This spring, something shifted. YC’s W26 batch produced multiple AI-native law firms — Vector Legal, Moritz (formerly Arcline), LegalOS, General Legal. Carta acquired a UK firm and launched Carta Law, calling itself the largest AI-native law firm for private capital. The term “AI-native law firm” went from niche to everywhere in a matter of weeks.
But here’s the thing: nobody agrees on what it means, and there are too many buzzwords around this concept.
That’s why I brought 50 people into a room at Stanford on May 23rd. For every community event, I aim for a precise mix: attorneys who practice, founders who build, and engineers who ship. This time, the room had 23 attorneys, 17 founders, and 10 engineers — from YC startups, Google, Meta, Big Law firms like Skadden and A&O Shearman, and a dozen other places.
I asked every speaker one mandatory question: What is an AI-native law firm, in your definition? Nobody agreed. But after two hours, three words kept surfacing:
Client-Facing AI. Self-Learning Data Loop. Flat Fee.
Word #1: “Client-Facing AI”
Keenan Venuti, CTO and co-founder of Vector Legal (YC W26), gave the most direct answer. To him, AI-native means the client interacts with AI directly.
Vector Legal built a platform called VectorOS. Their startup clients log in, upload their legal documents, and ask questions across them. They can review contracts, manage their data room, track equity — without calling a lawyer every time something comes up. The lawyer is still there, but the role has changed. Instead of being the person you call with every question, the lawyer designs the system, oversees quality, and steps in when something gets escalated.
This is a fundamentally different relationship. In a traditional firm, the lawyer is the interface. In this model, AI is the interface. The lawyer is behind the scenes.
My position takes this one step further: in an AI-native law firm, lawyers serve AI, and AI serves clients. That's not diminishing the lawyer's role. It's redefining it. The lawyer's core value shifts from delivering legal services directly to training, supervising, and optimizing a system that can deliver at scale. The lawyer becomes the architect of the machine, not the machine itself.
Linh Duong, founder of LegalScientist.ai with 20+ years of international legal experience, extended this idea from the other direction — the lawyer’s side. He’s building a platform for antitrust compliance, and one thing he’s seen is that AI gives lawyers a new kind of visibility into their clients. With AI continuously monitoring a client’s risk patterns, a lawyer can now go to the client and say: “I’ve seen this pattern. It means you might have A, B, C, and D. You probably haven’t seen this yet.”
That flips the lawyer-client relationship. Not waiting for the client to show up with a question. Going to them before they even know there’s a problem.
A senior engineer at Google X added a cold dose of realism. He’s working on legal-domain technical research — patent search, hallucination detection, document ingestion at scale. His point: for client-facing AI to actually work, context ingestion has to be solved first. The current connectors — Google Drive, Box, whatever you’re plugging in — are toys. They can’t fully ingest a client’s legal context into an AI system. Without that technical foundation, “client-facing” is just a nice interface sitting on top of a shallow model.
Fair point. The vision is clear. The plumbing isn’t there yet.
Word #2: Self-Learning Data Loop
This is the one I feel most strongly about — and the argument I opened the event with.
In my presentation, I walked the room through my 5-Step AI-Native Law Firm Roadmap and drew a line at Level 3: the self-learning data layer. Below it, you’re using AI tools — a chatbot here, a plugin there. Above it, the system learns from your work. Every output feeds back in. The AI gets smarter each time. Most firms haven’t crossed that line. Most firms don’t even know the line exists.
I shared a live example from my own experiment. For the past 80+ days, I’ve been publicly building OpenClaw Law — an AI-native law firm with two AI agents, Morgan and Cleo. Recently, I migrated my setup to Hermes Agent, which has a built-in self-learning loop. When I asked Morgan the same type of question across several sessions — how to set up a Delaware LLC — the agent naturally generated a reusable skill called “startup client intake brief.” No one programmed it to. The system recognized the pattern and created a shortcut on its own. That’s what a self-learning data loop looks like in practice: the work product becomes the training data.
I wasn’t the only one thinking along these lines. One participant put it simply: “AI should be learning, because that’s what a legal system is. You train the legal system, you train the AI to do the same thing.” He drew a direct parallel between training a junior associate and training an AI agent — you give it tasks, review the output, correct it, and over time it handles more on its own.
Another founder described what this looks like at the system level: their clients’ entire journey — from case intake to messaging, drafting, payments, and post-filing interactions — all flows into one system, building a deep context layer. From a single interface, you can ask an agent to handle any part of the pipeline. The data doesn’t disappear after each engagement. It compounds.
Word #3: Flat Fee
One veteran attorney and founder cut through the theoretical discussion with a blunt claim: a truly AI-native law firm must be flat fee.
The math is simple. If your AI drafts a first-pass memo in 10 minutes that used to take a junior associate 4 hours, you can’t bill by the hour anymore. The pricing unit has to change.
New firms don’t have this problem — they price however they want from day one. The real tension is inside traditional firms trying to adopt AI while keeping their billable hours model intact. That’s the part that’s going to break.
Other Moments That Sparked Discussion
Several threads stood out from the open floor.
How many lawyers does an AI-native firm actually need?
Daniel Aydin, founder of Lovie.co, raised a question the room couldn't shake: if one person armed with AI can own the entire client journey — from intake to documents to operations — how many lawyers does a firm actually need? His concept of the "Manager of One" suggests the answer might be far fewer than today. But he added an anchor: clients still want to feel taken care of by a real human.
When Lawyers Stop Reading Contracts
Jean-Pierre Guittard, a veteran attorney, made an observation that got the room talking: he’s noticed he doesn’t really read contracts anymore. AI reads first. He reads the summary. He acts on what AI flags. The old relationship with the text — clause by clause, word by word — is fading. It’s a real tension, though I’d note that verification has always been the lawyer’s job, and AI hallucination is improving rapidly. The deeper question isn’t whether AI makes mistakes. It’s whether lawyers will keep the muscle memory to catch them.
Access to Justice: The One Thing Nobody Disagreed On
On the other end of the spectrum, access to justice was the one topic where nobody disagreed. Leo Li (JSD candidate at Stanford) shared his research on AI transformation in courts — using AI to audit debt collection cases and finding that vast numbers of parties never received effective legal representation.
Sheikh Sultan Aadil Huque, a research associate at NUS Singapore, is building an AI platform for public international law grounded in UN primary sources, with a free tier for students, refugees, and developing countries. The biggest value proposition of AI-native law firms might not be in making BigLaw more efficient. It might be in reaching the people who could never afford a lawyer in the first place.
My Takeaway
Three words. Three shifts.
Client-facing AI changes the service interface — who interacts with whom.
Self-learning data loop changes how knowledge accumulates — from individual memory to institutional intelligence.
Flat fee changes how value is priced — from time spent to outcome delivered.
Together, they form something like a working definition. Not the final one. Not a universal one. But the one that 50 people in a room at Stanford — attorneys, founders, and engineers — kept circling back to in May 2026. Six months from now, the answer might look completely different. That’s exactly why documenting it matters.
What I’ll carry forward from this conversation: the industry is moving faster than any single framework can capture. But the firms and teams that will lead aren’t the ones adopting the most tools. They’re the ones rethinking the relationship between lawyer, AI, and client from the ground up.
Meanwhile, we’re planning our next conversation for late June or early July in New York City. The topic: What Is an AI-Native Legal Department — shifting the lens from law firms to in-house legal teams. We want more in-house counsels and founders in the room. Whether you’re already building AI workflows or just starting to use ChatGPT to review contracts, every voice is welcome here. No barrier to entry. Just one prerequisite: you care about what the legal industry becomes next.
Thank you to all participants for the conversation. Special thanks to Leo Li and Adrian Mak for their help in making this event happen.
Legal Tech Frontier — a community I started in August 2025 to bridge legal professionals and tech builders, and to empower lawyers to move freely in a changing world. We’ve now hosted 11 events across Silicon Valley, and Asia (Beijing, Hong Kong, and Tokyo), with 1,000+ members across the US and Asia; Join us through helenlab.com/community.