Building a Second Mind for Legal Work

Hanyi, a dispute lawyer with ten years of experience, started with the words “AI LAWYER” on the screen. Then they blinked out. “That’s my crisis,” he joked. For a lawyer, the term suggests replacement, an existential threat that he said keeps him and his batchmates awake at night.

On this Friday afternoon, a senior dispute lawyer and a computer science undergraduate are asking the same fundamental question about their craft. Hanyi’s answer isn’t to fight the technology. It’s to reframe the role entirely.

“So, I prefer now to think of AI lawyer, not so much as artificial intelligence lawyer, but authentically integrated lawyer,” he offered, before adding with a laugh, “So, that is a bit lame, I know.”

Know-What, Know-How, and the Hidden 80%

Hanyi’s argument rests on a distinction between two types of knowledge. The first is “know-what,” the explicit information in textbooks and manuals. This is the general knowledge of contract law, maybe 20% of a professional’s value, and the part most vulnerable to automation.

The other 80% is the “know-how.” This is the tacit knowledge below the waterline: the skill, judgment, and experience gained from practice. It is the craft. This know-how shows up when advising a client that while a certain action is legal, it might hurt their reputation or bring other commercial consequences. This is the human part of the profession. The journey from existential crisis to authentic integration begins by recognizing this hidden 80%. The challenge is not to protect it from AI, but to use AI to extract it, structure it, and build with it.

The room he is speaking to is a mix of legal professionals and AI builders. Lawyers and recent law graduates are trying to understand the implications of AI for their sector. Around them, builders are working on enterprise AI for finance and accounting, manufacturing, and environmental diagnostics, alongside founders exploring personal agents and second-brain tools. The trade-off Hanyi is about to name has already been made, in different ways, by most of the people listening.

Building a Verifiable Second Mind

The journey to structuring know-how often starts with a practical problem. For Hanyi, it was a High Court trial involving 30 gigabytes of evidence. “If I were to bring this out,” he told the judge, “it would fill two of those rooms.” That experience pushed him to build a stack, starting with OCR and search tools.

He brings up a very interesting point that is probably lost to non-domain experts: In law, the structure of the documents convey valuable meaning and must be preserved. Simply feeding PDFs or even clean Markdown into a model is not enough, as important context gets lost. How the clauses are nested and structured within a document give experienced lawyers a lot of context of how to interprete the document.

He is experimenting with translating know-how into formats like SQLite or knowledge graphs, because legal meaning often lives in the structure and relationships within data, not just the words.

This process enables a critical framework for human verification, which Hanyi breaks down into two parts:

• Provenance asks: where does this assertion come from? It must be possible to trace every claim to a specific document or statute. This can often be checked with a script.

• Support: Support is harder. It asks if the reasoning connecting the source to the assertion is sound. This is where human judgment remains essential, because reasoning can go wrong even when sources are provided.

This is why fishing for evidence, as he calls it, is so tricky. He joked about lawyers who fish for quotes out of context. But he also uses the metaphor positively: good legal research is a form of fishing, where know-how helps you find the right spot and use the right net. His AI-assisted workflow uses Boolean search queries as a kind of scientific fishing to find what matters.

One person in the room highlighted this tension. “I was going to ask you,” he said, “if in the back end, you had to make more refinements, maybe with RAG or GraphRAG or other architectures... because in law, I think you need very deterministic outputs.”

Hanyi’s response was immediate: “Yeah, I think you are exactly right.” True verification is not just about finding keywords. It is about preserving the structural meaning of the data so that a lawyer can validate the AI’s reasoning.

From Automation to New Capabilities

What stood out for this sharing was that Hanyi was not obsessed purely with automating existing work. It was his focus on innovation: giving lawyers and law firms new tools and creating entirely new capabilities. This is possible at 2 levels.

At the first level, by codifying and directing their know-how, a lawyer can also become a designer, a teacher, or a project manager.

Hanyi gave an example of learning from a friend who works in UX. “I gave it to my UX friend, so he gave me what the standards are,” he explained. “And I don’t know how to execute, but I extracted the standards from who I trust. Then, I pretend to be the UX GM. I go to AI and say, follow this experience for me.” The human provides the expert judgment and world-class standards, and the AI handles the execution.

At the second level, Hanyi references legal scholar Richard Susskind, who argues lawyers should focus on more than efficiency. The real opportunity is to innovate for clients and prevent problems from happening in the first place, to “really serve society, really try to make for a more just society.”

Hanyi shares an example that frequently arise in his practice:

“In construction law… you have a construction contract. Maybe the project is for $100 million in lump sum… usually in construction projects, there’s always overruns… delays… extra work. Then the client comes to him saying they have a $50 million claim. Sometimes what was found was that the claim was not just 50, but 60-70 million, but it went unnoticed because they had done extra work but had not properly protected their rights as a project manager / contract manager."

In this case of construction disputes, the legal claim often depends on project administration that happened months earlier. Variations, delays, extra work, notices, instructions, site correspondence, payment certs, contract clauses. If the contractor didn’t manage that trail properly, the legal right may still exist, but the evidence and entitlement pathway are messy or weakened.

My read of his deeper point is this: an AI-assisted lawyer is not only faster at dispute work after the fight starts. The lawyer, in the age of AI, could even provide preventive legal ops as a service. Take the construction contract and project documents while the project is live, structure them, and flag: “You’re doing extra work here; this may need a notice / variation claim / record / reservation of rights.”

This turns the lawyer from post-failure litigator into a system that helps the client preserve claims before they become disputes.

The $50m to $60-70m example is really a “hidden entitlement” story. The client comes in with the claim they can see. Hanyi looks through the documents and sees unclaimed extra work because the client was not acting like a good project manager or contract manager. The lawyer or the firm, with the assistance of AI, can now make that kind of document-aware, workflow-aware detection scalable.

The strategic question is no longer about being replaced. It is about building a second mind. It’s about taking the craft that lives in your head, structuring it in the world, and having it work together with your mind to become a better lawyer.

---

SQ Collective hosts Coworking Fridays for founders, operators, and AI builders working through real product questions in Singapore.

Join an upcoming Coworking Friday: https://lu.ma/sqc-friday Explore SQ Collective: https://www.sq-collective.com

Michael