AI at dinner: what's next for law firms
Vasu Aggarwal
On October 13, we co-hosted a dinner with Stubbs Alderton & Markiles at Eataly in Los Angeles. We invited legal leaders from southern California to talk about AI: what's working, what's hard, what's next. I opened with a fireside chat between Kevin Debre and Bob Green.
What I enjoyed most was how quickly the conversation got practical. Nobody at the table was asking whether AI works. They wanted to know where it fits, who checks it, and what it changes about how they run their firms and serve their clients. Those are harder questions, and they're the ones I've kept thinking about since.
Here's what stuck with me.
Where AI sits in the stack
We also walked through how AI is built, in layers. Hardware comes first: the chips and servers. The algorithm sits on top: the models from companies like OpenAI and Meta. Then the application layer, which is where users get the value. It is built from two things: the harness, which pulls the model and the data together to produce a response, and the data itself.
I find this breakdown useful because it tells a firm where its real choices are. Very few firms will actually train their own models. But every firm decides which application it trusts, and what data it puts into it. That's where the differences between tools show up, and it's where a firm's own knowledge and judgment matter most.
It's also where the next shift is happening.
Agents change the risk
We also spoke about AI helping people read: cases, contracts, large sets of records. And now, agents are writing a different story. They send the email, reserve the table or purchase online on your behalf. A draft can be checked before it goes out. A message sent to a client has already left. So firms need to decide which actions AI can take on its own, which need approval, and who owns the fix when it gets one wrong. And for responsible leaders, what’s insured.
That last question really stayed with me. When people start asking about insurance, it tells you they're no longer treating AI as an experiment. They're treating it as part of how the firm operates, with all the accountability that comes with it.
"Closed system" doesn't tell you much
If agents raise the question of what AI should be allowed to do, privacy raises the question of what it should be allowed to see.
It came up again and again. "Closed system" makes it sound safe, but it doesn't say where the data is processed, who can see it, whether it's kept, or whether it trains a model. A familiar vendor name doesn't either.
Start with what you're putting in. Public research is one thing. Confidential client documents are another. Attorney-client privilege is a different matter entirely.
To me, this is a more honest starting point than any label. The question isn't whether a tool sounds secure. It's whether a firm can clearly explain what happens to each kind of information it puts in.
Today's junior attorneys are tomorrow's leaders
Albeit briefly, we also discussed how private equity is now getting involved in law firm management through an MSO structure with the view of improving margins, or the new ‘partners-only’ law firm which claims to execute majority work through AI. We discussed how on one hand, junior lawyers are important and becoming tomorrow’s future; and on the other, a user of a CPA firm which was now acquired by private equity was providing better services compared to pre-acquisition.
We only touched on this briefly, but private equity is now getting involved in law firm management through MSO structures, with a view to improving margins. There are also new "partners-only" law firms that claim to execute the majority of their work through AI.
The table didn't land on one answer, on one hand, junior lawyers are important. I mean they're how a firm builds its future leaders. On the other hand, one guest shared that a CPA firm they use, since acquired by private equity, was providing better service than it did before the acquisition.
I don't think those two points cancel each other out. New ownership models and AI may well change how work gets done. But firms still have to answer how the next generation of lawyers will build the judgment that clients rely on.
Experience is what clients feel
AI can help a firm prepare and respond faster. It can't make an unsupported promise safe. Client experience is the most important part. The test is whether the client gets a better-informed answer and can trust the lawyer giving it.
The more I think about it, the more this feels like a test for everything else we discussed. Agents, privacy controls, new ownership structures, pricing models: none of them matter to a client unless the answer is better and the person giving it is someone they trust.
This now brings us back to clients.
Better work, not just cheaper work
There was no uniform push from clients to cut fees because a firm uses AI.
People talked about better work as part of the value. A faster first pass gives a lawyer time to test the argument or chase down an inconsistency. Fixed fees came up for repeatable work. Nothing at the dinner made the case that the billable hour is dead, or that one pricing model fits every matter. The hard part is telling the client what they get and why it's worth paying for.
It suggests clients care less about how many hours went into the work and more about what the work did for them.
Looking back on the evening, every conversation kept circling the same idea. AI can do more of the work, and increasingly it will act on a lawyer's behalf. But the questions that matter most are still human ones: who checks the output, what information is safe to share, how the next generation learns, and whether the client walks away better served.
The firms that answer those questions clearly will get far more out of AI than the ones that simply adopt it. That's the conversation I hope we keep having.
Thank you to the Stubbs Alderton & Markiles team and everyone who joined us
