On April 25, our community Legal Tech Frontier held its 10th in-person event — this time at Peking University School of Transnational Law, widely regarded as one of China's most prestigious law schools. 110 corporate counsel showed up. They had to switch to a bigger classroom.
I joined remotely from Silicon Valley for the opening remarks. What follows are the key threads from the day — reorganized not by speaker, but by the questions that actually matter.
The three-layer AI ladder — and where most lawyers are stuck
Koki — our community’s APAC events lead and a well-known AI content creator in China — opened with a live poll, sorting the room into three tiers of AI usage:
Layer 1: Chat. Using tools like DeepSeek or Doubao for Q&A. Most of the room was here.
Layer 2: Knowledge and research. Using tools like NotebookLM or Claude for structured legal research, document analysis, and synthesis. A smaller group.
Layer 3: Building. Using Claude Code, Cursor, or similar tools to create custom workflows — not writing code in the traditional sense, but using natural language to make AI do exactly what you need. Only a handful.
The gap between Layer 1 and Layer 3 isn’t about technical skill. It’s about a mental shift: from using someone else’s product to building your own system.
A few lawyers in the room had already made that jump. One team was using Claude skills to package their contract review logic. Another had AI auto-completing not code, but contract clauses inside VS Code. These aren’t programmers. They’re lawyers who realized they don’t need to wait for a vendor to build what they need.
This is what “vibe coding” looks like in legal — using natural language to tell AI what to build, and letting it handle the engineering. It’s not a programmer’s privilege. If your work has repetitive tasks, you can automate them yourself.
Buy vs. build: why Mr. Yuan ditched legal AI products for Claude
This was the sharpest practical argument of the day.
Mr. Yuan — a legal counsel at a robotics company — laid out his approach: stop buying packaged legal AI products. Use a general-purpose model (Claude) plus custom-built skills instead.
His reasoning was blunt:
Once a legal AI product is packaged, you can’t see its workflow. You can’t adjust it to match your judgment. You can’t iterate on it when you learn something new. And most Chinese legal AI products are built on older open-source models — their capabilities can’t keep up.
Mr. Yuan demonstrated, live, how to build a contract review skill in Claude from scratch — starting from “I have no programming background, how do I create a skill?” Claude walked him through the entire process step by step.
He then added a visual interface where he could accept or reject each suggestion individually. He even built what he called a “penetration test” — having AI simulate a contract’s entire lifecycle from signing to execution to delivery, checking for issues at every stage.
Mr. Yuan’s line: “You only need to output your judgment, your knowledge, and your preferences. AI handles the rest.”
Every time he found something to improve, he updated the skill immediately. His knowledge accumulates. His tool evolves in sync. He called it “being your own AI product manager.”
This practice was the best answer to the “buy vs. build” question raised earlier in the day. Not buying an off-the-shelf legal AI product — but using general-purpose AI to build a system that belongs only to you.
The irreplaceable lawyer: what AI can’t do
While several speakers focused on how to use AI, Mr. Xiong — a corporate counsel at a major manufacturing company — made the opposite argument. And it was the most important point of the day.
Mr. Xiong’s thesis: in-house counsel’s real value isn’t reviewing contracts. It’s shaping facts before they become legal problems.
He illustrated this with real scenarios from cross-border operations. His company was negotiating a factory lease in Southeast Asia. The deal had stalled for months over email — the other side kept delaying, going back on terms, running out the clock. Mr. Xiong flew out with the factory GM, sat down face to face with the landlord, and closed everything in a week. Decision-makers in the room. Every clause settled on the spot. Their competitor’s deal, handled remotely, took far longer.
In another case, a colleague overseas was on the verge of quitting over a landlord who refused to cooperate on permit changes. He spent ninety minutes on a call with her and realized the problem wasn’t legal at all — it was about stress tolerance and communication approach.
The pattern: AI can draft the contract. AI can research the jurisdiction. AI can generate a first-pass regulatory analysis. But in complex commercial negotiations, cross-cultural communication, and building trust inside an organization — AI is useless.
Mr. Xiong’s advice: use AI to save time, then invest that time in the things that make you irreplaceable. Getting to the front lines of the business. Building influence with decision-makers. Saying yes strategically instead of reflexively saying no.
He also shared his own Claude workflow: using AI to quickly learn unfamiliar business terminology, do preliminary jurisdictional research (for example, price control regulations in an Eastern European country), and pre-screen contracts based on his risk preferences — with roughly 90% accuracy.
But Mr. Xiong’s point wasn’t about the tool. It was about what you do with the hours it gives back.
The roundtable: optimists, pessimists, and one uncomfortable silence
After the three presentations, the room opened up for discussion. A few exchanges stood out.
“Is AI empowering lawyers or replacing them?”
All three speakers were optimistic. Then a senior law firm partner pushed back. He called himself an “AI pessimist” — if AI succeeds, practitioners may be digging their own graves. If it fails, it’s just another capital bubble. He added that US-China competition means neither side is willing to pump the brakes on AI development.
The room went quiet for a moment. He was saying something many lawyers think but rarely say out loud.
The responses came from different angles. One speaker argued that when productivity explodes, the rules governing how gains are distributed — which is to say, the law — only become more important. Another said that even if AI capability froze at today’s level, its impact on the legal industry is already irreversible. A third admitted that society may genuinely not be ready for AI’s success — but that doesn’t change the fact that we have to face it.
“Can China produce its own Harvey?”
The room was doubting it. Harvey is valued at $11 billion. That scale is nearly impossible to replicate in China’s legal market — not just because of market size, but because Chinese corporate legal departments are relatively powerful and prefer to solve problems in-house rather than rely on third-party tools. Several major companies’ legal teams are already deploying AI at a pace that would surprise outsiders. Trying to build a product business by serving corporate legal departments in China is a tough sell.
“How do you deal with AI hallucinations?”
The consensus: cross-verify with multiple AIs. Require citations. Build authoritative source libraries into your skills. Set your preferences to make AI ask questions about uncertain content instead of fabricating answers. The core principle: treat AI like an intern you need to manage — you must be able to judge every output it gives you.
What I took away
I’ve now hosted or spoken at events in Silicon Valley, Tokyo, Beijing, Shanghai, Hong Kong, and Shenzhen. This Shenzhen event reinforced something I keep seeing across all of them:
The lawyers who are pulling ahead aren’t the ones who know the most about AI. They’re the ones who are clearest about what AI should and shouldn’t do in their work.
That’s the real meaning of “AI-native.” Not a new tech stack. A new way of thinking about where your value actually lives — and ruthlessly using AI to free up time for exactly that.
This was our community’s 10th in-person event. Silicon Valley Legal Tech Frontier Community — born in Silicon Valley, now 1,000+ members across the US, Tokyo, Beijing, Shanghai, Hong Kong, and Shenzhen. We help legal professionals move freely in a changing world.
Next stop: “What is an AI-native Law Firm” @ Palo Alto (Stanford Campus), May 23. Learn more at helenlab.com/community.