I ran the company. I owned the P&L. I deployed $100M+ learning what actually moves a business — and what only looks like it does. Now I build the AI systems that do it. Not pilots. Not decks. Systems that change the decision. 我经营过公司，背过 P&L，部署过 一亿美元以上——因此我知道什么真的能推动一家企业，什么只是看起来能。现在，我亲手构建能做到这件事的 AI 系统。不是试点，不是 PPT，是真正改变决策的系统。
Every enterprise now has a working demo. Almost none has a changed P&L. The gap is not the model — the models are good enough. 今天每家企业都有一个能跑的 demo，却几乎没有一家因此改变了 P&L。缺口不在模型——模型已经足够好了。
The gap is that the two halves of the problem live in different rooms. The people who understand what the business actually needs have never shipped a system. The people who can ship have never been accountable for the outcome. 缺口在于：这个问题的两半，住在两个不同的房间里。真正懂企业需要什么的人，没有亲手交付过系统；能交付系统的人，从未为结果负过责。
I've been in both rooms. That is the entire proposition. 这两个房间，我都待过。这就是我的全部主张。
Putting a reviewer behind a model makes the output safer. It does not make the company smarter. What changes a company is when executive judgment — the thing that decides what matters — becomes part of the system's loop, and improves through use. 在模型后面放一个审核员，只让输出更安全，不会让公司更聪明。真正改变一家公司的，是让"经营判断"——那个决定什么重要的东西——本身进入系统的回路，并在使用中不断变好。
Fails as: dashboards nobody acts on.典型失效：没人据此行动的看板。
Fails as: AI that recommends, and no one who is accountable.典型失效：AI 给了建议，却没有人为它负责。
Fails as: a pilot that never enters the operating cadence.典型失效：永远进不了经营节奏的试点。
Fails as: nobody measures whether the decision was right.典型失效：没有人回头衡量那个决定是否正确。
Product teams have more signals than ever and no dependable mechanism for deciding what matters, where the opportunity is, or why a concept deserves investment. The cost is not a bad decision — it is a year spent on the wrong one. 产品团队从未有过这么多信号，却没有一套可靠机制来判断：什么重要、机会在哪里、为什么这个概念值得投资。代价不是一个错误决定，而是一整年花在错误的方向上。
I framed the product thesis, designed the operating workflow and architecture, built the system end to end, worked directly with the client, and extracted reusable software foundations after delivery. 我框定产品主张，设计运行流程与架构，端到端亲手构建系统，直接对接客户，并在交付后抽取出可复用的软件基座。
The moat is not the crawler or the model. It is the closed loop that turns evidence into decisions — and improves through use. 护城河不是爬虫，也不是模型，而是那个把证据变成决定、并在使用中不断变好的闭环。
A read-only mobile collection system using ADB and multimodal models to turn any app or mini-program screen into structured, reviewable product data — so category decisions rest on evidence instead of anecdote.基于 ADB + 多模态大模型的只读移动端采集系统，把任意 App/小程序界面变成结构化、可复核的商品数据——让品类决策建立在证据上，而不是传闻上。
Not a feature and not a pilot: the product, the workflow, and the operating cadence of the team redesigned together around AI — the full Executive-in-the-Loop model applied to a live business.不是加一个功能，也不是做一个试点：产品、流程、团队的运行节奏被一起围绕 AI 重新设计——Executive-in-the-Loop 模型在一家真实企业上的完整落地。
Agent workflows built around source traceability, structured analysis, and human confirmation — designed so an analyst can defend every sentence to an investment committee.围绕来源可追溯、结构化分析与人工确认构建的 Agent 工作流——设计目标是让分析师能为每一句话向投委会负责。
Turns hours of financial audio into transcribed, speaker-attributed, searchable evidence with quality gates at every stage — the unglamorous plumbing that makes downstream analysis trustworthy.把数小时金融音频转成带说话人归属、可检索的证据，每一环都有质量闸门——正是这些不起眼的管道，让下游分析变得可信。
A pipeline that holds character voice, pacing, and tone consistent across hundreds of thousands of words — the hard part is not synthesis, it is consistency at length.在数十万字的长度上保持角色声线、节奏与语气一致的流水线——难点从来不是合成，而是长度上的一致性。
Replaced a manual, frequently disputed royalty process with an auditable rules engine — so the company could sign artists it previously could not settle with.用可审计的规则引擎替换掉人工、频繁争议的版税流程——让这家公司能签下过去因为结算不清而签不了的艺人。
Scoring and recommendation logic that turns fragmented consumer and category signal into a ranked, defensible new-product shortlist the commercial team will actually commit to.把零散的消费者与品类信号，变成一份有排序、经得起追问、商业团队真的敢承诺的新品清单。
Separating symptoms from root causes, then translating the diagnosis into a prioritized set of operating interventions — the work that decides whether an AI programme is worth starting at all.把症状与根因分开，再把诊断翻译成有优先级的经营干预方案——这项工作决定了一个 AI 计划到底值不值得启动。
Every exceptional operator carries a small set of questions they ask before they decide. Munger inverts. Grove asks what an incoming CEO would do. Bezos asks whether the door swings both ways. Those questions — not their answers — are the actual asset. 每一个卓越的经营者，身上都带着一小组"决定之前必问"的问题。芒格做逆向推演；格鲁夫问"新来的 CEO 会怎么做"；贝索斯问"这扇门是否双向"。真正的资产是那些问题，而不是他们的答案。
They are more portable than any conclusion and almost never written down anywhere a working team can reach at the moment of decision. Wisdom OS distills them into modes a team can invoke — and puts them in the room where the call is actually made. 它们比任何结论都更可迁移，却几乎从未被写在一个团队在做决定的那一刻能够触达的地方。Wisdom OS 把它们蒸馏成可以被调用的"思维模式"，并把它们放进真正做决定的那个房间里。
“What would guarantee failure here? Now avoid exactly that.”「什么会保证这件事失败？那就精确地避开它。」
“If we were replaced tomorrow, what would our successors do first?”「如果我们明天被换掉，接任者上来第一件事会做什么？」
“Is this reversible? Then stop deliberating and decide.”「这个决定可逆吗？可逆就别再研究了，做。」
“What job is the customer hiring this for — and what are they firing?”「客户到底雇这个产品来做什么活？他们又在解雇谁？」
“Explain it without the jargon. If you can't, you don't understand it.”「不用行话把它讲一遍。讲不出来，就是没搞懂。」
“Forget our story. What happened to everyone else who tried this?”「先忘掉我们的故事。所有试过这件事的人，后来怎么样了？」
“What breaks this — and what would gain from the disorder?”「什么会击穿它？以及，什么会从这场混乱中获益？」
“Who owns the outcome of this decision — and does the system know?”「谁为这个决定的结果负责？以及，系统知道这件事吗？」
On method. Each mode is distilled from the published work of the person it is attributed to, and named as such. These are modes of thinking, not simulations of individuals — no mode claims to speak for anyone, and none is presented as a person you are talking to. The operator still makes the call; the mode only makes the question sharper. 关于方法。每一个模式都蒸馏自其署名者的公开著作与访谈，并明确标注出处。它们是思维模式，不是对个人的模拟——没有任何一个模式声称代表某人发言，也没有任何一个被呈现为"你正在与之对话的人"。做决定的仍然是经营者本人；模式只负责把问题问得更锋利。
Where AI actually moves your P&L — and where it only looks like it will.找出 AI 真正能动你 P&L 的地方——以及那些只是看起来能的地方。
One consequential decision. One system. In production, not in a slide.一个真正重要的决策，一套系统，上线运行，不在 PPT 里。
Your board and your team, fluent enough to lead this — not just to buy it.让董事会与团队真的懂到能领导这件事——而不只是采购它。
My AI work is grounded in lived responsibility for customers, P&L, product, frontline operations, organization, and risk. That is not a credential. It is the reason the systems are designed the way they are. 我的 AI 工作，扎根于对客户、P&L、产品、一线运营、组织与风险的真实承担。这不是一张资历证书，而是这些系统之所以被设计成这样的原因。
Premium mobility joint venture. Every decision on this list — pricing, fleet, service model, headcount, capital deployment — landed on one desk.高端出行合资公司。定价、车队、服务模式、编制、资本部署——所有这些决定，最后都落在同一张桌子上。
Multi-product-line growth across a 200,000+ driver network — operational complexity at a scale most software leaders never touch.跨产品线增长，覆盖 20 万＋司机网络——大多数软件领域的管理者从未接触过的运营复杂度。
Connected market research, customer insight, competitive analysis, and service strategy across China, the UK, and India.打通市场研究、客户洞察、竞品分析与服务战略，跨越中国、英国与印度。
Six Sigma Green Belt; co-inventor on a transmission patent. Engineering discipline is where the rest of this was built on.六西格玛绿带，变速箱专利共同发明人。后来的一切，都建立在这份工程纪律之上。
The practical value is not the certificates. It is that I can sit with a founder in Shanghai, a board in Singapore, and an engineering team in San Francisco in the same week — and say something useful in each room. 真正的价值不在证书。而在于我可以在同一周里，坐进上海的创始人办公室、新加坡的董事会、旧金山的工程团队——并且在每一个房间里都说得出有用的话。
If you are a CEO or an executive team facing a decision where AI genuinely matters — not a tool purchase, a decision — that is the conversation I want. 如果你是一位 CEO，或者一个正面对"AI 真正会影响结果"的决定的高管团队——不是采购一个工具，而是一个决定——那正是我想谈的对话。
Also open to a small number of operating roles in frontier AI. More → 同时对少数前沿 AI 公司的经营岗位保持开放。了解更多 →