--- id: "workshop:ai-legal-products-showcase" type: "InnovateUS workshop" title: "AI Legal Products Showcase" url: "https://innovate-us.org/ai-legal-products-showcase" date: "2026-05-27" duration: "60 minutes" series: "AI for Public Sector Legal Professionals" instructors: ["Jennifer Sellitti","Sean Harrington"] moderators: ["Beth Simone Noveck"] recording: "https://www.youtube.com/watch?v=cq3A73gJirQ" transcript: "automatic, from the recording" source: "InnovateUS (innovate-us.org)" licence: "CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org" --- # AI Legal Products Showcase InnovateUS workshop · 27 May 2026 · 60 minutes **This session showcases innovative applications of AI in legal practice. Participants will learn how AI tools are being deployed to streamline legal processes, provide legal information to underserved communities, and identify individuals who qualify for legal assistance. Through real-world case studies of AI deployed in legal practice, attendees will explore both the tremendous potential and necessary guardrails of using AI in the practice of law, with practical lessons for implementing similar programs in their jurisdictions.** ## Learning goals 1. Understand how AI technologies are being successfully deployed to improve access to justice, with specific examples including automated document preparation, legal information delivery, and service matching systems 2. Evaluate the opportunities and limitations of AI in serving self-represented litigants and underserved communities, including considerations around accuracy and the irreplaceable role of human attorneys ## Speakers - **Jennifer Sellitti**, Public Defender, State of New Jersey - **Sean Harrington**, Director, AI and Legal Technology Lab, Sandra Day O'Connor College of Law, Arizona State University ## Moderators - **Beth Simone Noveck**, Founder of InnovateUS and Director, Burnes Center for Change and the GovLab Part of the series [AI for Public Sector Legal Professionals](https://innovate-us.org/workshop-series/ai-for-public-legal-professionals-series-). Recording: https://www.youtube.com/watch?v=cq3A73gJirQ ## Transcript Automatic transcript of the recording: names and terms can be misheard, and the times can be a few seconds off. Each time links to that moment of the video. The transcript alone: https://innovate-us.org/ai-legal-products-showcase.txt (plain text) and https://innovate-us.org/ai-legal-products-showcase.json (every segment with its start and end second). **Jennifer Sellitti** [0:00](https://youtu.be/cq3A73gJirQ?t=0) And with gratitude for your help and to everybody for joining us today, as you can see, I am joining you from an airport. So I am going to be brief in my introduction, which just gives me a chance to turn it over to our fabulous speakers. I'll take the poll down off the screen if you've seen it, just making sure everybody's gotten a little bit of background around who's here and why people are here. Again, we heard only from a few people, but we're hoping to hear from you more because we will save a good chunk of time for questions. Again, instructions in the chat, but you can find everything at the bottom of your screen. And if you have any technical questions, please feel free just to send a message to Agata, to Christine, or to myself labeled with Innovate US. Also, feel free to rename yourself if you want to just let us know what office you're from, what state you're from, et cetera. Again, it's always helpful to know who's in the room. What we're going to do is hear from each of our speakers briefly, going to give a little bit of an intro presentation, then I'm going to ask them some questions, and then we're going to open it up to your questions. Again, this is the last in our current series on AI for public legal professionals. We've had three sessions before this. I see some of our past speakers. Thank you for joining us today. And we will also share the link to videos from the earlier part of the series. But today, we shift into a different mode from talking, as we did earlier, about AI basics and AI ethics, to now talking about building AI. [1:31](https://youtu.be/cq3A73gJirQ?t=91) That may seem like a very far stretch or a far leap for a lot of people here who may be thinking, I'm not very technical. But in fact, what's so exciting about these tools is that they can be driven by anybody with plain language. And so what we want to hear from both Jennifer and Sean is more about how they have gone about both themselves and working with partners, building AI, again, in one case, for and with partners, and in Jennifer's case, for her office with partners from outside. So this is a chance to kind of ask your questions about should you build, should you buy, et cetera. So let me introduce each of them, and then I'll turn it over to them to give a little bit of a presentation, and then open it up to your questions. And again, don't hesitate to ask if you have questions about how to ask questions. Jennifer Salidi is really a polymath. We will put her bio in the chat. But let me say briefly, she is public defender for the state of New Jersey. I have had the pleasure to work with her in my capacity previously in the government of New Jersey. She leads one of the nation's largest, if not the largest, office of public defenders in the country, with approximately 1,300 lawyers, investigators, social workers, professional staff serving all 21 counties in New Jersey. But I would be remiss if I didn't give you one other item from her bio, which is that outside of her legal work, she knows what I'm going to say. She is also a certified diver and a licensed ship captain and runs, I believe, Jennifer, I know you don't tend to promote this other piece of things, but give us the links for some of the work you do exploring wrecks off the coast of New Jersey. If you want to know more about that piece, you can read her book, The Adriatic Affair. We will confine ourselves to talking today, though, about legal AI, but please feel free to use the chat if you would like to know anything about exploring dives and wrecks and other super cool stuff. I feel like we'll come up with an Innovate US workshop where we have to talk about that instead. Jennifer will be followed by Sean [3:36](https://youtu.be/cq3A73gJirQ?t=216) Harrington. He's director of the AI and Legal Tech Studio at ASU, Sander Day O'Connor College of Law. He is also a 2025 Fast Case 50 honoree and a recipient of the Nag Exemplar Service Award. He is a leader in terms of the integration of AI and AI tooling into legal practice. So I'm delighted to have both of them. I'm going to turn it over to them one at a time. We'll spotlight Jennifer to tell us a little more about the work that she is doing building tools for the Public Defender's Office. Take it away. Thank you so much, and thank you, Beth, for that introduction. And it is funny that as I've been exploring AI tools that we can use as public defenders in the state of New Jersey, that I've also found a lot of shipwreck applications that I've been experimenting with as well. So this has inspired me on both fronts. But here today, we are here to talk about the Public Defender's Office and how we are using AI in the state of New Jersey and at the Public Defender's Office for building and improving the lives of the attorneys and the people that we serve. [4:55](https://youtu.be/cq3A73gJirQ?t=295) So we at the Public Defender's Office, in a little bit of an interesting place in the state of New Jersey, in that New Jersey as a state has really invested in putting AI at the forefront of government efficiency and looking at ways to make government more efficient using AI. And that, I have to admit, did give us a leg up because it enabled us to access tools that were already being provided by the state of New Jersey as we started thinking about how we can empower public defenders using AI. And one of those things, one of those very important things is the Office of Innovation, as it was then called, it's now called the Innovation Authority. But Beth, at the time, was at the helm of the Office of Innovation, which was created by Governor Murphy in order to kind of short circuit some of the red tape that happens when governments invest in technology and really put us on a path to adopting AI tools and AI products quickly. [5:59](https://youtu.be/cq3A73gJirQ?t=359) And when I first took office on February of 2024, I went to set up a meeting with Beth to talk about how we could use the services that that office was providing and was immediately struck by a comment that she made where she said a lot of government agencies are afraid of it. A lot of government agencies are trying to think about how they can protect themselves from AI. And she became really excited and I became really excited about the fact that what I wanted to talk about is how we could improve our attorneys' lives and the lives of our clients. And that really started with trying to figure out ways to use AI to give our attorneys more time to spend with the client, right? Anybody here is a public defender or a prosecutor is constantly saying, I need more lawyers. I need more lawyers. And at some point, we're not going to get more lawyers. At some point, we have to say, how can we make the lawyers that we have more efficient, enable them to handle more cases in an increasingly complex world? So one of the things that I found that was taking most of my lawyers' time was writing briefs, right? And we all know the struggle of sitting down, we have a topic, and we want to use bits and pieces of other people's briefs. And the way that used to work in the office of the public defender is each of our 21 counties had their own little file on the server where people would put briefs that they have done and put them into folders that people could look through. And you wondered, are those briefs written at three o'clock in the morning? Or were they written at three o'clock in the afternoon? Are they good briefs? Is it a really smart lawyer writing this brief? Or was it somebody who maybe was phoning it in that day, right? So it was really difficult to access the quality. And then also somebody in one of our, you know, I'll just name some counties in New Jersey, somebody in our Essex County office wouldn't have the benefit of getting any content of those briefs from our Camden County office. [7:55](https://youtu.be/cq3A73gJirQ?t=475) So what I wanted to do was create a way that we could use AI and put in all of the best briefs, edited by our appellate section, vetted by our appellate section, so we know that the law is good, that they're well written, put them into a database and allow every lawyer in the agency to do what I call Franken-briefing, right? Take a little part of this brief, a little part of this brief, a little part of that brief, and put it together the way we always do. But that process used to take hours. So now with our AI briefing tool, what our lawyers are able to do is type in a legal question and then pull from all the best briefs written from around the state and then see immediately where that appeared, where that content appeared in another brief and be able to very quickly put it together. And I know when I was a line attorney that the process of just cobbling together my first draft would often take hours and now that's something that the tool can do in seconds. So we're still, there's an iterative process and we're still working through it and one of the things we did with the Office of Innovation and with Beth was she set us up with Princeton University. So Princeton University students were actually the people who built this thing for us and they get a benefit because they get to experiment and understand how AI for the legal profession works. They get to use it as coursework, there's some papers being written about it, some scholarly work being written about it that they have kindly sent me to review but I don't even understand because it's all technical language. But so everybody is a part of this process is sort of getting something out of it and I think one thing if you're looking, if you're in a state whether you have a lot of resources like we do in New Jersey or you don't, universities are a really great place to go to maybe collaborate on trying to build some of your own AI products. So that is a product that we have built and I know that Beth may have some more specific questions that she wants me to answer about that later. We have also taken advantage of New Jersey's OIT where we have [10:01](https://youtu.be/cq3A73gJirQ?t=601) our own closed AI sort of chat GPT-like system where we can ask legal questions to that and we don't have to worry about confidentiality because it's closed within our department so we can feel more comfortable putting client data in there, asking specific questions. It is not connected to Westlaw, I saw somebody just pop that into the chat. It is not, this is proprietary for the state of New Jersey, something state of New Jersey has built and given to all government agencies to use. And then we also have purchased a Westlaw AI so we have that tool as well and we're starting to use that internally. And you know I think when you think through this buy versus build scenario and I think a lot of people out there may be thinking about buy versus build, to us it made sense to build because we had the tools to do it. Westlaw AI is great and we bought it and we are using that, but by building we were able to really customize a tool that met a specific need and is working on a specific set of briefs as opposed to Westlaw AI which is much more of the universe of legal resources that are out there. And then also you know one thing I think one downside and one thing that we have to think about when we're building is when you are building thinking about how is the product going to be used after that, right? Because we have been approached because of this sort of connection and this reputation that we've gotten from being out on the edge on AI, we've gotten approached by a lot of small businesses that want to volunteer to build things with us. And my first question is always like is this going to be sold to law enforcement, right? Is this going to be sold to police departments? [11:43](https://youtu.be/cq3A73gJirQ?t=703) Are you going to experiment with us? And it's tricky because I think eventually all of this stuff is going to be out there but we've sort of made a commitment not to build things that eventually will be passed along to law enforcement or prosecutor's offices. So that's something we are being careful with when we're building out AI. And I think the other things that are important to talk about when we talk about what we're doing is what we're doing now, where we're going next, and sort of what our vision for the future is. And the way that I have always seen AI is that first you have to work on using it internally, right? So you have to work internally on what are the systems that you need? What is the pain point? What is the need, right? And we can all think of a million reasons to do AI but what is a thing that is taking up a lot of time? What is a niche problem that you can solve? Which is what we did with the brief bank. And then we've trained all of our lawyers to use every available AI resource that they can and we are constantly teaching them how to use it to do their jobs better. But to me that's only the first step in the evolution of where I think our industry, for lack of a better term, is going to be over the next several years. The next thing is to me is scaling. How do you then scale AI across your organization and how is it going to change the way people are working? [13:13](https://youtu.be/cq3A73gJirQ?t=793) One of the other things that we have done with AI is eliminate the need for secretaries to open case files. We've automated the case opening process into our case management system using some AI tools, some more traditional IT tools. That could put people who have secretarial jobs out of a job. So what we noticed was that within our secretaria pool we had some secretaries that were very, very good at using technology. Very good at AI, very adept. We're working with the lawyers and helping them use technology that was out there, use our systems that are out there. So we actually created a new position within the agency, advertised it to the administrative staff to allow the secretaries to move from the secretarial track. I'm in state government so everything is civil service, there's tracks that you get into, promotions and salaries are based on the track that you're in. So what we did was open up positions not within the secretarial track but within our MIS department, so our OIT department, and allowed secretaries that were good with technology to apply for those positions. And they didn't change their work assignment, they still sit in the counties that they used to sit in, but now their job is to sort of be an administrative liaison between the office of our technology department and the lawyers working in their office. And by doing that we were able to give them better jobs, higher salaries, a better professional track, but also keep them doing what they were doing, which is teaching the lawyers in the office how to use the technology tools that are available to them. So that's just one way. I also see a world in which we have lawyers, just like if everybody remembers like 10, 15 years ago when forensic science was becoming... [15:00](https://youtu.be/cq3A73gJirQ?t=900) Very big in our cases and much more complicated than individual attorneys could understand or process in a big caseload. I do see a future where we're going to have AI attorneys, that we're gonna have attorneys whose specific function is to help lawyers navigate AI in the courtroom both offensively and defensively. So we're looking at all that. We are looking at, we don't see AI as a tool to be used. We see it as a way that we need to fundamentally restructure our workforce to harness and to make everybody better at their job. And then the last thing I'll say just very quickly is the future. And to me, the future is how do we use AI to improve the actual client communication? And we're not 100% there yet because we're still working on the other things, but we are looking at a lot of ways that we can allow attorneys to engage with the people that we serve more frequently, more often, and more efficiently by using AI tools that we can even put in the hands of both our clients and our attorneys. [16:04](https://youtu.be/cq3A73gJirQ?t=964) So that's the future. And that first future is a little bit less clear, but it's something that we're working towards. And we're always keeping our eye on all those things, internal use, scale, and then what does the future look like and the future is really empowering our clients. And I'll just say two more things before I pass it back. The first is that I think we cannot be afraid of this as lawyers. It is coming, it is here. We have to embrace it and we need to be thinking about it as a major, major change in the way attorneys, investigators, support staff do their job in doing what we do. And that the purpose of this is never to use technology for technology's sake, that the purpose of this is to figure out how to make us more human, right? To use AI to give us more time, to give us back the time that we crave as lawyers and that's time with the people that we serve. So that's always my question. [17:04](https://youtu.be/cq3A73gJirQ?t=1024) Will this get us closer to spending more time with our clients? And if the answer is yes, then that's something maybe we need to look at AI solutions for. The other thing I will say is you may be looking and thinking and say, oh, well, you have all these resources, you have all these things. At the time we were building these, we were also the last agency in the state of New Jersey that did not have wifi in every building. So I guess as a word of encouragement, I say you can build the foundation and build the house at the same time, you just have to be really creative about it. And I know that there's a lot more questions and I wanna hear your questions and I know Beth has some questions for us. So I'll stop here and pass it back to Beth or to Sean to kind of move on to the next piece of this. Great, we'll go right over to Sean. **Sean Harrington** [17:49](https://youtu.be/cq3A73gJirQ?t=1069) Hi, I'm Sean, I'm the director of the AI and Legal Tech Studio. I think I recognize some names in there from all the NAG trainings I've been doing this year. Okay, are my slides up? **Agueda Quiroga** [18:01](https://youtu.be/cq3A73gJirQ?t=1081) Yes. **Sean Harrington** [18:02](https://youtu.be/cq3A73gJirQ?t=1082) Awesome, thank you. Okay, so that's who I am. So I wanted to talk more kind of how you could use these actual tools in the practice of block, kind of get into the nitty gritty, roll up our hands, think of some ways that we can actually practically use these. This is obviously dependent on exactly what Jennifer was just talking about, which is what tool do you have? Do you have a frontier model like Cloud? Do you have open AI? Are you building your own? We're gonna walk through all those options and talk about the power of these things. So Co-Work is really powerful. And one of the reasons it's really powerful is not because it's like this unique feature of Cloud, but because it brings the technical aspect of Cloud code into an interface that most attorneys can use and feel comfortable with without worrying about coding and folders and all that stuff, development environments, none of that. It's really useful. So this is kind of what it looks like in a video. This is him saying, hey, can you organize my desktop for me? This is how my wife's desktop looks on her laptop and it drives me berserk every time I look at it. [19:04](https://youtu.be/cq3A73gJirQ?t=1144) But now you can just tell Cloud to do that. And you can see it's kind of grinding through, thinking about what it's gonna do. And then all of a sudden it organizes the guy's desktop for him. And this has been kind of, so on that in the top right, you see that they're hit list. Cloud has just been rolling out, and Thropic in general, has been rolling out huge numbers of features kind of since March, where they're just every week, they seem to have two or three new amazing features for laptop class professionals. And a lot of those would be really good for attorneys. The other thing is there are attorneys out there now using Cloud Code. It sounds technical or difficult to use. It's not, it's all completely natural language. It'll walk you through things you don't understand. You can ask questions. I would say if you haven't fired up either OpenAI's Codex or Cloud Code and tried to do something, this would be a great time to do it. They're both in a great place and easy to use and it's just 20 bucks. Hopefully your org has access to it. [20:07](https://youtu.be/cq3A73gJirQ?t=1207) So Anthropic made big news a couple of months ago when they launched what's called Skills and Plugins for Law. So I wanted to talk about what those things are. The hierarchy here is that a plugin has a number of skills. And if you can see that list on the left-hand side there, it has different kind of areas of law. So like the corporate legal law, for example, will have maybe a skill that does like redlining or something like that. A law student, maybe it would make like flashcards. The skills are completely natural language editable. So you can say, and you can make them yourself. This is, you can go into Cloud right now and you can say, hey, I would like to make a skill that takes all this discovery and makes a timeline out of it. And it will walk you through the process. It'll ask you some questions. And at the end of that, you have a skill. So what is a skill? A skill is something that the model uses when it thinks is appropriate. That's crazy. That's a whole new thing. So previously when we would do things like connect apps to other apps or other databases, it would be these very narrow API calls where you'd say, you may go get this one thing. [21:11](https://youtu.be/cq3A73gJirQ?t=1271) You may go post this one thing, these very narrow things. And now what we're doing, especially in the instance of skills for Anthropic, is they just go, here's a pile of skills. Use them when you think it is appropriate. And the models are pretty good at that. You can also invoke them manually if you want. You just hit a little slash and then you do a red line scale or whatever it is. And it will do it manually. But in general, it's pretty good at just figuring out what tool to use at the right time. So I was trying to think, this is Cloud Co-Work, not Cloud Code. I was trying to think of some things that you could do with Cloud Co-Work that might be kind of fun for you to do. So first of all, Cloud Co-Work, you leverage on a file folder, a directory in your actual computer. So you could point it at a folder that has your brief bank for motions in Lemonade. And you can say, here are the facts of my new case. I would like you to draft a new motion in Lemonade based on the templates that are in this folder. And it does a really good job most times. [22:13](https://youtu.be/cq3A73gJirQ?t=1333) And it will actually create the thing and you can watch it go through, use any available skills, and then it will deposit it into the folder labeled properly. And it's again, really incredible. You can also schedule these types of automations. So you can say, once a week, I want you to go into the discovery folder. I want you to look at all the documents and I want you to make a timeline of events or update the existing timeline of events. And it will just do that automatically once a week for you. Really incredible stuff. We already kind of mentioned skills, so I won't go over that more. But I would also like to point out that this was the thing that crashed the legal tech stock market. It wiped out $250 billion of investment on a single day from places like Thomson Reuters and Lexis. And I think what it made a lot of people realize is that a lot of the existing software that law firms and lawyers in government subscribe to is just software as a service. [23:14](https://youtu.be/cq3A73gJirQ?t=1394) And to the extent that these tools can replicate that, there is very little moat for a lot of these legal tech companies if you're not holding one of the proprietary data sets like Lexis or Westlaw or VLex. And then they took it a step further and actually dipped their toe into law. And that was another very surprising thing. So in Claude, you now have integrations with NetDocs, iManage, Harvey, Westlaw, Lexis. You could do legal research sitting in your Claude account. This was all very surprising to a lot of us that Lexis or Westlaw would be willing to do that. And yet here we are, where they have definitely taken the lead in kind of the legal market. The other reasons, so I wanna go over some of the other tools in case you have something like Gemini. Google Gemini right now has a 2 million token context window. So a million tokens is roughly the equivalent to two to 3,000 pages of text. So something that's 2 million, that's up to potentially 4,000 pages. So things like document review, discovery, you know, you get this massive discovery bomb. [24:18](https://youtu.be/cq3A73gJirQ?t=1458) Well, now you can use a model that can keep all of those 2,000, 3,000 pages of text right into its working memory, which is pretty powerful. If you have to do anything, visual editing, Google's Nano Banana 2 is truly incredible. I just gave it a picture of me, a picture of, is that Tatum Channing? Is that the guy's name? In one of the X-Men movies and said, put me in the X-Men outfit or whatever. And it just did it. You can remove things, you can make graphics. So all of these, these are all made, they're not actually slides, they're made in Nano Banana, all of the slides on the slide deck, or most of them. And so it's just a really, really powerful tool, really easy to use. Some of the image generators can get kind of complicated, whereas this one's like pretty intuitive and easy. They also, if you're, you know, part of the enterprise Google system, you're, you know, a Google shop, G Suite shop, instead of a Microsoft tenant shop, they have Google Agent Space. Google Agent Space is truly phenomenal. It's exactly what they promised us in 2022 and 2023. [25:20](https://youtu.be/cq3A73gJirQ?t=1520) And now they're making good on it, which is you're in one interface. You can say, go look in my email for the latest email from Jennifer, find any information online that would be relevant to her request, double check my NetDocs account for any relevant query, you know, any relevant documents, attach them and send it in the email. And it will go do all of those things without you having to go into any of those apps, anything that plugs in to Google Agent Space. So truly a very powerful platform. So the other thing that Google has launched is Google Notebook LN. And this is free. If you have a free Google account, you can use it. Just be aware that anytime you're using a free account, the data is theirs, especially in the case of Meta and Google. But it's really nice because the law students love this and the law faculty love this. So you can see on the left-hand side, those are all the sources that I put into, you know, the folder. And then in the center there, you have a live dynamic chat with the thing. Every one of those little footnotes that you can see, you can click and it will take you actually to the source. [26:21](https://youtu.be/cq3A73gJirQ?t=1581) So you can verify the context or the content. And so I think it's a very reassuring thing for people to be able to, you know, where did you get that information? Are you representing this properly? You know, just check the work of the AI. It gives you grounded legal answers, which is really nice. Then they have a whole suite of things where it'll turn into a podcast or into flashcards. MindMap, I saw a transactional attorney throw a giant M&A contract into there and use the MindMap feature so you could have a visual representation of it. So it does have applications for working lawyers, but just a great tool as part of the G Suite. Let's take a brief interlude because what I want to talk about is the white-labeled chat GPT problem. And so all those really cool features that I just showed you from the Frontier models, those are not the brain of the model. That's not GPT-5, GPT-5.5, OPUS 4.7. Those are app layer things, agentic harnesses that are on top of that brain. Well, what a lot of orgs and institutions and even big law have done is they have taken the brain of the thing and mistaken that for actual access to a Frontier model. [27:30](https://youtu.be/cq3A73gJirQ?t=1650) So yes, they have chat GPT, you know, five point whatever, but it looks like this. It doesn't have any of that fancy app layer stuff that would be really, really useful to attorneys. And so they're kind of perpetually stuck living in 2024 without having all of the cool stuff of 2026 that goes on top of these things. The other problem I think in law in general was that a lot of us, because we're stuck in Microsoft Office tenants, which there's nothing wrong with that. I love Microsoft Office. I'm good at, I used to, prior to this, I used to do trainings on tech competency and I've trained like attorneys how to use Outlook and things like that. But we know that most of us are stuck in these things and we all thought that Copilot was gonna be the absolute silver bullet with this technology because it sits in everything. So even, you know, the holdout senior partner or whatever, that was like, ah, this AI stuff, I've been hearing about it for years, doesn't do anything. Once he had that in Outlook or in Microsoft Word, then you thought maybe he would be more likely to use it. The problem is it dramatically under-delivered is I think the latest statistic I saw was that of all pilot projects for Microsoft Copilot, I think only 1.7% elected to go forward with procuring it for the entire institution. [28:45](https://youtu.be/cq3A73gJirQ?t=1725) So it's just really under-delivered in ways that are kind of baffling to a lot of us. But I think also that's a lot of attorneys if they're like, ah, I use the AI stuff, it doesn't work. Usually they were given some form of this, which is nothing near what the frontier models can do. And so I think it disappointed them. The other thing is, you know, obviously all of you have very unique cybersecurity and procurement, you know, hurdles that you have to jump over that somebody at a private law firm doesn't. And so when they see these national surveys and they look at government, a lot of the answers on here are just, we don't know. We don't have access to this stuff. Our procurement policies won't let us get it. Maybe we have a pilot for Microsoft Copilot, but that's kind of it. We're not using agentic AI. And I also am bullish on the notion of local models. So as Jennifer touched on, you can build your own. You know, open source is kind of three to nine months behind the frontier models. We do this here. We have, previously when I was at the University of Oklahoma, we had a NVIDIA Spark that we ran models, local models on, because we wanted to use FERPA data and the clinics wanted to use their client data. [29:51](https://youtu.be/cq3A73gJirQ?t=1791) Well, that's all in-house. That all sits on the model. It's not connected to the internet. And you're leveraging, you know, just three, four, five months old AI technology, which is good for a lot. Kind of simple AI workflow, you know, simple workflows for attorneys. And I would imagine that that will continue in the future as, you know, more graduates come out of computer science programs, diffuse into workplaces, and they can actually, you know, bring this AI tech stack to you and easily implement local models. I would also like to point out that there is an open source version of Harvey and Legora out there. This is Will Chen. He's an ex-Latham & Watkins guy. I think that works in a non-profit now, and he vibe coded it in about two weeks. You can also run this completely locally on your own servers. And yeah, the future is going to be weird with all this stuff, but there's a lot of options out there for you. So that's all I had for right now, Beth. I will get rid of that. **Agueda Quiroga** [30:48](https://youtu.be/cq3A73gJirQ?t=1848) Thank you so much. I think we more than got our money's worth, and then some. Even not being free, it's more than our money's worth. We would pay for both of these presentations. Fantastic. You already have a ton of questions in the chat. So I will be brief in asking you a few from me. That'll give you guys a few minutes to also scan over the questions. Anything that's like a one-word answer, please feel free to just type in the chat or in the Q&A. But let me ask both of you just to reflect really quickly, again, not all about the process, the problems that are best solved with AI and the ones that aren't. Here, I'm not talking about everything in the universe that could be done. But for people who are thinking about where do I start? What's the first thing I could work on for myself or with my team? What, in your view, is like one place that they should start? John, you've given us 50 different things. Just as a category of place to start, what advice would you give people about the kinds of problems that are best suited to AI and those that are not? **Sean Harrington** [31:56](https://youtu.be/cq3A73gJirQ?t=1916) Yeah. Well, first of all, if you haven't ever paid for one of the Frontier models, you should do that. I'm pretty agnostic on which to use. But if you've never... It's $20, and I think it's worth it just so you understand where the current state of the technology is. For one month, try out Chat GPT or Claude or one of those things. And watch an up-to-date YouTube tutorial from the last month or couple of weeks that runs down all the features very quickly. And try to imagine how you would use that in the context of your job. First of all, these things are really well-suited to things like summarization, manipulating text. So that's one of my favorites. It's kind of under-reported things, which is... Excuse me. I have a PDF. I need to get this into a Microsoft Excel document, and I want some charts and visualizations. It'll do that no problem. Very easy. So that's like a very tedious kind of data monkey type thing that just all of a sudden, AI just knocks out very easily. Sean, I want to just ask you about the... Real quick, before I go over to Jennifer, is the product space as opposed to your own individual use. When people are thinking about the kind of project that Jennifer did around brief writing, do you have a product that you think, this is either the plate... We're going to go partner with ASU. [33:13](https://youtu.be/cq3A73gJirQ?t=1993) What should people call you for? What thing would you like to build with them? Yeah. So we're building... We're currently doing things because AI creates problems at scale. So we have a... We're an initiative. We're working with Maricopa County Courts. And I hear this from AG offices too. So I'm going to be relevant to all of you, which is they're just getting tons of slop filings at the state court level, where it's not even that people are churning out 20 page complaints full of AI slop, which is definitely happening. But they're also doing things like procedurally, maybe it's at the stage where motion for summary judgment, and all they have to do is write a response to motion for summary judgment. Well, they'll write that, plus then a motion to strike, plus then some discovery motion, plus the sanctions motions. So they get like 500 pages that are even at the wrong stage of the litigation. And so we're trying to build something for the court clerks where they could feed that in, find out if there's anything useful, find out exact reasons to reject the filing back to the pro se litigant, and explain why. And do that quickly with the human in the loop, so that it would just get the human to exactly the spot, explain the justification, the human could go, yes, I agree, or no, I don't. And then they could send that out to attempt to triage this. I hear this with AGs, with things like FOIA requests. Like anything where you have a statutory window where you must respond to the thing is getting really dicey now, because you're just completely overwhelmed by the volume of the content. And so in many cases, I think, using AI to fight AI is probably the best if only option that we have here. But people have been slow to move on that. And I think for reasonable reasons, because it impacts things like democracy, people are like, we don't want AI making those types of very important decisions about humans. So we're trying to create, you know, either human on the loop or human in the loop systems for that type of thing. [34:58](https://youtu.be/cq3A73gJirQ?t=2098) And I would love to partner with any of you if you have ideas on how to do this. Jennifer, let me go over to you, sort of how did you think about the problem that you wanted to solve first? Obviously, it started with the problem rather than the tools. So how should people now, having been through a bit of this experience, where would you encourage people to start to say this is a problem we want to tackle where AI could help and where AI can't help? **Jennifer Sellitti** [35:23](https://youtu.be/cq3A73gJirQ?t=2123) I think by taking the AI out of it in the beginning, I think you think about what is what is the problem for us. And I think for many public defender, prosecutors offices, it's caseloads, right? We want our lawyers to be in a perfect world, our lawyers would have fewer cases. So for me, it was what are the things that people are spending a lot of time on in their cases, which was this kind of process of cobbling together draft briefs. I remember I went to a conference, actually, through rest introduction, went to a conference in Spain, on AI with different people from different countries there. And one of the speakers said something that I that I haven't forgotten that kind of crystallizes this for me. And he talked about an old Black and Decker commercial from back in the day for drills. And they were in the advertising room and the advertising executive was saying, well, the drill does this, it does this, it's pretty, it has this thing on it as this thing on is this type of motor and whatever. And the advertising executive said, that is not what we're selling. We're not selling the drill, we're selling the hole in the wall. So I think the the real question that we have to ask ourselves is what is the hole in the wall? What is the problem that is keeping our attorneys up at night? What is the thing that's keeping in my case? What is the thing that is keeping my attorneys from spending more time with their clients more time at the jail? And then start to say, okay, that's the problem. What is the potential AI solution? [36:49](https://youtu.be/cq3A73gJirQ?t=2209) And what is the right tool to create that solution? I think the mistake a lot of people make, and it's the mistake that I made myself in the beginning was as cool, it can do this, it can do this, it can do that it can do anything like we can make it I'm anything I can make anything better. But what is the problem? What are the contours of that problem? And then then start thinking about what is the AI? What is the technologically based solution? And who is the best person to build that? We have so many questions waiting. So I'm going to just do a couple quickies. Jennifer, let me start with you here. Maybe you could just walk us through a little bit how you explored that problem with your team. I'll be brief, because you know that you know what I'm trying to ask you about how you brought your team into really designing this. And I think that was what was a huge differentiator is this was not something imposed by a group of techies. It wasn't Princeton University who came to you. You really sort of prototype this with your team. And what did that look like? Who was around the table? What did you do? [37:54](https://youtu.be/cq3A73gJirQ?t=2274) Yeah, so we started by bringing in people from our MIS department, our technology department, but then quickly realized that this was a lawyer problem, right? So we went to our appellate section to the head of our appellate section and put together a group of, you know, for lack of a better term, super users, who started sitting with Princeton and the Office of Innovation and thinking about how we could build this thing together. And all along the way, we had appellate attorneys trying to break it. We had appellate attorneys testing early versions of this, trying to get wrong answers from it, trying to get weird answers from it, and then over time, use that group to refine it. So it really truly was a collaboration with the people who are the ultimate end users of this, with the students at Princeton, with the Office of Innovation and within our IT department, to over a period of time, populate this with a small number of briefs at first, test it, keep it from breaking, and then add and expand the number of briefs that were in the brief brinks until finally we had a product that we could start testing with larger numbers of people. And let me ask you both, maybe Sean, let me start with you on this one and then come back to Jennifer, which is about the ethics of all of this. We have a lot of questions also, there's a lot of dimensions to the ethical questions. We have a lot of questions coming in about workers and the role of workers and how AI can replace workers, etc. But let me just start by your views on thinking about sort of governance and ethics as people begin to adopt these tools. [39:23](https://youtu.be/cq3A73gJirQ?t=2363) Sean, what have you seen work? What are you excited about? What are the guardrails that you recommend putting in place, especially as you're showing us a lot of commercial tools? To be fair and realistic, these are the tools that governments have procured. We can want to live in a different world, but our IT shops have signed contracts with these big providers. What as lawyers do we need to do differently? How do we think about governance? What are the sort of most urgent ethical issues from your perspective? **Sean Harrington** [39:55](https://youtu.be/cq3A73gJirQ?t=2395) Yeah, I mean, as far as automating away human workers, I think, you know, get together, people should be getting together with all, you know, a representative sampling of stakeholders within your org, you know, like paralegals, for example. So paralegals are the ones like elbows, you know, rolling up their sleeves, elbow deep into the documentation. So they're frequently some of the best people to bring in as stakeholders to think this stuff through because they're the ones that are day to day grinding through it. And they know the pain points and what can be automated and that type of thing. But I think that the bigger question needs to be, as Jennifer said, number one, what are we trying to solve here? What's the real problem? Number two, what are things that we just never want AI to touch? What are, you know, kind of the three categories from the COINTELLIGENCE book that was viral for a while there? Things that we would never want AI to touch, things that we're happy with human in a loop interaction with AI and things that we don't care and just are perfectly happy automating away. And that, I think, gives you at least some strong guidelines on the things that you just never want AI touching or places where you just think the human component of it is just too important. And I think that's a generally good starting point. As far as policies, there's databases of policies for firms, for government. And the benefit of those, well, what I would do with those is take all of them, throw them into an AI system and have a conversation and kind of cobble together the policy that makes sense with your org. [41:22](https://youtu.be/cq3A73gJirQ?t=2482) Having worked now at four different law schools, I will say, and this is just law schools, so I'm sure government's completely the same, which is they're all their own little kingdoms with their own little pain points, with their own unique workflows. And so I think it's probably better to kind of take from what's out there, but really tailor it to your org to make it unique. Jennifer, let me pass the ethics question to you. And I think, again, it has many different dimensions to us. But what have you seen that gives you pause as far as how lawyers, especially in government, are using AI or not using AI and how it's being governed? **Jennifer Sellitti** [42:00](https://youtu.be/cq3A73gJirQ?t=2520) So it's one of the things that has kind of struck me lately is different states' attempts to create ethics policies, professional rules around AI, and as well as organizational policy around AI. In New Jersey, we had our state bar association and our judiciary both created committees about how to use, governing how AI would be used in courtrooms and how client data could be protected internally. And on the courtroom side, right, on the AI being used to create documents for court, I have seen states doing things like making you file a certification when you have used AI or requiring you to disclose to your client that you're using AI. And to me, that is that is not the right approach for a couple of reasons. I think, one, because half the time, we don't know if we're using AI or not. And I think another piece of that is AI is no different than any other tool. When I first started lawyering, I used to bang on the wall to my good friend, Alex Shalom, who had the office next door to me, and I would scream, Alex, here's my problem. What's the case law for that? And he would scream something over at me. [43:12](https://youtu.be/cq3A73gJirQ?t=2592) AI is just Alex, right? We do this every day. We rely on paralegals. We rely on colleagues. We rely on interns. We rely on Google and all sorts of things to get information to better serve our clients. But at the end of the day, attorneys are responsible for it. If I file it in court, I am responsible for it. So whether Alex said it or I said it, I'm responsible. That, in my opinion, should be the rule, that the same old tried-and-true ethical rules that we have right now apply. I am responsible for what I file in court, and I am also responsible to not put my client's data into systems where it can get out to people outside the firm. Now, again, in New Jersey, we're lucky when we're looking at products because the internal product that we have, it was actually built using Cloud AI, but it's proprietary to the state of New Jersey. So it has much the same feel and features, but it's built for us, and we can put client data in it. If you have to use a publicly available tool, you need to sanitize it in such a way that your client's information is not getting out there. And I think there are ways to do that or ways to pay for tools that will not put it out there, but that you have to be really sure of that your client's information isn't getting out there. But I think when we start looking at rules, when we start looking at how this changes evidentiary rules, how this changes what we're telling clients in courtrooms, I always go back to take AI out of it and substitute AI with my friend Alex next door. What would I be responsible for if I got the information from him? And I think that's the way that states and bar associations and courts and judiciaries should be approaching how we look at AI coming into courts. So I think you're 100% right. This always reminds me of how we used to call... [45:00](https://youtu.be/cq3A73gJirQ?t=2700) buying a book on the internet, e-commerce, and now we just call it commerce. So, you know, the ABA initial guidance says you have to disclose. Everybody talks about disclosing and presumably version two of the guidance will say, well, you know, that ship has sailed already. Everything is using AI and we have to get a little more sophisticated in how we think about these things. We have a lot of sophisticated questions in the Q&A and I want to turn to some of those. A few I'm going to put in the chat because they're just sort of one word answers and invite you or anybody else to kind of respond to them. But I want to ask you about sort of come back to the commercial tools versus public sort of buy versus build questions. There's a bunch of questions we have in here. Some people sort of questioning about sort of the wisdom of commercial products. You already told us about some open source alternatives. We have another question in here. Oh, sorry, there's so many coming in, I've already lost it. But there's a bunch of things in here about sort of kind of the wisdom of relying on these commercial tools. [46:00](https://youtu.be/cq3A73gJirQ?t=2760) And without getting us too far down the rabbit hole of the discussion about public AI, we'll put a link in the chat to a whole series we did on public AI. But anything as a field, do you think we need to move towards a sort of more open source ecosystems, smaller models that we build for ourselves? Do you feel like in the legal context in particular that there's something we need to be doing differently or do you think it's fine to use Claude for all the cool things you've showed us that Claude can do or notebook Helen can do or whatnot? **Sean Harrington** [46:34](https://youtu.be/cq3A73gJirQ?t=2794) I think that there's instances, especially like niche practice areas where having like a fine tune local model would be like provide a lot of value. The benefit of the local model is you have none of the data concerns, nothing's leaving the office. You can put in literally whatever you want and kill it with a hammer if you need to. Again, it's not super expensive to build these things. We did it with a $4,000 max studio at OU. We have, it's like a $5,000, it's like a Dell box. It's like their equivalent of that that runs AI here. And we've had up to 30 and 50 concurrent users using it and it moving just as fast as open AI and running a model that kind of produces, runs at about 97, 95% of the capacity of the frontier models. But I think we're just never gonna be able to compete with the infrastructure that it takes to power these things. You know, the nada about this would be like Facebook, chips, power, all of those. And so on some level, you will always be behind the times if you're using an open source model. [47:36](https://youtu.be/cq3A73gJirQ?t=2856) Let's say there's a lot of value in having open source models in the open source community in general building things. And I'm a huge open source nerd and kind of always have been. And so I think, you know, a lot of the world is powered on open source software. Every PC of every person who is, you know, on this call right now is running some form I think it'll be probably a mixture of both is what I foresee in the future is sensitive stuff, local open source AI that you, you know, control completely non-sensitive stuff where you need frontier level intelligence. You have no choice, but to go to one of the frontier models. And Jennifer, how are you thinking about sort of going forward similar questions and collaborating, for example, with a university to build something as opposed to if Harvey were to come to you and say, we're gonna sell you a tool, would you wanna buy it? Or do you prefer the role of your own and why? **Jennifer Sellitti** [48:26](https://youtu.be/cq3A73gJirQ?t=2906) I think I would much prefer to build my own in a perfect world to be able to have something that's truly customized to our attorneys needs. That said, like I said, we are using Westlaw AI that's just the product that the state of New Jersey uses for this. And so I agree with Sean. And I think the more niche and the more sensitive the information, the more you want to build something versus if you need something that's gonna scale, if you need something that is doing more generic types of work in the legal profession, you're gonna wanna buy it. I do, I think, wanna flag though, one of the things that we're seeing and one of my big fears in this brave new world that we're going into is that it's not the big law firms that are gonna get behind. And it's not, at least in the state of New Jersey. And I know that there are public defender's offices that are in very different situations. But it's not us that's gonna get behind because we have a government, a governor's office that is very committed to empowering us with these tools. [49:28](https://youtu.be/cq3A73gJirQ?t=2968) It's the small mid-sized firms that are really getting squeezed here. So one of the things that we are looking at, I saw a question for Laura Sutnick who, hi, Laura, I'm trying to answer your question. I can't 100% answer your question. But I think it is really difficult for the middle-sized firms to do this because it is expensive to build. And then it is expensive, really expensive to buy these very large products. So we have been thinking about ways to partner with other government agencies and to partner with the legal community to see, are there ways that we can help the people who will often take our conflict cases, our pool attorneys, help them get the benefit of some of these AI tools without having to make such a big expense? We don't have the answers to those questions yet, but I think this is one of those rare times where at least in my state, that the public defender's office, the government is not gonna suffer. And the big firms are gonna suffer. [50:28](https://youtu.be/cq3A73gJirQ?t=3028) It's kind of the people that are in the middle. So I'd like to see more solutions for those size firms so that everybody keeps pace together. There was a, let me shift to some of the questions actually about workers that are being asked. So the question from Brad Donovan essentially about whether we're going to be pushing people towards more and more productivity and what that's going to mean for workers. I'm not doing good justice to his question, but managing the doing more with less he talks about. Will you expect more output for the same headcount at the same pay? So I think the question is, what pressures are these going to create? Maybe Jennifer, this is to you and your former colleague, Robessara Angelo, also asked just as a side note about workers, how you created the new, how long it took to set up the new career track for the secretarial pool. So I'll just throw that in there besides, but I'm wondering if you can talk a little bit about what this is gonna mean for headcount, pressures on workers, and how do we sort of keep workers at the center and on top and AI on top? [51:34](https://youtu.be/cq3A73gJirQ?t=3094) Yeah, so I don't see this as, we're not looking to increase people's caseloads. Caseloads in New Jersey are what they are. I don't know if people are familiar, but a couple of years ago, there was a caseload study that was done in the public defense industry. Our caseloads at the time were about two and a half times higher than what the caseload study said that they should be. And that was actually a decent score because in some places there were people who were much, much higher than that. But we took that seriously. And it's not a silver bullet, right? AI is not the solution to caseload, it is a solution. One of the things we did was we brought up, we committed to bringing in more attorneys. We asked the governor's office to give us more attorneys. So we hired more lawyers. But there's a point at which you have to say, how can I take lawyers, stop their caseloads from going up, but make it easier for them to manage these cases that they do have, that even in a perfect world are not gonna go down to like private law firm levels. They're just not, that's not what the public defender's office is about, right? [52:35](https://youtu.be/cq3A73gJirQ?t=3155) So we're looking at using AI wherever we can, like the brief bank, like ways to go through video footage, surveillance footage, we're seeing, and I know everybody on this call is seeing it. It used to be, if you had a theft, you'd have like one surveillance camera from the convenience store. Now you have eight body cameras, three different camera angles, statements that are on tape, right? Every case is exponentially larger because of the volume of video evidence. And then forensics are complicated because now every forensic scientist is not just looking at things with their eyes, they're looking at it with their eyes, they're looking at it with software, and now they're looking at it with the latest AI version of that software, right? So it's all getting exponentially bigger. So what we have to do is as the amount of data in our cases and the amount of complexity of forensic science increases, say, how can we use AI to go through that data? And AI, like Sean said, is really, really good at going through massive amounts of data and summarizing it. [53:39](https://youtu.be/cq3A73gJirQ?t=3219) So it's not going to replace a lawyer who still has to watch all eight body camera footage things, but it is going to replace having to watch it again and again, because now I can say, now show me all the places the best face shows up and it'll pop up, right? So I am not trying to say that we're asking for more productivity from our lawyers or more cases from our lawyers. What we are really clearly focused on is how do we keep the cases as they are or continue to lower them and at the same time, allow our lawyers more time to spend time on the parts of the case that cannot be replaced by AI, which is human to human client contact, human arguing in courtrooms and the emotional parts of our cases. **Sean Harrington** [54:22](https://youtu.be/cq3A73gJirQ?t=3262) So I love that. I'm very, you know, that is very reassuring to hear because in other industries, what we've seen specifically, like there was this Berkeley study that came out where they gave people AI and then they checked back on them like eight months later. And it wasn't a huge sample size, but it was decent. And they discovered that at some orgs, what happens is they discover, oh my gosh, I have AI. I can now do things that I could previously never do at my job. I had to go find somebody else who could help me. I had to give it to the IT department, you know, something like that. And they discovered that, oh man, I can do all this stuff. And so while their productivity increased, their workload also increased like dramatically. And so that's some conversation to have when you're thinking about rolling out this stuff, which is, hey, you know, yeah, I'm gonna be able to do more. I'm gonna be more productive, but there has to be a limit to that. You know, like what is the real limit? You know, at what point are you just doing too much and gonna just completely burn out? And there's a lot of people in the tech space that I've talked to that, you know, they worked at OpenAI and now they're just like basically retired. They're like, I just can't keep that pace anymore. It's just too much. [55:22](https://youtu.be/cq3A73gJirQ?t=3322) So we have only a few minutes left. Lots of good questions. Again, thanks for answering some of them in the chat or asynchronously. Lem, we always wrap up these things with kind of a call to action. What one piece of advice do you have for people here? If they wanna start experimenting responsibly and ethically in their own shops, what would you recommend to them to do? What's the realistic first thing you want them to do, first project, et cetera? Sean, let me start with you and then Jennifer, you can close this out. Yeah, this is gonna be maybe different than what you expect, but don't do it at work, do it at home. One of these weekends on a Saturday or Sunday when you got some free time, fire up Cloud Code, I mean, OpenAI Codex or Cloud Code and try to just build an app. Just tell it, I wanna build a meal prep app. I wanna build something to help my wife do X, Y, or Z and let it grind for an hour or whatever it does, answer its questions and then discover, oh my gosh, I'm a software developer now because that's a light bulb moment. [56:25](https://youtu.be/cq3A73gJirQ?t=3385) I think not enough people in law and specifically in government have had, which is, well, these things are incredibly capable. So yeah, it's 20 bucks, but it'll change the way you think about this technology. **Jennifer Sellitti** [56:39](https://youtu.be/cq3A73gJirQ?t=3399) And I agree. I think one of the most important things is that people do start using it, to use the available tools in a responsible way and get really comfortable with it. I think particularly for defense lawyers because we're not only on the offensive use of AI, trying to use it for the betterment of our client and our client relationships, we're on the defensive side. Every defense attorney really has to know AI inside and out because we also have to be on our game about keeping bad technologies out of courtrooms to be used against our clients. So at least for the defense attorneys out there, get very used to using it, use it in your personal life and then start thinking, how can I use this? Even if it's just like last night, I was really angry at somebody in my personal life, not professional, and I typed an email really super angry. And then I said, can you make this sound really loving and reassuring? Cause I know this person had a good intention. And in two seconds, it may be the most perfect email that you could have imagined, right? Start doing things like that. [57:40](https://youtu.be/cq3A73gJirQ?t=3460) And then think about how could I use that in a case that with an opening or closing or something small, and then build from there and take it one step at a time, one project at a time, spoken from a girl who tried to bite off all the projects at once. Start with one thing, really work on it, refine it, get people using it and then move on to the next hole in the wall, if you will. **Sean Harrington** [58:02](https://youtu.be/cq3A73gJirQ?t=3482) That's really good advice. Having one good project that will get over a lot of people, other people's like worries, either ethical worries or their hesitancy with the technology. One that works really well will go a long way to more adoption. I think that's a really good point. It gives me a final chance to sort of put in a 10 second plug for if you want to get your teams around that table to talk about the problems you wanna solve, there are the recordings from this series as well as a one hour course on AI for lawyers that's all free. We will put links in the chat to the course. All of these videos will be available beginning tomorrow. And with that, I just wanna thank our speakers today and in the whole series and to all of you for joining us. Thank you so, so, so, so much. There's a two question survey you'll get afterwards. One of which is, did you like this? And the other of which is tell us what you want more of. So please tell us if you want us to do this again, have more conversations, have longer conversations and we will make it happen. Thank you to everybody for joining us and thank you to the team for putting this on and to you for your questions. [59:07](https://youtu.be/cq3A73gJirQ?t=3547) Please join us in the future. Thank you everybody. Take care. Bye-bye. * * * Original page: https://innovate-us.org/ai-legal-products-showcase · Licence: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org