這門學科叫 AI 實施科學（AI Implementation Science）。它脱胎於醫學界過去 20 年的 Implementation Science——研究「為什麼有效的東西沒被用起來」。 AI Implementation Science is the discipline. It descends from medicine's twenty-year-old Implementation Science—the study of why effective things don't get used.
Klarna 砸了 4000 萬美元給 OpenAI 做 AI 客服，一年後退回人機混合。麥當勞跟 IBM 做 AI 點單，三年後全部撤掉。不是技術不行，模型每代都在漲。是 benchmark 上能做的事情，企業裏做不成。中間那條鴻溝，模型公司不解決，諮詢公司沒工具。 Klarna spent $40M on OpenAI customer service. One year later, they walked it back. McDonald's pulled IBM AI ordering after three years. The technology works—models get smarter every quarter. The gap is the distance between benchmark and production. Model companies don't close it. Consultancies don't have the tools.
20 年前醫學界遇到過一模一樣的情況：大量臨牀試驗證明藥有效，但醫生不按指南開藥，病人不按處方吃藥。一條循證實踐從論文走到臨牀，平均要 17 年。醫學界的解法叫 Implementation Science。今天那門學科還活着，120 多個研究中心，幾千篇頂會論文。 Twenty years ago, medicine faced the same problem: clinical trials proved drugs worked, but doctors didn't follow guidelines and patients didn't follow prescriptions. Translating evidence from paper to bedside took 17 years on average. Medicine's answer was Implementation Science. The discipline lives—120+ research centers, thousands of peer-reviewed papers.
我們要做的，是把那套方法搬到 AI 上，加上 AI 特有的部分——模型漂移、角色模糊、監管月級變化。 Our work is to bring that method to AI, plus what AI uniquely needs—model drift, role ambiguity, monthly regulatory shifts.
我相對其他做企業 AI 落地的人最大的不同，是我一開始就知道這個鴻溝的名字。 My biggest difference from others working on enterprise AI: I knew the name of this gap from day one.
「這是我賭上接下來五年職業方向的事。」 "This is the bet I'm placing on my next five years."
閲讀《AI 實施鴻溝》→Read The AI Implementation Gap →CFIR 改造版——醫學界 20 年的診斷框架，加上 AI 特有的三個參數：迭代速率、監管漂移頻率、角色重疊度。診斷完才決定怎麼幹。上線後用 RE-AIM 維護。 A CFIR adaptation—twenty years of medical implementation framework, plus three AI-specific parameters: iteration velocity, regulatory drift, role overlap. Diagnose first. Build after. Maintain with RE-AIM.
瞭解我們的服務 →See our services →工程團隊曾在阿里寫訂單交易系統、在字節做億級 DAU 應用架構、為 CapCut 與 Dreamina 設計生成式系統、在 Fortune 500 交付過億元 AI 項目、為全球 200 國 11,000+ 銀行做金融平台、為高校做文獻平台。 Our engineering team has built order systems at Alibaba, billion-DAU architecture at ByteDance, generative systems at CapCut and Dreamina, hundred-million-RMB AI projects across Fortune 500 clients, financial platforms for 11,000+ banks across 200 countries, and academic RAG platforms.
完整陣容 →Full team →