Tag: ai-workflow
All the articles with the tag "ai-workflow".
-
I'm Stuck at Stage 2.5 of AI Adoption
After watching Boris Cherny break down the stages of AI adoption, it hit home: I'm stuck at stage 2.5, where limited attention plus low trust becomes a vicious cycle. This week I used late-night schedules, a morning dashboard, and herdr auto-spawning sessions to cut a 4-5 hour workflow down to 1-2 hours.
-
我卡在 AI 導入的 2.5 階段
看了 Boris Cherny 分享的 AI 導入階段框架後有感而發：我卡在 2.5 階段，注意力有限加信任不足變成惡性循環，這週靠深夜排程、晨間儀表板、herdr 自動開 session 把 4-5 小時的協作壓到 1-2 小時。
-
A Few Days Into Switching From Claude Code to Codex: The Quota Honeymoon
A running log of switching from Claude Code to Codex this week: quota I could not burn through fast enough, a $20 Sol Ultra run that beat a $100 Fable plan on a professor friend's paper, and a tool-chain swap to Hyperframe.
-
從 Claude Code 換到 Codex 的這幾天：額度蜜月期
這週從 Claude Code 換到 Codex 的體感記錄：額度多到花不完、Sol Ultra 20 鎂幫教授跑論文，比 Fable 100 鎂方案還划算，工具鏈也跟著換成 Hyperframe。
-
Picking a Brand Look as a Non-Designer: AI Made Prototypes Cheap, So Taste Became the Hard Part
I am not a designer, but I spent this week iterating on a visual system for AgentCrew Academy with Fable, GPT-image-2, and GPT-5.6-sol. Here are the four rejected directions and the final spec applied to the website, slides, and documents.
-
外行人選品牌視覺：AI 把提案做便宜了，難的變成你的品味
我不是設計師，這週用 Fable、GPT-image-2、GPT-5.6-sol 迭代出 AgentCrew Academy 的視覺系統，記下被否決的四個方向與最後套用到網站、投影片、文件的規格。
-
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.
-
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。驗收必須回到原始狀態。
-
"Automation" Isn't the Answer to Most Things — and How I Made This Video
Why do the people teaching AI keep bringing up "automation"? I don't think automation is the answer to everything, or even to most things. Plus a few words on how I made Ep27.
-
Claude Code To-Do Tiering: Auto / Review / Collaborate / Manual
A small discovery from using Claude Code: sort to-dos by repo, then into four tiers — auto, review, collaborate, manual. Stuff that used to drag on all day now finishes in an hour.
-
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.
-
A Week of Making Tutorial Videos With Claude Code: Remotion Animation and ffmpeg Pitfalls
Production pitfalls from a week of making two tutorial videos: subtitle restraint, font size, animation anchor auditing, Remotion render crashes, normalizing quiet narration, and opencc over-localization.
-
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.
-
I Built a "Tinghao SKILL" — Distilling a Finance YouTuber's Perspective Into AI
I spent my leftover Claude quota building a Tinghao SKILL: 17 finance episodes distilled, 3-4 rounds of blind testing, and a curious difference in how AI and humans deploy dirty jokes.
-
「自動化」不是大部分事情的解方——以及我怎麼拍這部影片
為什麼那些教 AI 的動不動就提「自動化」？我覺得自動化不是所有事情，甚至不是大部分事情的解方。順便聊聊我拍 Ep27 時的取捨。
-
Claude Code 待辦分級制：自動／審核／協作／手動四級分流
一個用 Claude Code 的小發現：把待辦先分 repo 再分成自動／審核／協作／手動四級，原本拖一整天的事 1 小時內就做完了。
-
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。
-
用 Claude Code 做教學影片一週踩到的坑——Remotion 動畫與 ffmpeg 後製
這一週做兩部教學影片踩到的製作坑：字幕克制、字級、動畫錨點稽核、Remotion render crash 拆段、安靜旁白正規化、opencc 過度在地化。
-
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。
-
我做了一個「庭皓 SKILL」——把財經 YouTuber 的觀點蒸餾成 AI
用沒燒完的閒 token 做了一個庭皓 SKILL：蒸餾 17 部財經速解讀、3-4 輪盲測驗證，還發現 AI 跟人類講黃段子的邏輯不太一樣。
-
The Blind Spot in AI Self-Verification — Why You Need an Uninformed Agent to Audit
AI is great at "looking done." To catch its blind spots, you can't rely on it checking its own work — you bring in an auditor that knows nothing, inherits no context, and gets only the rules and the output.
-
AI 自我驗證的盲點——為什麼要請一個不知情的 agent 來稽核
AI 很會「看起來完成了」。要抓出它的盲點，不能靠它自己回頭檢查——得引入一個不知情、不繼承上下文、只拿到原則與產出的稽核者。
-
Don't Brute-Force Sites That Need a Login—Your Token Is Your Account
Don't point an AI agent at sites that require a login. Your token is your account, and when the banhammer comes down you're gone. Pay a little for a third-party scraping API and offload the ban risk.
-
需要登入的網站別硬爬——你的 token 就是你的帳號
需要登入的網站，別叫 AI agent 去硬爬。你的 token 就是你的帳號，admin 一丟二向箔你就沒了。花點小錢買第三方爬蟲 API，把封號風險轉嫁出去。
-
The More Rules You Add, the Less Claude Listens — I Sent a Team of Agents to Trim My Setup and Cut 36% of Always-On Context
Late at night I got Opus 4.8, and my weekly quota happened to reset. The first thing I did was put my harness on a diet. A COO student who had burned through 88% of his 1M context made me face one thing squarely: the more rules you add, the less the model listens.
-
規則越加，Claude 越不聽話——派一隊 AI 重整設定，常態上下文省 36%
半夜拿到 Opus 4.8、額度剛好重置，我做的第一件事是減肥我的 harness。一個 COO 學員把 1M 上下文用到 88% 的案例，讓我認真面對一件事：規則加越多，模型反而越不聽話。
-
AI Tool Philosophy — Pick a Mindset, Not a Side
After making nearly thirty tutorial videos: tools change, thinking doesn't. Don't dismiss others' tool choices. What matters is the orchestration logic — task decomposition, validation planning, responsibility assignment.
-
AI 工具使用哲學——不選陣營，選思維
拍了快三十部教學影片後的感觸：工具會變，思維不變。不用否定別人的工具選擇，真正重要的是指揮的邏輯——拆分任務、規劃驗證、分配職責。
-
Claude Code Ecosystem Overview — Skills, Hooks, Agents in Full Bloom
Claude Code's ecosystem has grown from a simple CLI tool into a full platform: Skills supported across providers, Hooks intercepting commands, Agent View for multi-session management, and fork subagents with context inheritance.
-
Claude Code 生態系盤點——Skill、Hook、Agent 全面開花
Claude Code 的生態系已經從單純的 CLI 工具長成一個完整的平台：Skill 各家都支援、Hook 攔截指令、Agent View 多工管理、fork subagent 繼承上下文。連社群都開始用 Skill 搞創意了。
-
Turning Street Smarts into an AI Skill — When the Numbers Are Clear, You Stop Flinching
After turning Norm Brodsky's Street Smarts into a Claude Code Skill, every partnership opportunity gets a margin analysis first. Many appealing deals turn out to be money losers. When you know the numbers, you stop flinching.
-
把《街頭智慧》做成 AI Skill——算出來的數字就在那邊，波瀾不驚
把 Norm Brodsky《街頭智慧》做成 Claude Code Skill 後，每個合作案都先算毛利。看似很好的機會，算下去常常是賠錢。心中有底，對方再怎麼跑火車都不慌。
-
Cross-Model Review — Stop Letting AI Grade Its Own Homework
AI reviewing its own output has inherent blind spots. Using Codex for independent review or the Review Council skill to orchestrate a three-model expert team is the most pragmatic solution right now.
-
跨模型互審——讓 AI 不再自己審自己
AI 自己審查自己的產出，盲點難以避免。用 Codex 做獨立審查、或用 Review Council skill 組三模型 expert team，是目前最務實的解法。
-
Three moves to firepower-demo Agentic AI at your company—desktop cleanup, conflicting meeting notes, multi-version emails
If you want to make your manager or coworkers feel Agentic AI in one second—and then open the door to an internal training engagement—my three go-to moves are cleaning up messy desktops, reconciling conflicting meeting notes and emails, and producing multi-version drafts for different recipients. These three hit office workers right in the pain.
-
在公司火力展示 Agentic AI 的三招——桌面整理、會議紀錄、信件多版本
想在公司讓主管同事一秒有感、進而開內訓案的話，我自己最常用的三招是：整理桌面亂檔、整理相互衝突的會議紀錄跟信件、迅速生成不同版本的信件給主管同事客戶廠商。這三招對辦公室的人就夠有感了。
-
Why Obsidian is something special to me, and why I only made this video now
I'd been wanting to make this video for a long time, but kept putting it off. Obsidian's flexibility is so high that it grows into a different shape in every person's hands. The payoff isn't immediate either—it's only after the vault has grown for a while that you realize you can't go back.
-
Obsidian 為什麼是個特別的存在——我為什麼到現在才拍這部影片
這部影片我想做很久了，但是卻一直沒做，直到今天。Obsidian 的自由度太高，所以在每個人的手上會長成不同的樣子。他的成效又不是立刻就能看到的，而是越長越大你才會發現離不開它。
-
Covey's time management matrix, now with Claude Code—how each quadrant shifts
Covey's time management matrix from The 7 Habits of Highly Effective People splits time on two axes: important × urgent. Drop Claude Code into the picture and each quadrant's time allocation shifts. The long-term compounding from AI isn't in Q1, it's in Q2.
-
Covey 七個習慣的時間矩陣放上 Claude Code 之後——每一格的時間分配怎麼變
Covey《高效能人士的七個習慣》的時間矩陣是兩軸切四象限，重要 × 緊急。把 Claude Code 放進去之後，每一格的時間分配會被重新洗牌。Q1/Q3 被 AI 吃掉，多出來的時間搬到 Q2 才是 AI 協作的長期複利。
-
How to tune Opus 4.7's effort, and why I still keep non-coding tasks on Sonnet 4.6 medium
Opus 4.7's default adaptive effort is a disaster—it turns into a lazy emperor. You have to crank it up to high or xhigh to avoid that, but token consumption goes nuts. After testing, Sonnet 4.6 medium is the sweet spot for non-coding work.
-
Opus 4.7 的 effort 怎麼調，跟為什麼我非編程任務還是停在 Sonnet 4.6 medium
Opus 4.7 預設的 adaptive effort 是場災難，會變成偷懶摸魚皇帝。要開到 high 或 xhigh 才能避免，但 token 就開始狂燒。實測下來，非編程任務停在 Sonnet 4.6 medium 反而是最好的選擇。
-
1M context isn't about holding more — it's about managing better
Claude Code's context window grew from 200k to 1M. The instinct is to put more in. That instinct is wrong. Five session management operations, four strategy rules, and the two most common ways to break your own context.
-
1M Context 不是裝更多，而是管理更好：Claude Code Session Management 心法
Claude Code 的 context window 從 200k 長到 1M，但更大不代表更好用。這篇整理五種 session 管理操作（Continue / rewind / clear / compact / Subagents）、四大策略、以及最容易踩的兩個坑。
-
Claude Code this week: 1M context as a management problem, quota anxiety goes collective, SpaceX takes Cursor
Observations from 4/23–4/29: the right mental model for 1M context, Claude quota anxiety as a subscription design phenomenon, SpaceX beating Microsoft to Cursor for $60B, Claude 4.7 identifying a journalist from 125 words, and what happens when you let Claude Code DJ.
-
Two Worlds of Web-Based Claude Code Setup
Web Claude Code has two completely different runtimes — cloud VM and remote-control. Here is what I learned setting up both, the small bugs I hit, and why some user-level configs simply cannot be lifted to the cloud as-is.
-
See the Outside, Think the Inside: Two Claude Skills That Pair
I saw a competitor selling well and my first reaction was anxiety. But before letting anxiety win, I wanted to actually understand what it was. Lately I have been pairing two Claude Skills for this — H/V analysis to see the outside, first-principles to see the inside.
-
Claude Code 這週整理：1M Context 怎麼用、額度焦慮集體發作、Cursor 被 SpaceX 搶走
4/23–4/29 這週的觀察與整理：1M context window 的正確心態、Claude 額度焦慮進化成集體現象、SpaceX 搶在微軟前面拿下 Cursor、Claude 4.7 根據 125 字認出記者、live coding 做音樂的副作用。
-
網頁版 Claude Code 環境設定的兩個世界
網頁版 Claude Code 有兩種環境——雲端 VM 跟 Remote-control。這篇整理我實測下來的設定訣竅、會踩到的小 bug，以及為什麼有些設定不能無腦丟到雲端。
-
看清外面、想清裡面：兩個 Claude Skill 的分工
偶然看到一個競品賣得很好，第一反應是焦慮。但焦慮之前，我想先搞清楚它到底是什麼。最近用兩個 Claude Skill 把這件事做透——橫縱分析法看外面，第一性原理看裡面。
-
My Claude Code Automation Stack: /loop, Monitor Tool, the Skill System, Obsidian Reflection Reviews
From /loop clocking in for me at ERP, to Monitor Tool as a background event gatekeeper, to Google Map MCP saving browser clicks, to Obsidian + Mirror framework for weekly reflection. Here's my six-month Claude Code automation stack, plus a clarification of how commands and skills actually split.
-
我家的 Claude Code 自動化工作流：/loop、Monitor Tool、skill 體系、Obsidian 反思週報
從 /loop 指令幫我打 ERP 卡、Monitor Tool 做背景事件守門員、Google Map MCP 省瀏覽器操作、到 Obsidian 搭 Mirror 框架做反思週報——整理這半年我家 Claude Code 的自動化架構，順便釐清 commands 跟 skills 到底怎麼分。
-
Six Months with Claude Code: From First Affection to Hesitation, From Overtime to Detox, From Mandarin Circles to the Thai Community
From January to April, my state of mind with Claude Code went from "that PTT-generation affection for the CLI" to "20x hesitation" to "AI detox" to "practicing Thai at a Bangkok meetup." Here's the six-month arc, along with the nights agents worked overtime for me until 5 AM.
-
Claude Code 用了半年：從好感到猶豫、從加班到 detox、從中文圈到泰國社群
從 1 月到 4 月陪著 Claude Code 一起長大半年，我的心態從「PTT 世代對 CLI 的莫名好感」變成「20x 的猶豫」變成「AI detox」變成「泰國社群練泰語」。整理一下這半年的軌跡，連同那些對自己講越來越 nerdy、agent 幫我加班到凌晨五點的日子。
-
Two Months of Claude Code Harness Ops: Crashes, Zombies, Cache Misses, and Plan Mode Edges
From March to April, running Claude Code as a long-running harness surfaced a bunch of issues. Survival tactics during the outage, my hook for cleaning up zombie processes, the resume cache-miss fix in 2.1.90, the silent memory bloat in 2.1.100, and the Plan-Mode-question-density problem after 4.7.
-
Claude Code Harness 兩個月運維史：當機、殭屍、快取、Plan Mode 的邊邊角角
從 3 月到 4 月，Claude Code 作為長時間運轉的 harness，踩過的幾個點：當機時的求生策略、我自己寫的 hook 清殭屍進程、2.1.90 修好的 resume cache miss、2.1.100 的記憶體暴增修復、還有 Plan Mode 問題太多的 4.7 癥結。
-
Claude Code This Week: Offloading CLAUDE.md, Quota-Saving Tricks, Ghostty Click Fix
A handful of Claude Code settings and money-saving tricks I picked up this week — how Rules/Memory/Hooks offload CLAUDE.md, fixing cmd+click in Ghostty No Flicker mode, dispatching long-read work to Haiku for quota relief, and the cache difference between pasting images and editing prior messages after 4.7.
-
Claude Code 這週整理：分擔 CLAUDE.md 的進階設定、省 quota 技巧、Ghostty 點擊設定
本週 Claude Code 累積的幾個實用設定與省錢技巧——Rules/Memory/Hooks 如何分擔 CLAUDE.md、Ghostty No Flicker 模式下連結點擊的修法、長文本用 Haiku 分派省 quota、4.7 後貼圖片跟改訊息對快取的差別。
-
Mirror System: Weekly Reviews Without the Comfort
Normal weekly reviews let you make excuses for yourself. Mirror System doesn't offer reassurance — just feed your daily notes to Claude and watch it surface the patterns you've been avoiding.
-
Mirror System：用 Claude 做不帶安慰的週回顧
普通的週回顧，你會幫自己找理由。Mirror System 不要肯定，不要安慰，只要事實——把本週 daily notes 餵給 Claude，看它反照出你不想正視的模式。
-
Turning Books into Claude Skills: From Textbooks to Atomic Habits, the World Is Already On It
Turning books, frameworks, and methodologies into Claude Skills — making knowledge executable as interactive coaches. A roundup of popular examples and practical lessons.
-
把書變成 Claude Skill：從教材到原子習慣，國外已經在瘋了
把書籍、框架、方法論轉成 Claude Skill，讓知識變成可執行的互動式教練。國外的熱門案例整理和實作心得。
-
My AI Workflow: Convergence and Divergence
Distilling my AI collaboration philosophy into two modes -- convergence and divergence. Plus how I build a personal knowledge base, and my day-to-day practice of running multiple agents in parallel.
-
我的 AI 工作流：收斂與發散
把跟 AI 協作的心法整理成「收斂」與「發散」兩個模式。加上個人知識庫的建構方法，以及多 Agent 並行的日常實踐。