How agentic artificial intelligence could shake up the legal industry

Episode Notes

Transcript

Announcer:

Welcome to theABA Journal Legal Rebels Podcast, where we talk to men and women who are remaking the legal profession, changing the way the law is practiced, and setting standards that will guide us into the future.

Victor Li:

What is a agentic ai? According to IBM, it refers to AI systems that are designed to autonomously make decisions and act with the ability to preserve complex goals with limited supervision. So yeah, that definitely sounds like it could be really exciting or really scary, or maybe both. And for lawyers, AI agents could completely change the way they do their jobs, handling things like legal research, document creation, and managing workflows with little human supervision. But if we learn anything about AI over the last several years, the potential benefits of agent AI can come with risks and potential consequences. My name is Victor Li and I’m assistant managing editor for the A B ABA Journal. My guest today is Tom Martin, he’s founder and CEO of Laro, an automation company that creates chatbots for lawyers and law firms. Tom is also CEO and founder of Deep Legal Consulting and co-founder of the Academy of Legal Innovation. He recently served as an adjunct professor at Suffolk University Law School where he taught generative AI legal services to students. Most importantly, he’s a longtime friend of the program making his second appearance on the show. So congratulations.

Tom Martin:

Thanks, Victor. Happy to be here,

Victor Li:

And he’s here today to talk about agentic AI and what it is and what we can expect from it now and in a not so distant future. So thanks again, Tom. It’s good to talk to you again.

Tom Martin:

Yeah, good to talk to you too. It’s always a pleasure.

Victor Li:

I gave the brief elevator version of your bio. For the listeners that might not know much about you, can you talk a little bit about yourself?

Tom Martin:

Happy to. I do have a background. As a lawyer, I have over 25 years of experience now actually practicing law. I still have the firm and have attorneys that work for me, but I get to focus my time on my baby, which is Laro. Been doing it for nine years. It’s been an evolution where we have started out with rule-based chatbots and layered over different AI elements as they’ve become mature. And that’s where we’re at right now. Everybody knows that it’s just getting crazy with the amount of new innovations coming out almost daily.

Victor Li:

So how do you balance your role as a lawyer and then also your role with Laroy and whatnot? How do you maintain that balance between the technology aspect and then also the practicing aspect?

Tom Martin:

Yeah, so I guess I wouldn’t characterize it as a balance so much as the feedback loop because in a sense, I am my own target market as a practicing lawyer, I naturally have a feel for how things work on the ground and then that informs how one might build a solution to make it better, more efficient, and actually better quality. And so I think that’s a competitive advantage and it really helps and makes it a lot of fun to do.

Victor Li:

And actually, I’ve always been curious about this. I don’t think I’ve asked you this before. So you’ve always been interested in chatbots for a long time, and I remember, I think one of the first stories I wrote for the a b ABA Journal was about a illegal chatbot you designed for Facebook users, and you’ve seen how chatbots have grown and evolved over the years and just the quantum leap that it’s taken over the last several years with generative ai. What made you decide to get into this field in the first place, and did you think that they would become as prevalent as they have?

Tom Martin:

Wow, I only wish that that were true that they would, but wow. And you take me way back, the Facebook chatbot, that was about nine years ago now as an experiment to help people incorporate their businesses in California. No, in short, I didn’t know where it was going. A lot of the talks that I gave back then about AI was pretty much talking about AI as it would be now and where it’s headed. And it was frustrating for both sides for me as someone who wanted to build tools to be more intelligent and do more. And for attorneys, frankly too, at the time, they kind of saw things or wanted expected things to be like magic. And back then it was impossible to meet that expectation. Yet at the same time, what it could do back then was revolutionary. The ability to get even a rule-based bot to actually perform a function and do virtualized work was extremely impressive. The reason I’ve said it a few times, and my initial inspiration was Joshua Browder with his work on do Not Pay, he was as a 17 or 18-year-old in London, helping people get out of parking tickets with his bot.And When I read that story, I was like, man, this kid, he’s onto something and I need to take advantage of this and bring it into the legal world.

Victor Li:

So if you were to look back at one of your speeches from nine years ago about ai, would it be pretty close to what you see now or would you be like, I don’t even want to go there at this point?

Tom Martin:

No, I mean some of it was good. One of ’em that was just completely, almost practically on point was this one. It was a gif, this little video graphic that I used to put up, which actually came from a Mother Jones article that was talking about AI crossing the threshold to general intelligence.

And it used this analogy of, I think it was Lake Michigan and that Lake Michigan holds the comparative volume of how many synapses we have in our brain or neurons. And basically it was showing it getting filled up and it was pointing out the fact that it’s not linear, it fills up right at the end. You drop, drop, drop to fill in Lake Michigan or you start populating the number of neurons that the brain has, and it’s right at the end where it just goes parabolic. And the year that it came up with when that would happen is this year 2025. And by most estimates, having AI with general intelligence is not too far off. It’s probably between two to five years based upon our current progress. And so it’s kind of crazy that it wasn’t too far off. Yeah, that’s crazy.

Victor Li:

So let’s talk about agent tech ai, and I promise I won’t make any secret agent puns or any, I won’t make James Bond references or seeing secret agent manner or anything like that. So what’s the difference between agent AI and generative ai? Is it genetic ai the next step from it, or are they different from each other?

Tom Martin:

So it’s funny because actually agents use generative AI to accomplish what they need to. In fact, generative AI is what empowers agents to take multiple steps to accomplish goals. So they’re not two different things. In fact, they rely on each other. And so how would a lawyer use an AI agent and

Victor Li:

How would it differ from how they use generative ai?

Tom Martin:

Okay, so one thing I tell my students is that it’s important to understand the before and after picture.

And so before picture, before we get into agents and workflows and all that stuff, let’s think about the before picture. Well, it used to be that you would code a rule-based solution where it would go through multiple steps. You would define what you wanted to solve. Let’s say it’s tagging up a document where the old fashioned way, the true before picture is that you would have an associate sit down and actually tag up a document saying that there’s different maybe legal issues that come up in this motion. So the next step is that you would likely have a computer OCR it so that it understands the text, and then you would have it programmatically find keywords and then tag those keywords as representing different legal topics. That’s just a rule-based workflow. So then the next step is that you would have an AI workflow where you basically define the flow of what you want it to do.

You want it to tag different legal topics, but then instead of having it use something like RegX, which can identify different keywords within a document, you would rely on an AI model like let’s say Claude or GPT-4 oh or 4.1. And so it would be the AI model that’s actually understanding that this is a particular legal topic and tagging it as such in the document. So we’ve gone from a traditional workflow to now an AI workflow. So then now we’re crossing over into an agent workflow as you’ve heard many, many times before. And so I should point out there’s two types of agents. There’s ag agentic workflows, and then there’s true AI agents. And so the ag agentic workflows are ones where you actually don’t tell the AI in particular, this is the step-by-step that you’re going to do. What you do is you provide it with a goal.

You say, I want to tag up this document based upon the different legal topics presented in this document. And so you just give it that goal and then it creates a plan of how it’s going to do that. It has access to certain tools, maybe it has a highlighter, a way that it could highlight the text. It has another way where it could actually apply tags. And so it has different tools available to it. As I mentioned before, it’s relying on an AI model. And then it also typically has some kind of knowledge base where it could draw information from maybe examples of other documents that have been tagged in similar ways. And so then it’s created its own plan. It goes through the document, it tags it all up. It has a feedback loop to just make sure, have I done all of the steps that I’m supposed to do?

And then if it has, then it rests, it’s done, and then it presents the report or the outcome to the user. Now, a true AI agent is different in that essentially it’s like a multi turn workflow where it comes back from the feedback loop and then it can create a new plan based upon a new set of circumstances, just like we do if we’re walking down to Grand Central Station, but then we get interrupted by somebody who’s starting to talk to us about something else, then maybe that takes us in a different course. So that different course we plan for, we think through how are we going to deal with this? And in the same way an AI agent would think through that new circumstance and then plan a way to solve that. And so these new AI agents, they actually can continue that loop for up to eight hours, not all of them, but in the philanthropics latest report, it was reporting that they using their own clot agents can go through that multi-term process up to eight hours. So the big advantage, the big benefit to using agents is that unlike what we’ve seen before, you could actually focus on outcomes, not just step-by-step processes. And so that’s a big advantage for lawyers because you can give it assignments where it can work on those and hopefully take the work off your plate.

Victor Li:

Do you think that would make lawyers nervous though, to give up so much control or it just relies so much on the agent getting it right? I mean, do you think that that might be an obstacle to widespread adoption?

Tom Martin:

I agree. I think it is a concern. That’s one of the concerns I have about ai, like true AI agents, because anyone who works in this field will tell you that each inference step each time, it has the ability to plan and make those decisions on its own. There’s the possibility of error. We all know about hallucinations and that sometimes they come up in terms of citations and there needs to be a correction for it. But beyond the error, I would point out there’s a real cost to it, literally because when AI agents are thinking through, they’re planning, they’re doing the multi-step, they’re feedback, and then coming up with a alternative plans, all of those are multiple calls to the AI models. So you have multiple calls to the AI models, each one of those calls and inference steps cost money, and so it could get very expensive very quickly, especially there’s yet another step that I didn’t mention, which is a multi-agent architecture. And the multi-agent architecture is like, one example is Claude’s research agent, where basically it uses this Opus four as kind of the orchestrator that divvies up work to sub-agents, which are Claude Sonnet four and the subagent conduct research by, they break down the task into Subparts, and each subagent does research reports back to Opus four, who’s the orchestrator, and it goes through this entire loop. So you can only imagine how quickly you could burn through money

Doing it that way.

Victor Li:

Well, so now I think about all the agents, what I seen in the matrix where NEO has to take on all the agent Smiths.

Tom Martin:

Yeah, you don’t want to end up being taken down by multiple agents that way. All right, so before we continue, let’s

Victor Li:

Take a quick break for a word from our sponsor and we’re back. So let’s talk a little bit more about, I guess just AI in general, specifically generative AI and energetic ai. So how prevalent is generative AI within the legal profession right now? How many, I mean, I know it’s tough to put a number on it, but is it pretty widespread amongst lawyers and legal professionals?

Tom Martin:

So I can only rely on reports that I’ve seen. For example, the Walters clearer report from last year was saying that up to 86% of in-house counsel, and I think it was 78% of law firms were using a generative AI weekly. And so that one distinguished itself from the clear report in that the Clio report, although it showed a high percentage, I think it was 76 or 78%, they didn’t really specify use weekly or not. It was a whole range of use from using it once a month to using it every day. But Walters clear was much more specific about the frequency of use. And so that was last year. And so I could only imagine the uptake has been greater anecdotally from talking with a lot of people who want demos of laroy and other customers and just people in the community, they’re using it maybe not to its best use to actually help them to accomplish tasks regularly. They are using it for some basic things like summarization and helping to draft with the appropriate diplomatic language, but to actually bring it into an entire workflow. There are larger firms that I know are tackling it really well,

And some of them have built even their own custom AI and even ag agentic workflows, but not everyone I think is using it to its full advantage. And it’s true, the old saying that the future is here, it’s just not evenly distributed. It’s definitely true for law firms.

Victor Li:

Right. Yeah, I mean, I was going to ask you, are you it more in the research phase, like people using it to draft memos or is that still kind of, I mean, did the stuff by hallucination scare ’em off a little bit, you

Tom Martin:

Think? Yeah, I definitely think there’s been a backlash to using it for legal research in a way where you just rely on it and for good reason, you shouldn’t just solely rely on it like that and then take it as the words of the Bible and just file it with the court. That should never have happened. We can’t obviate our fiduciary duty of competency supervision as lawyers. That’s what people rely on us for, but it can be used responsibly. I’m not saying not to use it. There’s really wonderful products and they’ve gotten so much better now. They rely on qualitative source documents. And when that is done, which I want to point out, maybe it’s an obvious point for a lot of people listening, but when you use chat GPT, like the consumer facing version, that is just relying on what it’s been trained on unless it’s searching the web and the web itself is not entirely reliable, but that’s the extent of what it knows. It’s not relying on a vetted golden set of information that’s being provided to it. So that’s why it’s more likely to get things wrong.

It’s more likely to hallucinate. But if you actually, a lot of, I would think most legal AI vendors, they do have a reliable source of data and documents. Of course, the Thomson routers and Lexus of the world have that within their database. When they rely on that information, it is much more likely to come up with a reliable and accurate output, but you still need to review it and you still need to make sure that beyond the review, that it’s accurate, that it actually fits within the argument that you’re trying to make and works well for furthering the interest of your client.

Victor Li:

Yeah, I think someone described it to me as sort of for the people who are like, oh, if it’s on the internet, it must be true. Then chat GBT just kind of takes that to another level because then it just spits out whatever this is on the internet. I mean, I think I asked it about myself and it said that I wrote a book on Kim Jong I, I’m, I think I’d remember that if I did that. I probably also have other types of agents following me, not AI agents. I think I probably have real agents following me if I did that. So yeah, obviously beware. And I feel like for lawyers though, it’s just like they should know better than to just rely on something like that. Just even going back to law school, just having to check your citations and check all your cases and whatnot, they should know better. But it’s been strange to us just in the journalism field, just seeing all the people that just don’t do it.

Tom Martin:

Well, let me point out, one thing that I’ve done to try to tackle the hallucination problem with citations is that through logy, I’ve actually developed a tool called Sitecheck ai and Sitecheck ai, it should eliminate the problem because you simply upload a pleading, you upload a motion, whatever the document is that has citations in it, and it thoroughly reviews it by checking it against a database of cases to ensure that those cases actually exist. And so my point is that these tools can be used to eradicate these problems as well, and that lawyers really don’t have an excuse anymore for having to worry about getting sanctioned or having inaccurate citations in their documents because there are tools available to prevent that.

Victor Li:

Right. So talking about agent to I again, so how prevalent is that within a legal issue? Is that still very much in IT infancy?

Tom Martin:

I believe so, yes, because there’s still a lot of questions about how to responsibly use it, and I think you definitely need to have a level of understanding sophistication to be able to employ it

Victor Li:

Responsibly. Gotcha. And we talked a little bit earlier about what it can do. What are things that it can’t do?

Tom Martin:

Okay, it can’t make a coffee for you, it can’t make ethical decisions. Well, look, it actually can make ethical decisions for you.

Would you want it to? Absolutely not, and this is where an AI strategy plan or an AI procedure and policy for a law firm would be extremely useful because it is early days because we are trying to wrap our heads around what it’s good to be used for, what it’s not. We really need to sit down and ourselves think about what we want people to be able to use it for. And I’m not saying banning it, I’m not saying for people to tell people not to use it, because that’s probably most likely when other people will use it when they’re told not to use it. But we all have to find a responsible way to work and live with AI because it’s here to stay and we need to have rules and principles that guide how we interact with it.

Victor Li:

Gotcha. Alright, now let’s take another quick break for a word from our sponsor and we’re back. So now let’s take a look into the future. So you could put your prognostication hat on and hey, you saw the chatbot thing coming well a decade ago, so I expect you to get these predictions right or everyone’s money gets refunded, right?

Tom Martin:

I don’t know about that.

Victor Li:

So where do you see, when it comes to HN ai, where do you see the technology heading? What do you think we’ll see within the next couple of years or so?

Tom Martin:

Well, I think what’s been bandied about lately is that 2025 is year of agents. I don’t think that’s true for the legal industry yet. We’re wrapping our heads around it just like we did chat, GPT and generative ai, and we’ll learn how to get it right. I do think that once we do it is going to become extremely prevalent in law firms. Lawyers will use it without a second thought. All of this becomes much more interesting when it becomes boring,

When it becomes just another tool. And we’re not there at all yet for AI agents or agentic workflows. The thing I see five years out is that what this technology does is it allows lawyers to be proactive in a way we’ve never been before. Our work as professionals has largely been reactive. Somebody comes with a lawsuit or they want to file a lawsuit. It’s already after the fact that there’s been some kind of damage and a lot of it has to do with that type of reaction. I think the pro action, the ability for lawyers to actually have a much closer relationship with their client will be enabled by Jaron of ai. And so the concerns about the billable hour and have less work to do, keep in mind that’s all within the reactive category, but this proactive category is actually kind of a blue ocean that in a lot of ways is filled by insurance. Where insurance kind of covers that uncertainty that has to do with what can happen before you have a legal problem.

But where lawyers can actually take up some of that space is by using these tools to actually work in real time with their clients. What do I mean by that? I mean that one thing that’s become clear is that these models can actually conduct some pretty decent legal analysis and real time, practically real time. I remember that I fed in a Supreme Court decision and then asked for an analysis of it and got one that was pretty decent from, I think it was son at 3.7 at the time. Within about 10 seconds I reviewed it myself, thought through the reasoning, compared it to the actual opinion, and it was pretty on point. Now with Ford, it’s gotten even better with these reasoning models even better from OpenAI. My point being is that there’s a lot of activity happening with clients on a daily basis, many of which has legal implications.

Now an innovative entrepreneurial firm could, and some are already thinking this way, build into a client’s daily activities, legal analysis, red flags, check-ins of everything that’s happening, and have these risk loops informing decisions of the company as well as the firm on an ongoing basis. And so that ability to deeply understand a client’s business and be involved in it and providing the value that they want, which is avoiding risk, while at the same time giving lawyers an even greater field of business opportunity, I think is a big part of the future for lawyers. And I really hope they embrace it because that’s actually a way to expand what we do and even be more valuable to our clients than the kind of future that we’re worried about and fearful of right now is that we’re going to be replaced by AI because we’ll be much more in the position of being strategic and helping them in deeper ways. And likely all of that will lead to even more jobs for more lawyers because there’ll be many more support roles that are necessary to coordinate it.

Victor Li:

Yeah,

Add, that was my next question. I was going to ask you, is this going to replace lawyers? That’s what everyone’s always afraid of. That’s always the big million dollar question. Is this going to replace lawyers or even get rid of certain types of lawyers or maybe of a certain level of seniority or whatnot? If we were to place the first years or the summer associates or things like that, or people on the lower end of the seniority spectrum, I mean, do you think there will be some kind of effect like that where, I mean, if you can do the work of a one L or of a young associate, then why would law firms have to hire so many? They could just hire a small few and then rely on the AI to carry the rest?

Tom Martin:

So it’s not a straightforward answer. It depends answer, and somebody that I really admire, professor Richard Suskin has been writing about this for decades and he’s anticipated what’s happening now, which is that I agree that lawyers as they’re currently defined, will become less and less necessary, but that’s not the end of it. What’s going to happen is that lawyers, as we know them, are going to have a redefined skillset, like the job description that they have now is going away, but they’re not going away. The job description is changing.

And also there’s going to be explosion of legal support roles that kind of explode the traditional artisanal lawyer into many different roles that work in tandem with ai. So like legal ai, data scientists, legal engineers and technicians as Suskin put it, and many other roles that we haven’t foreseen. One analogy that I’ll give you that really drives it home is drones. When drones came onto the scene, people were really worried that the traditional artisanal pilot would go away entirely, and you wouldn’t need pilots anymore, and it would just be drones doing everything. And actually what’s happened is that the military employs a lot of drones successfully. And actually what happens, a drone, A drone doesn’t just fly itself. It actually requires a support crew, and you have many teams of soldiers that are assigned to operating drones. There’s lots of information that needs to be reviewed. There’s lots of support roles that need to provide the resources and supplies necessary to run and make drones fly. So it’s not the case that it’s eliminated jobs. If anything, it’s multiplied those jobs and actually human pilots are still quite in demand.

Victor Li:

Yeah, I mean, given the reluctance of people to adopt self-driving cars, I feel like pilots will probably be safe for a while because people aren’t just going to be willing to trust a pilotless aircraft, at least getting on one for a passenger aircraft. I mean, it just doesn’t seem like they’ll be able to make that leap of faith for many, many years until long after the technology is proven to be safe, or at least the risk is minimized compared to a human pilot. I mean, that would be a huge leap, I think, right?

Tom Martin:

Yeah. In the same way for human lawyers,

For a client with a bet, the farm case or even one that’s not bet the farm, but it’s just their livelihood. They’re going to want to have a human in the loop. And so having us as humans in the loop who are ensuring responsible approaches, ethical approaches, strategic ones, and keep in mind, it’s kind of a crude way to put it, but at the end of the day, we carry malpractice insurance. We are reliable because not only do we give our word that we are, but we back it up monetarily. And I don’t know of any AI agent or model that’s backing up what they do with indemnity at this point.

Victor Li:

And so there’s one more thing I wanted to ask you just about access to justice. I mean, obviously with the LSC budget possibly in trouble, I guess that’s a nice way to put it, the A two J gap, if anything might become even greater in the not so distant future. So how can AI agents help people who can’t afford a lawyer or need access to legal services and can’t do it?

Tom Martin:

I think they absolutely can help. I’ve been very lucky to have been working for the past year and a half, two years with Legal Aid of North Carolina and many other legal aid organizations. In fact, log, our closest partners have been legal aid organizations, state courts, and even state bar associations in developing solutions for them that help them to provide legal information to the masses. And most people can’t afford lawyers. And these types of tools allow organizations to amplify and multiply the number of people they can reach because it’s just not physically possible. Even if we try to throw human lawyers at the problem, there just aren’t enough lawyers to do it in the states. I believe it’s close to 1.5 million lawyers that are out there. And even if you had every single one provide pro bono hours, it just wouldn’t be able to amount to the need that we have. And so you need to have technology like this working 24 7, 365 to help people when they need it, for the needs that they have.

Victor Li:

And to wrap up, if our listeners want to get in touch with you either having questions about agents or they want to talk to you about automation for the firm with bots and whatnot, what’s the best way for them to do that?

Tom Martin:

Well, thanks Victor. The best way for them to reach out to me, I’m willing to give them my personal email address. It’s [email protected]. I think the way to keep on top of the latest developments and my thoughts about AI in the law is my substack. It’s called la manifesto.com, laro manifesto.com, and yeah, I write there every single week, have some podcasts myself, and yeah, I really enjoy our community. We have the American Legal Technology Awards coming up in October, and that’s always a great opportunity to get together and just recognize people that are doing good in our community.

Victor Li:

Yeah, I went to that event last year. It was a lot of fun. You guys always put on a good show, so thanks for that.

Tom Martin:

Yeah, thank you. It’s great to hear.

Victor Li:

Yeah. Thanks again for joining us today, Tom, it was great to talk to you again. It’s always nice to hear from you.

Tom Martin:

Thank you, Victor.

Victor Li:

Pleasure. If you enjoyed this podcast and would like to hear more, please go to your favorite app and check out some other titles from Legal Talk Network. In the meantime, I’m Victor Li, and I’ll see you next time when you be ABA Journal Legal Rebels podcast.

Announcer:

If you’d like more information about today’s show, please visit legal rebels.com. Legal talk network.com, subscribe via iTunes and RSS, find both the a b ABA Journal and Legal Talk Network on Twitter, Facebook, and LinkedIn, or download the free apps from a b ABA Journal and Legal Talk Network in Google Play and iTunes. The views expressed by the participants of this program are their own and do not represent the views of nor are they endorsed by Legal Talk Network is officers, directors, employees, agents, representatives, shareholders, and subsidiaries. None of the content should be considered legal advice. As always, consult a.

Continue Reading