biggest opportunity is not another tool but "the textbook that threads the fragments together." A coordinate system is scarcer than a component. Whoever writes the book defines the vocabulary; whoever defines the vocabulary captures attention; attention is where all downstream monetization starts.
Worth watching. This is not a product to buy; it is an event that is defining industry vocabulary. It validates a transferable path: during a technology boom, the textbook is the most underrated influence lever. The star count has proven the demand; what remains to verify is whether the vocabulary gets adopted.
05
Use these searches when the official site is missing or the current link is only a lead.