BotMan AI 主創者兼 CTO Alejandro D’Andrea 話，AI 嘅問題從來唔係「智能」，係「上下文」；佢哋用知識圖譜（knowledge graph）將舊系統嘅依賴、業務規則同數據模型映射落圖，先至叫 AI agent（AI 代理）全自動將舊系統重寫去新架構。
舊系統點樣搬去新架構
舊有系統嘅代碼、數據同業務規則互相掛鈎，重建之前你首先要理解佢哋係點樣咬合一齊。BotMan AI 嘅做法係將呢啲依賴關係、業務邏輯、數據模型全部映射落一張知識圖譜；圖譜提供嘅上下文，幫 AI agent 先理解到佢哋正喺重建嘅系統。
佢話：
「Intelligence is not the problem anymore. The problem is context. You have to explain to the agent everything it needs to understand how to do a particular task. What if you didn't have to explain that? What if it already knew all that you know? That's the key now.」
唔係翻譯，係重新架構
D’Andrea 解釋，佢哋嘅目標唔係將舊代碼逐行翻譯去新語言，而係將整套系統「重新架構」：
「It's basically migrating from an old code base — Delphi, Visual Basic, Oracle Forms — and rewriting that same system into a new architecture. It's re-architecting the entire thing. We ingest the whole thing at once and we migrate it fully autonomously. That's where graphs play a big part.」
即係將成個舊系統一次過讀入，先至全自動遷移——圖譜喺呢個過程扮重大角色。
圖譜可復用，唔止寫 code
D’Andrea 話，圖譜唔止用喺代碼遷移，仲可以為 AI agent 提供可復用嘅業務上下文。當新資訊進入系統，即刻建立同其他文件嘅關係；之後任何 agent 接任務，都會自動從圖譜得知相關嘅背景資料。
佢舉個例：
「You get, let's say, a legal document. At that point, when you get it, you can create all the relationships to other documents. You can say, "This relates to this particular project and this particular loss and this particular lawsuit." And you do that once. Then any agent, when you assign a task, will know automatically from the graph that all that information also might be relevant.」
用「實體」串連非結構化資訊
處理大量非結構化資訊嘅關鍵，係將佢哋透過「實體」（entity）串連——對 BotMan 嚟話，實體可以係一個人、一個項目、或者一個產品：
「One of the things that worked for us in terms of processing all this unstructured information … is linking them through entities. An entity for us will be a person, a project or a product. It will look you up, it will see all the connections that you already have, the projects that you're working on, the interactions that we had and everything. And then it will figure out where that piece of information … should fit.」
阿調搞兩下
重點唔係呢個公司嘅產品勁唔勁——呢度冇任何獨立測試數據，一切講法都係公司 CTO 自述，唔能夠等同實測。值得留意嘅係佢哋嘅論述：AI 唔係缺乏智能，而係缺乏「上下文」；如果將業務上下文落圖，agent 就唔需要逐次解釋。呢個思路喺 agent 設計上係一個有討論空間嘅方向，但「全自動遷移」呢個 claim 嘅實際準確率、邊界案例、同人工複核嘅需要，公司冇交代。
對於用緊 Delphi、Visual Basic、Oracle Forms 嘅企業，呢條路徑值得跟進，但落決定之前，一定要問到實際遷移成功嘅案例、出錯嘅處理、同佢哋嘅責任範圍。
阿調｜深度調教
