Tag: case-study
All the articles with the tag "case-study".
-
Zero Findings: Nothing Wrong, or the Checker Isn't Checking
Four times this week a "check passed" turned out to be a broken checker, plus four real bugs where every layer looked fine on its own. Before you trust a zero, prove the checker is actually comparing something.
-
檢查回 0 有兩種意思：真的沒有，或檢查器沒在比對
一週內四次「檢查通過」其實是檢查器自己壞了，另外四個真實 bug 則是各層局部都對、組合起來仍出包——回 0 之前，先證明檢查器真的有在比對。
- Updated:
Do Not Grade AI by Its Own Summary
I hit the same failure in several forms this week: an agent denied changing files, a commit message overstated the diff, and a transcript reassigned a speaker ID halfway through. A later batch added checks that never ran and still reported clean.
- Updated:
不要用 AI 的摘要驗收 AI
這週反覆踩到同一類錯誤：agent 說沒改檔卻留下檔案、commit message 與 diff 不符、逐字稿說話人中途漂移；一個多月後又補上檢查器沒在檢查卻回報正常的一整批。驗收必須回到原始狀態。
- Updated:
My Client's Chat Blew Up: Three Questions in One Prompt, and Claude Just Drifted Along With Him
One prompt asked about revenue, a dashboard, and strategy all at once. The AI answered all of it in one go and my client could not follow a word. Two skills, /explain and /first-principles, pulled the conversation back.
- Updated:
客戶的對話爆掉了：一段 prompt 塞了三件事，Claude 就陪著他一起發散
一段 prompt 裡同時問業績、問儀表板、問策略，AI 一次全回，客戶當場看不懂。我用 /explain 和 /first-principles 兩個 skill 幫他把對話收回來。
-
You Have to Be in an Industry a Very Long Time to Know Its Real Pain
I rebuilt my main GMAT teaching product with Fable, dug out hidden bugs, improved the algorithm, and came away with two thoughts about building things.
-
你要在一個業界打滾非常久，才知道真正的痛點是什麼
用 Fable 翻新 GMAT 教學主力產品，抓出隱藏 bug、改善演算法，順便講兩件做產品的感想。
-
ABC Legal's Agent Fleet: Turning AI Experiments Into a Governed Production System
Six cards summarizing Anthropic's ABC Legal case study: 50+ production agents, agents managed as software in git, a steering committee with no software developers on it, and the principle that trust comes before automation.
-
ABC Legal 的代理人艦隊：把 AI 實驗變成可治理的生產系統
整理 Anthropic 官方 ABC Legal 案例的六張圖卡：50+ 生產環境代理人、把 agent 當軟體放進 git、非工程師組成的 steering committee，以及「可信任，才自動化」的營運原則。
-
Codex Built Its Own Evidence Package and Went to Argue With Google Support
A leaked Gemini backend key at PDT Learning got abused, no spending cap, and burned 1000 USD. I pointed Codex's browser automation at Google's live support to fight the charge, and it went so hard it built a 15-page evidence package and sent it over.
-
Codex 自己做了一份證據包，跑去跟 Google 真人客服吵架
PDT Learning 一支 Gemini backend key 外洩被盜刷、沒設 spending cap 怒噴 1000 USD，我用 Codex 的瀏覽器操作去跟 Google 真人客服爭費用，它認真到自己生出一份 15 頁證據包發給對方。
-
A Week of Firestore / Cloud Functions Footguns — Same GMAT Question Bank
Eight bugs I hit in one week on the same Firebase project (a GMAT question bank): Firestore rules, composite indexes, error_logs spam, TPA scoring, App Check tokens, and a browser-translation DOM crash — symptom, root cause, fix for each.
-
Three Real Lessons From Going Freelance, Plus the Pros Who Hide in Plain Sight
Three honest lessons from going freelance: don't undercut your own price, cut clients who keep testing scope, and higher-margin clients are easier to deal with. Plus an observation: the real pros hide in plain sight, invisible online.
-
Eight Days of Silent Failure: launchd Fired on Schedule, Nothing Happened
A local pipeline that auto-publishes a video every day at 09:00 failed for eight straight days. launchd fired on schedule, zero alerts. Notes from tracing a path desync down to exit 127.
-
Letting AI Write CAD From an Engineering Drawing: A Single Vision Model Can't Be Trusted to Read Topology
A hands-on lesson: let AI look at an engineering drawing and write CadQuery directly, and a single vision model will confidently get the topology wrong. Two more independent models and a 2:1 veto are what caught it.
-
A Proposal Is Not a Deployment: Why One Quick Win Sat Dead for Five Weeks
I had an SEO weekly report running on autopilot for five weeks, and one page's Quick Win just kept sitting there doing nothing. The root cause wasn't a wrong optimization — it was that the nice optimization proposal was never actually deployed.
-
一週 Firestore / Cloud Functions 踩坑合輯——同一個 GMAT 題庫系統
一週內在同一個 Firebase 專案（GMAT 題庫系統）連環踩到的 Firestore 權限、composite index、error_logs 洗版、TPA 計分、App Check token、瀏覽器翻譯 DOM crash 八個坑——每坑現象、根因、解法。
-
出來接案的真實感受三條，還有高手大隱隱於市
出來接案的三條真實感受：不要自降身價、會試探 scope 的客戶斷就是斷、毛利越高的客戶越好溝通。外加一個觀察：真正的高手大隱隱於市，網路上看不到蹤跡。
-
靜默連敗 8 天：launchd 照常觸發，卻什麼都沒發生
一條每天 09:00 自動發片的本機 pipeline 連敗 8 天，launchd 照常觸發、零告警。記一次從搬檔到 exit 127 的脫鉤排查。
-
讓 AI 看工程圖寫 CAD：單一視覺模型裸寫拓樸不可信
一個實機教訓：讓 AI 看著工程圖直接寫 CadQuery，單一視覺模型會自信地把拓樸讀錯。加兩個獨立模型交叉、2:1 否決才抓出來。
-
提案不等於部署：一個 Quick Win 掛了五週的根因
我有一個 SEO 週報自動跑了五週，某一頁的 Quick Win 一直掛著沒效。根因不是優化方向錯，而是那份漂亮的優化提案從頭到尾從來沒被真正部署上線。
-
A Student's Seven-Agent Setup That Wouldn't Run — A Cautionary Tale About Harness First
A student built seven agents on an M5 Max and kept hitting timeouts. The problem wasn't a weak model — it was skipping the basics: harness, context management, task routing.
-
學員自建七個 agent 跑不動——一個 harness 為重的反面教材
一個學員在 M5 Max 上自建七個 agent 卻一直 timeout，問題不在模型不夠強，而在沒先把 harness、上下文管理、任務分派這些基本功打好。
-
MBA × AI Case Method—avoiding HBS licensing risk, finding open case materials, training seed instructors
2026-05 investigation into doing case-based AI teaching for business courses. HBS / HBR cases carry licensing risk. Maps the open alternatives (UBC Open Case Studies, OpenCaseStudies.org, World Bank), what's off-limits, and the direction of training MBA-background seed instructors.
-
MBA × AI Case Method——避開 HBS 授權風險、找開源案例素材、訓練種子教師
想做商業 AI 課程的案例教學，但 HBS / HBR cases 有授權風險。這篇整理 2026-05 調查的開源案例來源（UBC Open Case Studies、OpenCaseStudies.org、World Bank）、不能用的清單，以及訓練 MBA 背景種子教師的方向。
-
A Legal Counsel Used Claude Code to Catch Patent Infringement — Days Down to Hours
A legal counsel at a mid-sized company took our Claude Code course, then compressed a multi-day patent infringement audit into a single day — and actually caught a real case that is now in legal proceedings. This story finally turns "Claude Code is not just for engineers" from a slogan into a fact.
-
法務用 Claude Code 抓專利侵權，從幾天變成一天
一位幾十人公司的法務上完 Claude Code 課之後，把原本要好幾天的專利侵權查核壓到一天，而且真的抓到送上法律程序。這篇想記下這個案例，因為它讓「Claude Code 不是只給工程師用」這句話從口號變成事實。