Tim Kowal And now your hosts, Tim Kowal and Jeff Lewis.

Jeff Lewis Welcome everyone. I am Jeff Lewis.

Tim Kowal Well, Jeff, we wanted to start today with ClioCon 2026. And our audience members might be wondering, what did I miss at ClioCon 2026? Does the idea of Cleophone give you FOMO or Panopticon vibes? Or maybe you didn't even know ClioCon was happening. Maybe you don't know what cle what a ClioCon is. So Jeff, we'll ask you to report on that later. But first, we wanted to take stock of where we are. In our AI journey as attorneys, and I guess maybe large more broadly as a species. It's now Q4 2026. And where did we start with AI and how is it going? In 2023, AI first passed the bar exam in 2024. Many or most of us, depending on what poll you look at, most many or most attorneys were already using ChatGPT, and the use accelerated into 2023. When the judiciary struck back and started issuing sanctions that have not really abated. In 2026, we were introduced to AI skills, a feature within AI platforms that allows lawyers and everyone else to not only perform a task but to follow a workflow. And lawyers can now build their own tools, including to catch bad citations. And building your own custom tools is where Josephine Petrick fits in. Josie is a Practicing appellate lawyer, an advisor to a legal AI company, and a builder of her own AI and legal tech tools. And this year she was on the winning side of one of California's Supreme Court's biggest decisions. Josie is a partner at the Norton Law Firm in Oakland. Josie is in Los Angeles, and she's a certified appellate specialist. She advises BriefCatch as its chief of AI products, a part-time role alongside her practice. She clerked for Judge James Dennis of the Fifth Circuit and practiced appellate law at Hanson Bridgett before joining Norton. And she's also built her own really cool AI tools that I want to ask her about. She's built Askruly.com, an AI search tool that draws from a verified database of California court rules. And she built the California Supreme Court civil case tracker. Both of those tools are now run on Lexponent at Lexponent.ai. Josie, welcome to the podcast. Thank you for joining us.

Josie Petrick Thank you so much for having me, Tim and Jeff. It's it's a real pleasure. I'm a huge fan and it's such an honor to be here.

Tim Kowal Well, we've been talking about bringing you on to the podcast for a while. You and I first met just about a year ago at the at the appellate section holiday mixer in Los Angeles last year. I remember at that time I was still using ChatGPT and you said you really should give Claude a try. And it was just a couple months later that I think they announced like I think it was Cowork, it was their Claude Code product that was announced for the for the masses, people who are Who are going to be you know shocked and appalled at seeing a terminal window could still achieve the same kind of effect using Cowork. And they announced the use of these skills that you could use to create these workflows. And I had had this database of like all my all my ChatGPT prompts that I would copy and paste depending on the task. And now you could build your own workflows directly in. That was That was a huge inspiration. So and I and I always remember now that I use Claude for everything now, and I remember our conversation when you first told me that you should really start using Claude. And I wondered you know, AI is like a it's like a Rorschach test for attorneys and I guess for everyone else. It either inspires dread or excitement. And lawyers, we lawyers by nature are, you know, we're we're we're not very tech embracing by nature as a profession. And add on to that our, you know, tendency toward OCD and security and ethics concerns on top of it. How did you come to hold an as you know both an esteemed place in the profession while also embracing AI?

Josie Petrick That's a great question. And it's so funny that you have now embraced Claude. I didn't realize I was the maybe the inspiration for that. I love Claude for my legal practice and for my startup work. And I also use ChatGPT still. I just can't seem to quit either one of them. So they each have their beloved place in my in my repertoire. But let's see, I started playing with it when it ChatGPT actually when it was still a sandbox before it was a chat. Chatbot and I just you know I got tipped off by friends who were in the tech sector and they said my gosh it can it can produce code check this out and I said I wonder what can it do what it can do and it was really bad at producing poems I don't know why that was one of the it was horri like abjectly horrible. It would make stuff up I found out right away. So this hallucination business was not a surprise to me. I was never blindsided by that fortunately. But it was also really good at tracking down information. And it was just shockingly plausible as a conversation partner. And I saw the potential. And I was a linguistics major. My father was a mathematician. And so just this sort of subject matter always just really interested me. And I just used it a lot. And one day a couple of legal tech founders approached me as a potential customer. They were building for they were building appellate tech. For a record review. They ended up getting YC backing. They were called ABL. And I was just such an enthusiastic customer and would provide them all kinds of feedback. They eventually asked for me to become an advisor for a pro product advisor for their board. And they didn't have a lawyer on staff at the time. So I gave a lot of product direction. But and that one thing led to another and I just fell in love with building my own tools. And you know, you said we're lawyers are a not necessarily a tech forward bunch, but we are problem solvers. And we're creative problem solvers. And I think that now with large language models, what we what we generally call AI these days, it's such a natural fit with law. I think it's gonna keep you know, we work with words, words are our code, and now that we can talk to computers in that same language, I think there's so much potential for lawyers, even those who those of us who consider ourselves tech averse or maybe shy about using tech. Any of us can innovate solutions to problems. So that's that's how I really got into it. But I remember you and I bonded over not only Claude versus ChatGPT, but also use of a Notion database or Notion databases to keep track of various practice pointers or unwritten rules. So that was I s actually still use Notion for that as well.

Tim Kowal Yeah well you built you built the Supreme Court civil case tracker on a Notion database. And w I want to ask you more about that in a little bit. But let's talk let's talk about the way that that we problem solvers you know first you know thought to leverage AI, which is to I guess to To solve the problem of writing a brief very quickly and without apparently a lot of thought or oversight going into it, which led to a rash of sanctions and a lot of consternation in the profession. Back in April, we had Ross Guberman, the founder of BriefCatch, on the podcast. And we called that episode the end of hallucinated cases, with the with the introduction of the RealityCheck product. But the sanctions haven't quite stopped. This spring, the Sixth Circuit sanctioned a lawyer whose fabricated quotations came out of the AI tool built into Westlaw, and SB 574, which the governor signed in September, will now require lawyers to personally verify every citation in a filing. So Josie, what's what's your verdict with BriefCatch’s RealityCheck product and others? Like the checkers built into Westlaw and Lexis, and everything else out there on the market, including the types of tools that attorneys can build on their own. Is there any excuse left for filing a c a brief with a fake citation?

Josie Petrick No, Tim, and I don't think there ever really was an excuse. I mean, there's there are explanations, sure. You know, use out data. It and it's not limited to text generated by AI. It can be misuse of a case. Maybe you're citing to and this is this has been a problem for decades, you know, citing to a party's argument that's being described in a court opinion instead of the actual holding, r inverting a holding that happened recently in the California Supreme Court Lawyer got sent to the state bar for that. Over overstating authority, getting a quotation slightly wrong. I mean, these have been problems for a while, but they've gotten worse. And so I do think, you know, there's never really an excuse because we owe duties of candor to the court and professional responsibility. But there are certainly explanations for them. But the good news is, like you said, these tools are coming out that are making it a lot easier to conduct that verification. Not to substitute verification, but to really facilitate it. You know, pull up all the cases in the brief at once, you know, flag ones that don't look right, provide links so you can easily jump to them and identify you know, maybe it was text produced by a colleague or a local counsel or you know, maybe from a template that's outdated, or maybe AI for those of us who are drafting with AI. And I think facilitating it is really a huge step in Improving the these outputs, but and also educating the bar and the bench about the limits and proper uses of these tools.

Tim Kowal I wanted to ask you about well another question about the RealityCheck product and other related checkers. And the question is it's it's a critical question about the checkers because I, you know, I embrace them as someone who wants to both leverage AI for you know for broadening and enhancing the research and analysis that goes into a brief, and adding that analysis without having to, you know, necessarily personally type every Every citation and every id and make sure every period behind every id is italicized and the like. Is using RealityCheck or a related checker a substitute for verifying? And isn't it going to be and I part that that question is probably rhetorical. I'm sure it no, I'm sure it's gonna be the answer, but isn't it going to incentivize attorneys to say, well, I ran it through RealityCheck where I ran it through the Westlaw litigation analyzer and it came back with a clean bill of health. Like so I guess it's ready to file.

Josie Petrick Yeah, I mean audit reports are certainly going to be part of the proof of compliance. And you meant you mentioned the new California law effective January first. Audit records are going to be important, but just like you said, just using these tools and running it through them, they facilitate the attorney's review. They can't substitute it. So you know, I think that's a complete answer, but it's it really, you know, not just making not just catching hallucinations and Know the types of errors, sort of mechanical errors, but also like a senior a lot of these tools, not all of them, are offering a way to check for overstated authority, like get a senior partner's read on am I fairly characterizing this case? Is there a fact I you know maybe conveniently or not so conveniently omitted in the drafting process. Maybe I was trying to make the brief shorter, more succinct. Did a critical detail get lost? I mean these tools now are getting so good that they can point that out And really improve the work product, not just do a CYA check at the very end of the process. So I do think it's a process in which attorneys should be very closely involved and not just hand it off to an yet another tool.

Tim Kowal Does that does that that that seems to be part of the answer to the question I was gonna ask you next, because you on another podcast or an interview, you had said that we shouldn't treat human only work as a gold standard. And I thought that was interesting. And you pointed out that humans also misquote cases. They also they also overstate cases and authority, especially when they're doing so on a deadline. And the standard that you proposed instead was layered review. And I thought that's interesting and that struck a chord. I suspect that's right, but I wonder if you could unpack what you mean by layered. How does an attorney conduct a layered review?

Josie Petrick Yeah, I totally agree. I mean we've actually run RealityCheck what we're current the BriefCatch’s verification tool on briefs that were briefs and opinions that issued before AI came out and we're finding the same types of errors, right? Like overstatement, getting a year wrong, kiss wrong, holding inverted. You know, hu humans are fallible, just as AI currently is. You know, we can talk about the jagged edge, but if we can combine Forces, you know, have the human judgment, human sort of taste and discernment, human checking, and then add this other layer of AI verification, make sure we're not missing anything. I do think that the gold standard shouldn't be measured by who is conducting the review, but what the output is and how reliable is it. So that's that's really eliminated by that. I mean there are there are layers within the AI review, which I think is also an interesting Interesting topic, but yeah, when it comes to verification, really it's just I think we work better together.

Tim Kowal Yeah. Well can I get your can I get you to weigh in on the new requirement under SB 574, the one that requires lawyers to disclose their use of generative AI to the court whenever they're submitting a brief that made use of generative AI. And I wondered if it seems like there might be might be two camps forming around it, or I don't know, maybe there's only one camp. The only camp I know of is that You know, you just use a boilerplate that says, Yeah, like everyone I use generative AI, you know, do your worst.

Jeff Lewis It's like the it's like the disclaimers on the pharmaceutical ads, you know, if you take this pill, you may suffer from one of the following ten disastrous side effects. You know, yeah, we may have used Westlaw, Lexis, or Clearbrief to make this brief. Cover all your bases.

Tim Kowal And I suppose another alternative might be to, yeah, be more even more specific and say we used AI for the purpose of researching, but not for drafting or for something else.

Josie Petrick Yeah, I think that's where it would come down to. I mean, at some point, you know, you might s some firms might take the approach of over-disclosing, and so it's, you know, over-disclosure is kind of like disclosing nothing. I mean, this may be a cheeky response, but I picture these disclosures starting to sound like unpaid vendor advertisements. Like, you know, our firm uses the latest advances in natural language processing, including tools that leverage large language models, Westlaw, including its AI overview. Lexponent to assure compliance with procedural rules, BriefCatch to refine drafts and verify. You know, it's you could go that route. You know, we don't delegate work to you know, whatever the, you know, the other component of SB 574. You can't delegate the practice of law to an AI, whatever that means. But all output and citations are verified by the responsible attorney. And I think that's the important piece there. And it's not just AI generated output, it's now it's all citations, which I think is an important piece of that law as well.

Tim Kowal I've personally I've I've gotten even more paranoid about checking my citations. In the past I might have just copied and pasted from, you know, the standard of review from another case. And I didn't reverify it for this particular brief. But if I'm ever going to use AI to assist draft a standard of review section for example, you y you'll bet that I'm gonna look up the that case for that you know, that that perfectly anodyne statement of law because I'd Do not want to be a cautionary tale. So I don't know. I guess maybe there are maybe there's like a bell curve effect or something with AI. It's lawyers are either being more derelict in not checking anything or they're being you know more aggressively dil diligent in checking everything even when it's unnecessary.

Josie Petrick A hundred percent. And you know, maybe in a sort of an ironic way, maybe it's improving the or maybe it will continue to improve l lawyers' briefs and the quality of the output. But yeah, I like your Bell curve analogy. That's really good.

Tim Kowal So you're a you're a practicing appellate partner. How did you end up advising BriefCatch on its as its chief of AI products?

Josie Petrick Well I knew Ross Guberman by and BriefCatch by reputation. I use it you have long used it in my own practice. And we connected when BriefCatch was expanding beyond rule-based editing skills, what we call like the deterr deterministic rules, their core rules engine, to a holistic editor that could assess legal writing more substantively and based on Ross's best selling books, point made and point taken. And he needed experienced practitioner judgment to evaluate the quality of AI suggestions and help refine the prompts. And so I started with a focused sort of consulting project, just giving him feedback, reviewing outputs against how I would critique briefs and how courts have sometimes critiqued. We had a test set of briefs that we that were known to have been critiqued by courts as an as an appellate specialist. So I loved the work a lot. And what I ended up doing was not just testing, but Suggesting changes to the prompt, to the prompt architecture, and from there moving on to accuracy, verification, and user experience and just eventually organically involved into my role leading AI products. So I you know I'm not leaving law practice behind, but was I'm really enjoy this opportunity to apply it at scale.

Tim Kowal Well, what I was so excited about with your elevation to that role, i mean it was a it was a confirmation of the great work you're doing and creating your own your own legal tech products. I think that that's so exciting for an attorney not only to be, you know, sitting at the consumer end waiting for the big guys to r release an AI product, but you're you know, you're leaning in to it and realizing that that look with AI you don't even have to know how to code. So you can create your own products. And I wanted to talk about your Supreme Court civil case tracker on Lexponent.ai. So if you're listening and you're and you've got a you got your phone or if even better if you're in front of a computer, type in Lexponent.ai and then find the link at the bottom to the California Supreme Court tracker. And it's it's just An amazing resource and it's something that that I use and my firm uses periodically to when we want our attorneys to go find some interesting cases that we might be able to offer a unique angle or unique perspective by filing an amicus brief. And so we've we found one recently that we're gonna look into submitting an amicus brief on an interesting issue of appellate procedure and you have it helpfully categorized by what type of issue is involved. Incl including appellate procedure or attorneys fees or FAA preemption, those are the kinds of ones that that draw my attention. Or there's water law or there's insurance fair plan which I both of which I know nothing about so I will not get within ten feet of those. But it's it's such a great resource to help keep the public and attorneys Apprised of what's going on there. Can you tell me what inspired you to create this Supreme Court tracker and how did you do it?

Josie Petrick Well, first of all, thank you for the shout out. I sometimes wonder if anybody's ever looking. I haven't looked at, you know, page views or anything for the page and I figured, you know, I was making it for a particular c similar to what you're using it for, identifying amicus opportunities, things to write about and also to stay informed about what's currently pending, because that's not always as obvious. The court issues, a pending issue s issue summary, PDF, but You know, to really keep track of the context and you know, is this gonna be applicable to one of my cases? It's a great resource for that. So I'm happy to hear that you're using it and finding it useful. I yeah, I originally you know I started tracking them for client work, like I said, and just to stay on top of everything. But I approached it at first I had this big idea, I sort of wrote a product requirements document for myself, like okay, I want to try I want a tracker, it's gonna have all these things. It's gonna be a website, I'm gonna vibe code it. And then I took a step back after I'd written this long document. I said, okay, can this be like don't vibe a call don't vibe code a whole app if it can just be a Google form, right? So like think really small, small batch it. And luckily I took my own advice and did that and actually did a lot of the first the f initial steps by hand to see, okay, well, what are the pending cases? What are the recent cases? What are the actual workflows involved in setting up this tracker? What's gonna am to actually matter to an attorney? Because if you just hand it off to like Claude Code to figure out, like they're gonna probably over-engineer it, have all these like categories and tags that practicing attorneys aren't gonna care about. And so, like, okay, what's the really the minimum viable product for gonna be useful for attorneys? So I didn't literally make it a Google form. What I did was make it It was a Notion database, like you said. It's still accessible on Notion, but we migr I migrated it to the Lexponent page so we could customize it, make all of the fields searchable, not just the ones that Notion makes available by default. And to provide the most just to highlight the most useful information. So it's been really fun and it's been really rewarding to be staying on top of these cases and see starting to see patterns and what you know what When review is granted and sort of the themes that have been emerging in the in the courts in the courts activity. So

Tim Kowal A after you told me about the civil case tracker w after we met last year, and I realized that you had you had you had built it on Notion, I think you were the first attorney outside of Jeff and myself that I know who has made any use of Notion. And I credit Jeff for turning me onto Notion years ago. And it has a little bit of a learning curve. And but especially with AI, you can just tell AI if you have it plugged in, you know, have your AI plugged into to Notion, you can have it you know. Make a lot of your information live on Notion and it can build databases. Maybe one of these days we'll we'll have you back and kind of compare notes on the types of tool other tools that we built on Notion. Because I built my own cite check tracker tool. I built an appendix generator tool and I just vibe coded it. Vibe coding me meaning I just turned on Wispr Flow and told Claude, wouldn't it be cool if there was a tool that did X, Y, and Z. And it said, you mean like this? And it produced it produced an entire four volume appendix just to ask me if this is what I had in mind. And I thought, holy crap, you know, like look what it would it what it can do almost by accident. So it's an amazing it's an amazing tool.

Josie Petrick Yeah, I love I love Notion. I was I think you two are the also the other only other journalers I've met who use Notion, but it's fantastic, especially now that as you mentioned, it has MCP access. Me is b which is basically for those of us who don't know, it's basically like an API, but for AI agents. And it's yeah, it works seamlessly in my experience these days. So it's really, really powerful tool.

Tim Kowal You also built another tool called Ruli or Ask Ruli, which is now lives under Lexponent.ai. Tell me tell our listeners about AskRuli and why you decided to build that. It's it's another tool that it's based on a curated database of authenticated authorities, but just allows provides a an easy user interface for California practitioners to get quick answers to To California court rule questions.

Josie Petrick Yeah, thanks. Yeah. So we I started building what we now call Lexponent. We sort of outgrew the name Ruly and Lexponent now covers all manner of procedural rules, whether they're in the local rules, for in California, California rules of court, the statutes. Eventually we're gonna have standing orders as well. So stay tuned for that. But I started building it because I kept encountering the same procedural friction. You know, as appellate attorneys, you probably know how much time goes into finding and applying. Court rules, local requirements, standing orders and how this whole vertical of okay, well, there's a local rule on point. Is there a rule of court that supplements does it supplement a rule of court? Is there something in the in the code of civil procedure that covers it? And then on the federal side, you know, the rules in different courts. So sometimes the answer exists, but it's scattered and extraordinarily inconvenient to get at. The PDFs sort of spread across the whole internet. So I had experimented before with vibe coding appeal calculators, filing checklists, procedural research, and case tracking. Lexponent actually started as custom GPTs, which was a ChatGPT feature. I don't actually think they have it anymore. I looked the other day and couldn't find it in my sidebar. But they just weren't satisfactory because you had to upload the local rules PDFs yourself, make sure that they're updated manually, and it gave pretty decent answers. And they would hyperlink them, but it would not you it wouldn't pop up and show you the rule in context. And I'm very much like a don't trust verify kind of attorney. And so anytime an AI product gives me an answer, I want to look to make sure I basically I just want it to like synthesize it quickly and show me the actual page so I can verify it myself, almost like a shortcut. And so the custom GPTs couldn't do that. And my husband, who is also a lawyer but mostly a software engineer and he also has a lot of product experience, he said, I could I could code that up for you in a weekend. And it took longer than a weekend, but he did code it up. And by the way, he's the reason I got turned on to Notion. He uses it prodigiously in his own work as databases and task trackers as well. So now it's it's not just a web app, although I still use a web app occasionally, it also has MCP connections. So anybody who uses ChatGPT or Claude or other services that use MCPs, which I think soon will include Harvey and Legora and Perplexity can you know and they have a Lexponent account, they can just connect it to their chat bot and ask questions. Or they can say, okay, does this document comply with the local rules? Propose fixes. It's becoming really, really powerful. You can generate checklists, you know, anything you can imagine. I actually used it to build an appellate simulator to train my appellate associates or any of any of my colleagues who are just who might come across like an appellate question if I'm not available. They'll just know it helps them to issue spot. Like, okay, is yeah, the appeal deadlines different in federal and state court. I should probably check. Right. And so I use Lexponent as the MCP as the procedural rules engine that I could draw on as a database to help make this simulation, which it's just so fun to think of new uses that these tools allow. But anyway, that's the that's the short answer. So yeah, thanks for the shout out.

Tim Kowal If someone is listening and thinking, well, that that all sounds great. I'd like to build my own tools, but I don't have the level of sophistication that Josie has. So is this just not for me, or can you give our listeners some easy you know remedial steps that they can use, like Claude skills or something, and what kind of tasks should they start looking at trying to solve for themselves using some of the skills that that don't require knowing a lot of the terminology in knowing The concepts that you know.

Josie Petrick Absolutely. And I do think that they're becoming more user friendly and that's such a good thing. And for anybody who's thinking, okay, is there something I can I can I start using this in my practice? There are two pieces of advice I would give. One is think start small. You know, don't don't write a whole appellate brief. I would I wouldn't even hand off drafting to an AI tool. I think that's that's probably not the that's the first place people usually think of because they are so good at producing text. But I would actually look at your workflows, what's repetitive? Tim, you mentioned earlier creating an appellate appendix. That's a great example. I actually use these tools to make my own appellate appendices now. To, you know, if a paralegal is jammed up for time, I'll I'll help them out with that tool. So think about things that are repetitive, and you think, hmm, is this something that I could hand off? And then experiment with it, just small little tests. And then the other piece that I would Strongly encourage is make sure that anytime you use one of these tools, these AI tools, that they're adequately connected to a good database. So for example, if you're gonna do legal research and supplement, maybe round it out using Claude or ChatGPT, I would strongly recommend connecting a database like CourtListener or Midpage or some of us have access to Westlaw, Thomson Reuters MCP. It's in the list of connectors. And that way the AI has a has a database to draw from. It's not going to use its general training data. It's not going to mix stuff up or use maybe unreliable information that's available on the web. And if you if you provide it with that that database, it can actually ground its answers in real things. And that's what we're trying to do with Lexponent as well, would be a database access as well as the skills. So those are the yeah. Those are just the two pieces of advice. I would just play with it. You know, Ethan Mollick has a great book about this. I think because we were talking about him earlier, or before we before we go before we got on, but he has a great book called Co-Intelligence. And then I think his latest book is called Coexistence. But he has very good advice for this for the uninitiated, for people who are maybe just starting out to use this technology and figure out what it does. Just use it a lot, experiment. Sometimes it'll do well, you'll say, maybe that was a good use for this. Sometimes it'll do something a little funky. That's what we call the jagged edge, because it's really, really good at some things, even superhuman at you know, certain what you know, recall and synthesization synthesizing. But certain things it'll just get wrong and you can really only learn those boundaries if you if you play with it a lot. So that's what I would say.

Tim Kowal Here's one suggestion that I'll that I'll offer, and it might be controversial, but it is to give your AI a lot of context. Let it know everything about you and let it know everything about your as much about your firm that you are willing to give, and certainly within ethical constraints and privilege and confidentiality restraints and all the rest. Make sure that you're on a paid plan, obviously, and that you have training turned off and the like. But the more context you give it, the better it's going to be. So for example, if you were to just go into your AI of choice and say, tell me everything you know about me, and it all it can return is you are an entity logged into this session on Friday afternoon. You have not given it enough context. But if you have it plugged into so it knows all of your workflows, if you have SOPs, you could just ask it, hey, Take a look at my firm's SOPs and tell me which ones I should turn into skills that could be automated or made better use of through AI. And it'll tell you. I had it just the other day. I was playing with Claude now, has a has a design module. It can create slideshows and images and videos. So I'm I've been developing this kind of AI practice management system on Notion. And I said, you know, give me a teaser video based on what I'm building, as if it was like a product I was gonna launch, but just make it internal. And it put out this very slick two-minute video using the firm's logos and color palettes and everything, because it picks all that stuff up from my from the context. And it produced an amazingly slick short promo video. And I feel very bad for all my graphic design friends because I don't know what is left to do in the graphic design world now that AI has entered the has entered the chat.

Josie Petrick Yeah, that's a really good suggestion. Providing a lot of context, I think goes hand in hand with giving it database access. You want to you want it to understand the goal. Like, you know, you say, like, what is my motivation? You're an actor of it. Like it needs to know like what is our motivation? What is our goal? What are we trying to achieve? What's and the more context you give, like, yeah, California versus you know, federal. You know I'm an appellate attorney, I'm a I'm a trial lawyer. All of that I think is really, really good advice. I actually the other piece that your comment reminded me is spring for the better models. Don't just use the like free consumer grade ones. I do think performance it's fine to you know to check it out, but you know I'll use my dad as an example. He's in his mid 70s and is still practicing. He's a math professor. He's a number theorist. And still does research, even though he's emeritus status. And he works on problem, you know, sort of open problems that he's identified in number theory and group theory with his working group. And they all had started playing with AI. And he said, Well, I'll ask it questions and sometime, you know, just say mix stuff up. And I said, Okay, well, have you tried kind of like the conversation I had with you, Tim, have you tried Claude? And get the paid plan, okay. And see if it does better. And then he said, yeah, it is better. It's better. And then he tries ChatGPT too. And then and then I got he said, yeah, it's really good. And I and then there was a period where the there was a Claude plan called Fable, as some of us will be familiar. And some of us got free Fable credits as this kind of promotional thing. And I thought, dad, give me one of your open math problems. I'm gonna see if Fable, I'll use my Fable credits to see if we can solve it. And it worked for like eight hours. I had to keep, you know, it would pause because I would run out of like whatever tokens and it would resume. So the whole process took like eight hours or overnight or something. But at the end of it, it produced a memo basically, and it didn't pr purport to solve the problem, but it made a lot of progress. And I could kind of tell, I'm you know, I'm not a math major, but I kind of understand what it was doing in broad strokes because you could see its reasoning. And I gave it to him and a freecy said, Yeah, this looks This looks like nothing. And then you said, wait. That's a good idea. Right. And it was like the nudge they needed to it was pulling from a different domain. And it was like the nudge they needed to actually solve the problem themselves, which they did not using AI, but it's like this thought partner. And now he's part of the Claude for Researchers program and is all in. And it they've solved like a lot of their open problems.

Tim Kowal Because it can take concepts from other domains and it doesn't have any inhibitions. It's not afraid of being wrong or professionally ostracized for offering a dumb idea. And it might offer a lot of dumb ideas, but then it'll offer a pretty insightful idea every now and then. Well, let's for a brief palate cleanse from talking about AI, let's briefly talk about an actual legal case. We've actually talked about the Family Violence Appellate Project versus Superior Court case on the podcast. But because now we have one of the authors of the one of the amicus briefs on the on the prevailing party side in that case, I wanted to get your take on it. Josie, you your you and your firm, the Norton law firm, wrote on behalf of 14 law professors, including Erwin Chemerinsky and Judith Resnik. In support of the petitioners and the ultimately successful outcome. And this was one of the at least in our domain, I think one of the most consequential decisions in as a procedural matter. In that in that case for our listeners who don't remember, the court held unanimously that when a litigant cannot afford a private court reporter, the trial court has to provide a verbatim record by electronic recording if no reporter is available. And the significance of that is that there was and still remains on the books the statute is it was six nine five four or something that prohibits trial courts from recording or using a recording for purposes of creating the official record. And how did you spot that opportunity, Josie? Did that come across your Supreme Court tracker?

Josie Petrick Remember if it did. I know that, yeah, I think it did. And I actually the opportunity came my way through Michael von Loewenfeldt, California at Calg, he's an appellate attorney, and he introduced me to Aaron Smith, the co-founder and former CEO of the Family Violence Appellate Project, who was the petitioner in that case. And I had authored amicus briefs on behalf of law professors before in the US Supreme Court a couple times, and I really love working with law professors, so I jumped with the opportunity. And Professor Resnik was sort of the lead on driving the themes behind that brief. But like you said, what appealed to me was you know, obviously for appellate attorneys, it's huge. And for an from an access to justice perspective, it's huge. It's a very practical access to justice crisis and it had a deep institutional history. I mean, so our where the court ultimately landed, as I know you've discussed before in previous episodes, was a pretty narrow. Ruling, although it gave I think the petitioners pretty much everything they had asked for, which was based on the in forma pauperis doctrine. Like if a litigant can't afford a private court reporter and the court is not providing one, they need to be able to have a means of getting a r a qu a reporter's transcript for the appeal. It's it's indispensable for an appeal. And we took in our brief we took a position was consistent with that and how the cool rule but the court ruled, but we We sort of anchored them a little higher. We said, look, all courts of record must provide verbatim accounts of what happened of what transpired in court. And this goes back to Blackstone, goes back to the common law of England, it's the law in pretty much every state that I'm I'm aware of. It's in the California Constitution, the courts of rec about courts of record. And so we said, look, it shouldn't matter what you know, whether The litigant is of means or not. You should be able courts should be allowed to require litigants of means to defray those costs, just like a filing fee. That's not an impediment to accessing justice if they can afford it. But we thought everybody should be entitled to a verbatim transcript. But the court, and I think that was sort of an anchoring argument. We didn't expect the court to take it, although we do believe that that's true. And I if the pro another case comes around raising that, I think we would probably Probably weigh in again. But the court took a narrow rule adopted a narrow approach and I think it was a good one and a very, very good outcome for access to justice.

Tim Kowal Yeah, it was well it was narrow in the way that you mention, but it's in another way it's broader in the sense that that the constitutional argument means that everyone gets access to a core to a to an electronic recording in all of a particular class of cases that involve you know important rights or constitutional concerns, but not necessarily like every breach of contract case, for example. But under the in forma pauperis doctrine, it would apply to all cases. You know, even, you know, no matter how mundane, no matter how constitutionally trivial it is, you would be entitled as long as you can establish that you cannot afford a court reporter. And that one is not available. And I wonder if you so that's interesting. You mentioned that that issue you know might still come up again, and you'd be you'd be you suggested you'd be raring for that fight. If it were to come up again.

Josie Petrick Know if it would come up again because I think the in forma pauperis doctrine really just might take care of it and moot other challenges. I suppose I c I suppose I could envision an edge case where a litigant of means was for some reason unable to obtain a court reporter, even though they tried and were denied. You know, I think it's it's certainly possible, but we'll we'll have to cross that bridge when we get to it. But as you said, it is a broader ruling in some ways, and then that's the sense in which I think it is actually consistent with the arguments we made in our brief, even though of cour of course the court didn't necessarily adopt our courts of record reasoning. But our position was this is a duty that courts have. It's based on, you know, the common law and the California constitution and the very nature of what it means to be a court of record, which includes all superior courts, appellate courts, not what used to be the municipal courts necessarily. So But you know, they don't have to transcribe everything. But it from our perspective it shouldn't matter what kind of case it is. It matters what the court is doing and what the litigants need to be able to do to access justice.

Tim Kowal My recollection from researching this years ago was that at one time there was in federal courts a First Amendment argument asserted, a First Amendment right to have a recording to be able to record trial court proceedings. And my recollection is that there were two the and that that claim was denied, and my recollection is that there were two grounds for denying it. First was that, well, you can get a reporter's transcript, so you don't need to make your own recording. And And that may be moot or that that that ground may no longer apply if there is a nationwide shortage of court reporters and you're in you're in a state like California where it's it's pretty much the only game in town, be before the FVAP case. And then the other ground was, well, you can you can attend and listen to the proceedings, so you don't need to make a recording. That's That seems that seems a little bit cold comfort because if you wanted to re to also find out what's going on in the department down the hall, you can't be in two places at once.

Josie Petrick Right. Yeah, that's a very interesting line of argumentation. And you know, I think the court avoided having to weigh in on that kind of that kind of creative and interesting argument through the constitutional avoidance. And I don't think the First Amendment argument was raised here, but yes, definitely interesting, especially from a right to participate perspective.

Tim Kowal The FVAP case is already you know, it's it's already on the move in the sense that the third district picked it up in a case involving an appellate bond where the where the appellant decided, well, you know, this bond would cost me over $20 million and that would that would wipe me out. The most I can afford with my insurance is a bond of a little over a million dollars. So this wasn't even a literally indigent or not technically indigent from the standpoint of this person could not get could not get a cost waiver, put it that way. But you know this person was a was a low millionaire, just not a not an not an eight-figure mil millionaire. And but that's a so that was enough in fr in relation to the to the cost involved, that person was indigent. And so I wondered if you If you think like I do that the that that doctrine might still have room to travel, it might maybe it might be applied to still further situations.

Josie Petrick I do. I think it certainly opened the door, as the Guzman case showed, the opened the door to making arguments about ability to pay that's relative to the actual costs. And I think that makes a lot of sense. I mean, litigation's expensive, court reporters are very expensive. You know, just because the judicial council has set certain thresholds for indigen indigency for purposes of filing fees doesn't mean that Cost of litigation and will be just astronomical to litigants of means that exceed that threshold. So yeah, absolutely. That's a yeah, please.

Jeff Lewis I was just saying on a practical level, I'm starting to see in the trial courts where before you know trial judges were hesitant to even push that button. I'm not even sure they knew where the button was in their courtroom to hit record. And now I'm seeing trial courts. I was on a law motion calendar today where the judge, before he called the first matter, he said, you know, my finger's ready to press this button if anybody wants it. You don't have to meet the poverty standards of a of fee waiver. Just tell me if you want me to press the button and I'll press the button. I think we're gonna start seeing a dramatic effect from trial courts feeling more willing to press the button and any lack of any negative consequence to a judge for pressing the button. So appellate lawyers' lives are gonna get a lot easier, I think.

Josie Petrick That's a great data point, Jeff. Good to hear.

Tim Kowal Okay, and next I wanted to I wanted to riff with you on the standard line back to AI. Okay, the standard line that AI should never replace attorney judgment. And it's one of those statements that's like a truism, like how could you disagree? But it doesn't really help you much. Like what because the word judgment is doing all the work, or attorney discretion or whatever it is. Where is that line? Where something requires attorney judgment. And so I was thinking about this you know, after we had a convers Jeff and I had a conversation with John Stobart touching on the same the same vein a couple of weeks ago. And I was thinking, in preparation for our discussion, I wanted to ask you how you define it, and I want to offer you my line of thought, the tentative line of thought. Okay, but I'll express it in terms of a of a of a rule just for sake of making it clear. Okay, if an attorney or if something needs to be filled out on a court form, then it could be filled out by AI. Okay, but because a form is a form is what the court or the judicial council creates when it says we just need to process the information. Okay, the a form basically has the has an implied premise that no judgment necessary, no attorney discretion necessary, just give us the information. Okay, so you should check to make sure that the that the data is correct. But you know, but other than that, that doesn't require attorney judgment, does it?

Josie Petrick Yeah, that's a great question. And the form the form the form hypothetical is really interesting too because just yesterday I was filling out mediation screening questionnaire and a civil case information statement for a California appeal. And those are actually quite substantive. I actually wouldn't just delegate the I mean you could use AI to help to help shape and you know, maybe streamline your responses or think of think of things that you hadn't thought of, but I mean The civil case information inform you want to check the right box you want to cite the right provision for which ccp provision you're invoking for the appeal you want to make sure that you're correctly advising the court when the notice of appeal when notice of entry was served when the notice of appeal was you know when all of that you want to make sure a hundred percent accurate so yeah actually a great use for ai to make sure it's accurate but I wouldn't actually delegate that wholesale although Other forms certainly can be. Those are probably the exception that proves the rule. But I mean, this may be surprising for somebody who is so steeped in AI, but I do actually believe that law, lawyering, and judging are fundamentally human endeavors, and we should not be delegating our judgment to machines, just like we shouldn't be delegating our judgment to Martians. I mean, the law expresses and continue continuously refines our social compact, how we want to behave. How we treat each other, the consequences when we fall short. And those are normative choices. And they're bound up with democratic legitimacy and institutional accountability and human values. So yeah, I don't think we should casually outsource them to algorithms. And I wouldn't, you know, just like I wouldn't casually outsource them to Martians to tell them how we should run our affairs. But to your point, where do we draw the line as a practical matter? Typically we aren't, you know, we aren't delegating the practice of law or Or judging or for the most part, to AI. You know, will someday could these tools become even better at lawyers and judges at getting at the right decision based on some empirical some empirical standard? Maybe, but you know, as in Silo the show Silo, that day is not today. So you know, like yeah, I think like I said, you have to play with it and see You know, but there is see what it's good at and see where it fails and just be very, very cautious. That I don't know if that's a more amorphous answer than you're hoping for. I mean, I guess I could go talk task by task, but I think you know, like we were saying earlier, the status quo sh should not be the gold standard. Cases can take years. The cost of meaningful legal access excludes many people. And the goal is to equip responsible humans with better instruments, you know, not necessarily preserve needless inefficiency. If so if there's something we can do as lawyers or judges and do it more quickly and more effectively, more accurately with these tools, then I think we owe it to our clients and to ourselves, frankly, to society to use those tools as long as those criteria are met. And if it's introducing the potential for error, if it's if it's actually outsourcing our judgment and our normative values, then I think that's a place where I would definitely wanna draw the line.

Tim Kowal Yeah. Well, it does become such a difficult question. Like where attorney judgment actually lives, and is it different from just fact checking? Like, for example, like the back to my form hypothetical. The easy example would be a proof of service form. Okay, like the attorneys already never look at the proof of service. There's they leave it to their staff to take care of. And an attorney, I would wager, Never lays eyes on a on a proof of service unless there is some issue of service and they're looking at the other side's service. And then another example might be you know, you know, that that no appellate justices are listening, but the like applications for extension. Okay. I'm filling out paragraph nine. Okay, I'm filling out paragraph nine, the good cause. The rest of it, though, I'm counting on my paralegal to make sure that.

Josie Petrick That's a great example. Yeah.

Tim Kowal That she's correctly, you know, identifying the number of volumes of appendix and pay number of pages in the reporter's transcript and how many previous extensions have been sought and days granted. Frankly I'm not looking at it.

Jeff Lewis Tim, I got news for you. The appellate justices are likely using AI to review your extension request for everything except paragraph nine, just to make sure you've gotten everything else right. So don't lose any sleep.

Tim Kowal That's what I mean. Yeah, if AI is going to be processing the information, then AI should be able to produce the information.

Josie Petrick Yeah, I think that's a really good really good example of where I think a skill or a or a form could be very much in process could be very much enhanced by using AI. As long as you give it all the proper context. Give it the actual rule, right? That the good cause factors. Here is the text of the rule. Give it access to your calendar, you know, what your recent matters have been, what your upcoming matters are. Yeah, as long as that's it's code compliant and the information is accurate, I think that's a That's a fantastic task to outsource to AI.

Tim Kowal Do you have opinions on the on the judiciary's use of AI? I believe that some superior courts are reportedly piloting AI tools for research and tentative rulings. I wonder if you have opinions. Should litigants be told whether and when AI tools are being used in their cases? Do they do people have a have a right to know, or is that not an issue?

Josie Petrick Hmm. Yeah, I think it's definitely an interesting pilot. I think, yeah, LA Superior Court, last I checked, was using a bench memo product called Learned Hand, which I think a Clio acquired. And then it some courts are not using it all, like I think the first district is not using it all, but whether somebody has a right to know, I think maybe generally, but you know, I think it kind of raises a question of how much is the writer's process, the judge or litigator. How much is our process protected by sort of confidentiality rules? You know, like you know, I used it to brainstorm and I asked it these questions and it came back with these answers. I think that's going to be privileged or in the context of the bench, you know, subject to sort of judicial confidentiality rules. But I think general disclosures are appropriate. You know, and as far as the judiciary's use of AI tools, I think I think it's a good experiment as long as You know, and I would basically apply the same measure that we've just been talking about that I was proposing for attorneys' use of it. You know, we're not deleg if as long as the judges aren't delegating their own judgment and they're using it to serve up justice better, more efficiently and maybe more accurately, surfacing information more quickly than they could have otherwise been able to do. I think it's great. You know, we've seen a couple of instances where there was not good use of it, where I think Some judges who unfortunately made the news had probably outsourced too much and not checked the output and there were fabricated name names and cases and that that's it. But it's it's a good example of what not to do and why we have a long way to go to educate ourselves about proper use of AI and using professional grade tools, not necessarily, you know, I don't know what tools were used in that case. I think that's that that wasn't revealed, but or in it's been more than one case, but But it's important to use the tools that have high degrees of accuracy. They use multiple rounds of verification and can be audited. I think those are probably good guidelines.

Jeff Lewis Yeah. It's all about the marketing in terms of building confidence of how reliable these tools are at ClioCon this weekend, or this earlier this week, they announced something called judiciary grade artificial intelligence and this Learned Hand product. And Pablo, the former guy behind Casetext, one of the couple of co-founders, is spearheading the judiciary product of Learned Hand used in the LA Superior Court and I think all these sanction stories have scared a lot of lawyers and judges off of AI. And so with this branding of judiciary grade AI, I have no idea what that is. They're trying to restore some measure of confidence in the products that judges aren't gonna screw people over if they're using AI. It's interesting to see where it's going from the lawyer space to the judiciary space.

Josie Petrick Yeah, I mean I think bench memos, yeah, I love that love that note about branding. I think that's that's absolutely where the market's gonna go. And I think, you know, anything that I you know encourage someone to build would be lots of verification. So lots of verification under the hood, lots of rounds of checks and balances. And but also, you know, if you're citing source material, like if you're if you're generating a bench memo or on the loy on the litation side, like you know, some other kind of work product. Any factual verification should be, you know, you should be able to click on a hyperlink and jump directly to the source. Right. And same with the legal verification, jump directly to the source so the reviewer can check it themselves. Don't trust, verify. Yeah.

Jeff Lewis Yeah. The software that's gonna have the best chance of succeeding is gonna have two qualities. One, it's gotta be passive checking, like spell check puts a little squiggle underneath a misspelled name. You'll get a squiggle underneath a case that has a questionable existence or holding. And frictionless cite checking of facts and law, you know, whether it's Clearbrief or a Westlaw product that splits the screen or with hyperlinks, you can quickly flip from brief to the source. That's going to be the key to these pieces of software. And I think in the year to come, we're going to see a lot of software differentiate itself in the high friction software that requires you to like upload a brief and wait three hours and then you get this product that is in 10 parts and doesn't have hyperlinks. Those pieces of software are not going to survive. And the Clearbriefs of the world and the BriefCatches of the world and RealityChecks of the world are going to do really well.

Josie Petrick Yeah, I definitely think ease of use is gonna be the name of the game. And working the way attorneys and judges already work, like a word add in, you know, don't make them go to another whole other website or learn a new way of working. You know, like does it integrate into our existing tools, our existing workflows, and does it make it as easy and seamless as possible? Does all at the end of the day, does it reduce our stress? 'Cause litigation can be so stressful, even for appellate lawyers. So Jeff, were you gonna say something?

Jeff Lewis I was gonna push back. Appell we write appeals on a beach in our pajamas. It's not stressful.

Josie Petrick Sh don't don't tell them.

Tim Kowal Well, we had so many t other topics we wanted to get to, but we're we're already at an hour. Let me just ask you in closing, Josie, you know, we covered, you know, how it started and how it's going you know, as of Q4 2026. Can you give us any predictions for where you think we will be this time next year in 2027?

Josie Petrick My gosh, Tim, it's the market is moving so fast and things are evolving so quickly. And I really do think that the tools are becoming very and I assume your question was about AI since that's what we were just talking about. But yeah, the tools are become becoming a lot more reliable than they were. So I think in 2027, I think frankly, probably could have put could have these tools, the profession professional great AI tools generate A brief draft or a motion or an opposition. You know, I don't think it's necessarily quite there. I think it would take a lot of tweaking, but I think by this time in 2027, for firms, for attorneys that are comfortable and for judges who are comfortable, they probably could have it write the first draft and then review it. The attorney or judge would then review it the way an attorney y you would review the work of a law clerk or a junior associate. And you still need to get up to speed. Still need to, you know, you can't delegate understanding, but I think probably it's gonna be there. But more broadly, I think the tools are just gonna become a lot more familiar and a lot more grounded in the in existing workflows. So agents are gonna get better, MCPs are gonna get more, and again, that's a model context protocol, kind of like the APIs for AI agents, but you're gonna have more databases connected and you're and I think attorneys and judges are gonna realize like. This actually really does enhance my existing workflows. I have a bigger picture, it's easier. This doesn't feel like this cluey weird technology that only my nerdy friends like Josie know about. So yeah, th those are my predictions. I'd be curious to see here if you if you too have any.

Tim Kowal So are you saying that Jeff and I are not gonna be able to fill up entire episodes talking about AI sanction cases?

Josie Petrick I certainly hope not. I did see a drop in Damien Charlotin. Damien Charlotin keeps a tracker of hallucinations, as you guys I'm sure you know. And I did notice that there was a precipitous drop, at least in I think the first half of 2026, but that might have been I might be getting that wrong, but in the there was recently a precipitous drop. I won't say exactly when the range was, but that might have just been because the quarter had just started. But I choose to be an optimist.

Jeff Lewis Does he use Notion to track those cases?

Josie Petrick Actually ask it.

Tim Kowal That's right. All right, Josie. That that was such a fun conversation. We are we're going to have to have you back, because there's so much more that that we didn't get to, and there's so much more that's that's coming, both on both on AI and in the Supreme Court. So you're gonna have to be part of our you know Supreme Court Roundup so that we gather what's what's been going on in the in the California Supreme Court and what's coming down the pike since you are the proprietor of the civil case Supreme Court tracker who can help us distill those trends. So thank you again Josie that's gonna wrap up this episode. If you have suggestions for future episodes, topics and guests, please email us at info@calpodcast.com. In our upcoming episodes, we're gonna get a full download from ClioCon on from Jeff and look for more tips on how to lay the groundwork for an appeal when preparing for trial.

Jeff Lewis See you next time.

Josie Petrick Thank you so much. Such a pleasure.