KNOWLEDGE NOTESKnowledge Notes
Explore curated knowledge, course notes and methods, with their existing content and sources preserved.
This index lists complete English editions. Additional Chinese notes and bilingual updates remain available in the Chinese index.
350knowledge notesA self-contained bilingual essence of Emily Campbell’s *Shape of AI* AI product UX pattern library (all 49 .md files fully read), reorganized into 36 specific patterns across 7 categories plus 1 set of overarching design guidelines, following a queryable structure of “Pattern Name → What Problem It Solves → How to Use It → Examples”. The seven categories are: ① Wayfinders (4 patterns: Follow up / Nudges / Suggestions / Templates) — solves the problem of new users not knowing how to use a blank input box; ② Inputs (8 patterns: Open input / Auto Fill / Inline action / Madlibs / Remix-Blend / Summary / Synthesis / Token layering) — feeds user intent and context to the AI; ③ Tuners (8 patterns: Filters / Parameters / Inpainting / Model management / Personal voice / Primary sources / References RAG / Workflows) — lets users fine-tune or restructure results; ④ Identifiers (6 patterns: Color scheme purple-green / Symbols icons ✨🔮🤖 / Name / Personality / Disclosure / Initial CTA) — distinguishes AI content from human content; ⑤ Governors (9 patterns: Citations / Controls stop/continue / Footprints / Prompt transparency / Regenerate / Sample response / Show the work / Token transparency / Variations) — preserves user agency and explainability while the AI works; ⑥ Trust indicators (3 patterns: Incognito / Memory management / Watermarks C2PA); ⑦ Dark matte hidden patterns (3 patterns: Data Ownership / Caveat disclaimer buck-passing / Rating) — gray-area designs that are questionable or erode trust. For PMs building AI chat or interactive narrative products: this is a quick reference manual for “which pattern to look up when you encounter interaction problem X next time”, with emphasis on Why / When to use / Counterexamples. It complements nngroup-ai-ux-design — the latter offers NNGroup’s conversational interface research perspective (6 types of conversation / no fluff / anthropomorphism / sycophancy), while this piece presents Shape of AI’s perspective as a “reusable UI pattern parts library”.
Wang explains agents through models, environmental feedback and tool-based action. In a product, these form a continuing process: observe the current situation, invoke a tool, obtain a result and use the new information to decide what to do next.
A collection of 66+ B2B vertical AI cases from Shensi Circle. Unicorns have emerged across 8 major tracks (sales/customer service/recruitment/legal/healthcare/insurance/HR/finance), plus a case database. Representative cases: AI legal unicorn (25-year-old esports player, $700 million valuation) / AI sales company (2 rounds in 9 months, $52 million) / Chinese-founded AI customer service company ($1.5 billion valuation) / AI insurance company ($30 million raised across consecutive rounds in 6 months) / AI medical Copilot ($24 million Series B) / Peter Thiel's $35 million bet on a sales Claude Code to end Salesforce / Sequoia US: "Stop selling tools to lawyers; rebuild law firms directly with AI"
Shensi Circle 39 AI Coding company case studies. Cursor ($50 billion valuation + training as product + chief designer + Composer)/ Lovable ($50 million ARR in 6 months + 6 growth playbooks + anti-traditional growth)/ Bolt / Cognition Devin / Replit (SaaStr zero-engineer case)/ Codex / GitHub Spec Kit / Open-source AI coding tools ($32 million)/ Readdy.ai (Chinese-founded / $5 million ARR in 4 months). Covers coding Agent business models, growth playbooks, technical inside stories, and the 80→100 make-or-break line
A collection of VC-perspective content from Shensi Circle: 362 out of 659 posts contain financing/valuation keywords. Covers a16z (Marc Andreessen / Ben Horowitz / supply-driven markets / Substack investment logic / 10x judgment / three paths to PMF) + Sequoia (US Arc incubator 4 PMF questions / Sequoia US vertical bets on Agents / Sequoia India AI companies) + YC (W24–W25 trends / YC partner's 15 secrets of Vibe Coding). Includes representative views from 49 a16z posts / 20 Sequoia posts / 16 YC posts. **Unique perspective from Shensi Circle: primary-market VC perspective ("which company is worth investing in / whether the valuation is reasonable / whether the founder is reliable"), distinct from Wang Kai's secondary-market RSI investment approach**
A collection of 74 Shensi Circle case articles tagged with keywords "ethnic Chinese/China/going global". It covers model cases of ethnic Chinese founders: Readdy.ai, Lanhu, MasterGo (5 million ARR in 4 months) / customer service company valued at 1.5 billion (200 million in funding) / recruitment company with 20 million (in an era where applications surged 3x) / designer-turned-algorithm indie developer on the MIT path / AI company employee + mother of two (Pretext, 21 million views) / Stanford Chinese dropout / Berkeley Chinese (AI marketing automation, 28 million) / former frontline Google and Meta AI scientist, 43.5 million / former Amazon executive, 15 million / Cambridge PhD, 5.35 million, AI parallel world / 22-year-old MIT Chinese dropout, 21 million / Manus going global / Linkloud Global NEXT 2026 Going Global Conference
A collection of 71+ pieces on leading company moves from Shensi Circle. OpenAI (19 pieces / Sam Altman as Oppenheimer / Sora / OpenAI fund layout / Operator) / Anthropic (10 pieces / 130 billion revenue / moat = permissions / foray into finance / valuation doubled in three months surpassing OpenAI / marketing self-marketing / 80% retention of overseas talent) / Google (Gemini / Privacy Sandbox / AI rewriting the control layer) / Apple (AI iMessage / iOS integration) / Perplexity (6 pieces / 450 million ARR / 5 major growth engines in three and a half years / business strategy) / Microsoft (Entra ID / Security Copilot) / Karpathy (2 pieces). Covers strategic inflection points, business models, and competitive landscape of leading AI companies
Cross-platform complete works index for the author of Shensi Quan (Shensi SenseAI) (659 unique works / 2022-06 → 2026-05). Perspectives on AI entrepreneurship/Agent/VC, with one in-depth Silicon Valley piece per day. Come here when you need to look up "what does a16z think about XX / how is this AI company valued / how exactly does a certain AI agent company operate".
A complete historical case library covering horizontal and vertical agents, coding products, infrastructure, business models and delivery boundaries, with all published additions through October 21, 2026.
A complete historical guide to AI industry narratives, emerging categories, enterprise software, infrastructure and forecasts, with evidence limits and all published additions through October 22, 2026.
A complete historical guide to startup judgment, product-market fit, early growth, go-to-market and scaling, with the published additions on consistent delivery, audience feedback and historical-score shortlisting through October 23, 2026.
Listen: 08:45–16:39. The team first avoided a crowded general-agent market by choosing vertical agents, then sourcing: finding creators, candidates, experts, or customers. All involve finding people and seem to share search capabilities. But that starting point answered how to avoid competitors before identifying a payer’s most important task.
Meng used a mature training platform inside a model company. Starting a business meant building platforms and finding a complementary opportunity. Turning training into a service simplifies data submission and execution without automatically making the data valuable.
Key takeaways from Duolingo's official culture and product methodology playbook (refined over 14 years of trial and error, signed by Luis von Ahn). Its core is not a set of value slogans, but a transferable operating system for "how to build products": ① Take the Long View — priority order of retention > growth > revenue, betting on technology (machine-generated speech started rough then improved), long-term hiring (better to leave roles open than hire the wrong person, generalists, long average tenure), treating mascot Duo as a competitive moat, enforcing retention thresholds for notifications and ads, and the 2022 switch to the linear Path that sacrificed short-term DAU for actual learning outcomes; ② Raise the Bar — excellence is the baseline not the goal, four product standards (Useful / Intuitive / Delightful / Polished), build V1 not MVP, shake-to-report during dogfooding, be hard on work easy on people (focus on what not who), clear individual Ownership; ③ Ship It — weekly updates plus hundreds of parallel experiments, Clock Speed to shorten interval at every step, Ruthless Prioritization to cut low-impact items, the "99 bad ideas" culture, using mission to resolve the "good and fast" contradiction (meet the bar first then prioritize), using product review PR (borrowed from code review) to quantify process; ④ Show Don't Tell — Numbers Are the Story, TL;DR = conclusion + evidence + action, no PPT for prototypes, good products need no explanation, Disagree and Commit, Trust Battery is charged by results; ⑤ Make It Fun — must be entertaining to compete for attention with TikTok/IG, Committing to the Bit by taking a gag all the way, Duo's dual personality, viral April Fools' stunts, 5-second Super Bowl ad. Plus the 6-step Green Machine work method (assign people → define success → set guardrails → build + feedback loop → urgent high-quality execution → scale what works, stop what doesn't). Target audience: PMs building gamified/AI interactive narrative products; each section includes Why + real Duolingo cases + transferable insights for gamified AI products.
A compilation of official original sources for the 5 major milestones in Google Search evolution: RankBrain (2015, first deep learning ranking system / apex predator example) + BERT (2019, Transformer understanding word-to-word relationships / 10% of US English queries affected / "for someone" example) + MUM (2021, T5 / 1000x BERT / 75 languages / multimodal / Mt. Fuji example / average 8 queries for complex tasks) + HCU (August 2022, site-level ML classifier / "written for ranking" detection) + AI Overviews (May 2024, custom Gemini / multi-step + planning + AI-organized + video search). **Pandu Nayak's February 2022 overview = the only official article that covers all 4 major milestones together**. These 5 are the "constitution" of Google ranking; E-E-A-T and Google's stance on AI content are all derived from them.
Host Qu Kai used Suno to generate an opening theme, selecting one from almost a hundred candidates. This already demonstrates both the attraction and a cost of AI music: an ordinary person can produce it, while a satisfying result still requires selection.
This entry archives a local textual research report from June 2026, distinguishing between industrial background, internet labels, rumors, and location-based marketing speculation; it does not judge residents on the basis of regional labels.
Canonicalization concepts + Google's automatic canonical selection signals (HTTPS/redirect/sitemap/rel=canonical/hreflang cluster); priority table for 4 specification methods (redirect strong → rel=canonical strong → sitemap weak); complete HTTP status code table (2xx/3xx/4xx/5xx/special 429) + impact on indexing and crawling; 5 redirect types + Apache/NGINX/PHP configurations; Permanent vs Temporary semantic differences; 2 site move playbooks (host change without URL change / domain change with URL change); for temporary outages, prefer limiting functionality over full shutdown; dual paths to ask Google to recrawl; troubleshooting for 6 major canonical issues + 14 anti-patterns.
Google's 3-layer mental model for crawling/indexing/rendering control. robots.txt (only controls crawling, not indexing) + 3 major limitations. noindex (meta tag + HTTP header) implementation and debugging. Full table of 11 valid rules + 3 historical rules for the robots meta tag. Element-level fine-grained control with data-nosnippet. X-Robots-Tag (Apache/NGINX configuration + regex adaptation for non-HTML files). The strictest rule wins in conflicts. The classic pitfall where robots.txt + noindex cancel each other out. Decision table for 4 methods of controlling content. Image removal special case (Googlebot-Image UA). Full list of 9 user-triggered fetchers (which ignore robots.txt). rel=sponsored/ugc/nofollow outbound link qualifiers.
The 6 surfaces where Ecommerce content appears on Google (Search/Images/Lens/Shopping tab/Business Profile/Maps). Pros and cons of 4 launch strategies for new sites (Grand reveal / Home page launch / Launch with stock unavailable / Soft launch). 2 paths for Product data: on-page structured data + Google Merchant Center feed (automated/manual/Content API); **Merchant Center is mandatory for the Shopping tab**. 6 major relevant SD types (BreadcrumbList / LocalBusiness / Organization / Product+ProductGroup / Review / VideoObject). Ecommerce-specific rules for URL structure (use `?key=value`, do not use `?value`, do not use fragment variants; variants can use path segments or query parameters; use the query-free version for canonical); 3 common mistakes (treating fragments as variants / multiple content pieces on the same URL / infinite URL space from timestamped URLs). Comparison of 3 pagination modes (Pagination / Load more / Infinite scroll) + requirements for unique URLs + sequential links + do not set the homepage as canonical + **next/prev is deprecated**. 13 review guidelines (expert perspective / quantification / competitor comparison / evolution / affiliate disclosure).
