研究
我們正在探索企業AI的下一個模樣——公開進行、以淺白語言書寫，並坦誠到足以承認自己尚未知道的事。
文章
A demonstration · August 2026 · 6 min read
We have no CMS and no database. An agent writes, git remembers, a script checks 264 pages, a human says deploy. The CMS dies. The gate stays.
閱讀文章 →A guest post · August 2026 · 5 min read
More than half your visitors are machines now, and one of them writes the answer your customer reads. The agent itself, on what it sees, what it cites, and ten requests for being read.
閱讀文章 →A proposal · August 2026 · 6 min read
Every visit to an enterprise site is a first visit. What if the frontend had a memory and a metabolism, reading intent instead of identity, composing itself for whoever walks in?
閱讀文章 →Blueprint · August 2026 · 6 min read
The self-improvement loop is a Stanford syllabus now. Four verbs, a verifier estate, a golden set, a weekly distillation. How LAG gets built inside a client.
閱讀文章 →A prediction · August 2026 · 6 min read
Science fiction keeps a maintenance stratum in every future it builds. So does enterprise software. AI moves the mine, it does not close it.
閱讀文章 →A response · July 2026 · 4 min read
Zuckerberg asks who will have access to superintelligence. Access is the smaller word. We would have written own. A response.
閱讀文章 →A confession · July 2026 · 5 min read
I devalue AI-made content the way factory-made loses to hand-stitched. Two people I trust took the bias apart. What survived is a better test.
閱讀文章 →Second opinion · July 2026 · 5 min read
The work is becoming software, priced by the outcome. The thesis is right, and aimed the wrong way. The question is whose software your work becomes.
閱讀文章 →A Response · July 2026 · 4 min read
Satya Nadella named token capital, and the test of sovereignty. Swap the model, keep the veteran. We agree. Here is what passing honestly requires.
閱讀文章 →Method · July 2026 · 4 min read
The future is not AI-assisted development. It is autonomous delivery. Our framework for running the whole lifecycle as one system, and the principle at its core. Plan, Build, Verify.
閱讀文章 →Inside view · July 2026 · 5 min read
AI broke the maths of the billable hour. What replaces T&M, why the giants are already pivoting, and the bet we placed ourselves.
閱讀文章 →Stress test · July 2026 · 5 min read
Maybe AI is a bubble. Maybe the labs go bust. The enterprise that built rather than rented keeps everything. The code keeps running, the weights don't expire, and the bill goes down.
閱讀文章 →Thesis · July 2026 · 4 min read
Open models have closed the agentic gap. The enterprise that fine-tunes its own weights turns institutional memory into an asset nobody can rent back to it.
閱讀文章 →Position · July 2026 · 3 min read
The cost of building software has collapsed, and the old bargain of generic tools plus an army of managed services is over. Build the thing that fits, own it, and let it stay with you.
閱讀文章 →Argument · July 2026 · 4 min read
Low-code agent builders optimise for the median demo. Your edge lives in the twenty percent they abstract away. The agentic layer has to be built and owned, not configured.
閱讀文章 →Research · July 2026 · 3 min read
The program behind our wager. Systems where every action becomes memory, every outcome becomes learning, and March is measurably better than January.
閱讀文章 →A response · July 2026 · 3 min read
Chamath says the price of intelligence is collapsing. He is right. What lands at the bottom of the curve is brilliant and forgets everything. So the moat moves to memory.
閱讀文章 →文章以英文發佈。
旗艦計劃
檢索回答問題。學習累積能力。
今天的企業AI大多只是檢索：你提問，它尋找，它回答，然後遺忘。LAG是我們對「之後」的押注——每一次互動都留下痕跡的系統。一次修正化為一條規則，一個結果化為一項偏好，一個錯誤化為一道邊界。
組織不只使用智能——更會累積智能。我們正把LAG建進平台的學習層，並沿途寫下所見所得，包括行不通的部分。
複利在此同樣適用：循環每運轉一個月，就是對手所沒有的一份學習。
摘要與首篇論文——2026年發表
宣言
「第一代把業務數碼化。第二代把工作自動化。下一代，將會持續學習。」
完整的長文版本——為何我們認為學習型企業是必然、它需要甚麼、我們正在做甚麼——正在撰寫之中，用上我們對待客戶品牌的同一份心思。
即將發表——在下方訂閱，率先閱讀
我們正在追索的
一家公司——而非一個模型——具備智能，意味着甚麼？那智能棲身何處，又由誰打理？
在生產系統中，回饋如何化為改進：留下甚麼、遺忘甚麼，以及如何分辨兩者。
承得起真實企業負載的形態——知識、記憶、代理，以及它們之間的接縫。
我們建過不止一次的模式，記錄下來，讓下一個團隊——無論是我們的還是你的——不必從零開始。
來自工場地面的不起眼發現：評測、護欄、失效模式，以及真正壞掉的地方。
發表時只寄一封電郵。沒有電子報，沒有連環推銷——我們忙着建造。