Maximum Lawyer

Watch the YouTube version of this episode HERE

AI is rapidly reshaping the legal industry but where does it actually help lawyers, and where does it fall apart? 

In this episode of Maximum Lawyer Live, Tyson Mutrux breaks down a thought-provoking article shared by Alex Su about how lawyers are using AI tools like Claude and ChatGPT in real legal practice.

Tyson walks through real examples of where AI excels like client communication, summarizing information, and reviewing documents and where it still struggles, including due diligence, negotiation strategy, and complex legal drafting. He also highlights a critical shift many lawyers are missing: moving from chat-based AI usage to process-driven automation inside firm workflows.

If you’re a lawyer experimenting with AI or trying to integrate it into your firm more effectively, this episode offers practical insights, cautionary examples, and a smarter way to think about AI implementation.


Timestamps

 01:30 – The two stories that show AI’s strengths and failures
03:40 – Comparing AI tools: Claude vs ChatGPT
05:00 – Using AI for client communications and simplifying legal language
07:15 – AI for reminders, prioritization, and managing overwhelming inboxes
19:00 – When AI drafting goes wrong in complex agreements
22:00 – The danger of generic AI answers in legal research
35:30 – Legal nuance AI often misses (release language example)
39:30 – The biggest takeaway: build process-driven AI workflows
 42:30 – Practical automation example: AI pulling weather data for cases


Tune in to today’s episode and checkout the full show notes here. 


Resources:

Creators and Guests

Host
Tyson Mutrux
Tyson is the founder of Mutrux Firm Injury Lawyers and the co-founder of Maximum Lawyer.

What is Maximum Lawyer?

Maximum Lawyer is the podcast for law firm owners who want to scale with intention and build a business that works for their life.

Hosted by Tyson Mutrux, each weekly episode features candid conversations with law firm owners, business experts, and industry leaders sharing real strategies and lessons learned in the trenches.

If you're ready to grow your firm with less stress and more support, this is your next must listen. Subscribe today.

Tyson Mutrux 00:00:01 Welcome back to maximum lawyer Live I'm Tyson Matrix. Today I'm going to be talking about an article that I found on ex Alex Sue. If you follow him he's on all the different, social media social media platforms. He shared an article written by someone I don't know their actual real name. It's at seller's council. So seller's council, and apparently so in their bio they're corporate M&A lawyers. So they buy and sell and buy and sell side. He's also a fractional GC. So it's an interesting article because it says how I actually use AI in my law practice and where it falls apart. I would have never given this any thought if Alex had not shared it. Alex, a very smart guy. Someone who, does really good content, shares really good things. So I went through this and I figured I'd kind of give my thoughts on some of the things that are in this. I think I agree with some things, I disagree with other things, but that was kind of an interesting thing. So I'm gonna mix it up a little bit today from what I've been doing, I but I this was just an interesting one.

Tyson Mutrux 00:01:16 It was. So I usually talk about things that are top of my mind. This one's kind of top of my mind right now and that's what I want to talk about. But all right. So I'm going to read part of this just to give you some context. Okay. I'm not going to read the whole thing because it's it's long. I'm going to give you, bits and pieces, but I'm going to read part of the beginning. Okay. So on a recent side by side transaction, I use Claude to help explain working capital. True, up to an inexperienced sell side council. What the other side had deemed an impossible issue was made simple and easy to understand. In just a few minutes, I was able to describe both the theoretical and practical aspects of a working capital. True, up in a fraction of the time it would usually take all thanks to AI. We closed shortly after the explanation. Okay, that's part of it. Next part. That same month, I asked Claude to review an agreement as part of due diligence for a transaction.

Tyson Mutrux 00:02:17 He told me assignment was freely permitted. It was plainly permitted. Prohibited. When I was pushed back. When I pushed back, it doubled down. I had to paste the exact contract language into the chat before it acknowledged the mistake, at which point it told me it hadn't actually reviewed the clause, but had assumed assignment was permitted based on the type of agreement it was. Both of these stories are true, and together they're my honest answer to every lawyer who asks what I really think about AI. It depends. Okay, I can relate wholeheartedly to to this I many of us probably have that are more, I would say the at least moderate to power users of of AI. I've had this experience with ChatGPT. What was kind of surprised me is that I've never had an experience like this with Claude. but in fairness, I've used over the years now, significantly, I've used ChatGPT significantly more, but as of late, I've certainly shifted to Claude. so it's, interesting. I've already I've already given the background, so I'm gonna skip over that part of it.

Tyson Mutrux 00:03:30 My experience with AI, he's got experience with AI. All right, so here's where, like, the heart of the conversation, the part I wanted to talk about here, things AI is good at. and he says one note. Over time, I have found that the best luck with Claude in my legal practice. So that's the model I will refer to. I still use ChatGPT for some things, but I generally prefer Claude's outputs and UX. So user interface. I agree with that. that is that is kind of why I've shifted to Claude is I still use ChatGPT for some things. but the most of my stuff is being done in Claude. All right. So client communications, he talks about how client communications is good at summarizing things when it comes to client communication, synthesizing the information. It really can. Condensing it down. If I'm in a hurry, I will tell you, I will actually sometimes take a big block of an email that a client who sent me, and I will just have it summarized and like, give me the bullet points.

Tyson Mutrux 00:04:33 I'm not. I don't have the bandwidth today. Just give it give it to me straight kind of a thing. And it does does a pretty good job. And so he talks about that, he says I can write out a dense email with lots of complex information, and Claude can make it easier for the client to understand and highlight action items in a fraction of the time. I've done something like that too, where I'm like, okay, I know I'm using a lot of legal jargon here. Make this, make this so that the layperson can understand it. I, I've used it to that extent, too. All right. The second part of this reminding and prioritizing, I have a busy life and an inbox that looks like a war zone. And he's an assistant. That sounds like. Or better filters. as many systems and procedures as I try to put in place, something is always at risk of falling through the cracks. Claude goes through my list every single day, and reminds me of things that I may need to follow up on, or that I committed to and have not yet delivered on.

Tyson Mutrux 00:05:27 This means greater client satisfaction and less stress for me, which makes me a better lawyer. Okay, so I agree with this approach. I use an I actually have an assistant that does this, so I don't use it for that. I have tested out that that is something I have tested out, both with Claude Bot which is now multiply, which was now something else. Open bot or something like that. It's changed multiple times. So, I don't use it anymore. I've disconnected it, so I don't. It doesn't matter, but I can see how you can do that easily with Claude. coworker on your desktop. You could easily do that. That's that's something like easily done perplexities product could do that for you as well. Their email product, that's something you could easily do. So I, I like this idea. I think he's on the right track. Great. Perfect. Good stuff so far. Technical reviews I work at a small firm. I don't have a paralegal. Oftentimes my eyes are the only ones on a document before it goes to a client.

Tyson Mutrux 00:06:31 Before AI, that meant the occasional typo or glitch would sneak through that. through that a second set of eyes would have caught. That's a tough sentence to read. Now, AI performs a technical review on every document before I send it. It's like having a dedicated associate to double check my work, but there's no additional cost to the client when all around. Okay, good. Great. I think that that's completely fine. Issues lists. Oftentimes during the course of a deal, you'll go through 4 to 5 versions of the purchase agreement before arriving at a few key issues that you can just, that you just can't get past. At this point, I usually find it helpful to step away from the contract and work from a document that identifies each issue, each party's position, and proposes potential resolutions. That document is incredibly valuable, but can take a couple of hours to put together. Claude can do it very well in just a few minutes. Okay. Love this. I. I guess I'm gonna kind of get I'm gonna tease sort of my bigger issue with some of the things that he's doing.

Tyson Mutrux 00:07:44 he's using it the way that he's using AI, the way that it's we've sort of been trained to to use it over the last few years and that's you enter a chat, you get a response, you enter a chat. So there are a lot of one offs. And this is you have lots of delegation to the AI. Great. Okay, good. However he's not I mean none of these are like workflow driven things. These are all just one offs. He's still the one pulling the triggers on all these things. He's still the one doing all the work. One of the like one of the things and this is specific to what he just talked about. We have an AI built in to our case management system where it will it can analyze the entire file and in real time, give us the issues where we're not having. We can just look at it. You don't have to actually like click a button or say, hey, can you do this and plug it all into the AI Now he's using an agreement.

Tyson Mutrux 00:08:47 So I understand a document. This might be a little bit different, but in the matter that he has, he can still identify these things probably pretty easily where you just can when the information is in there, you can have the AI assess it for, hey, here's some red flags you need to look out for. you're missing these things in your file. Well, let's let's, let's fix those things. so and that's more process driven instead of one off driven. That's a big difference. Significant difference. And so that's kind of my biggest bone to pick with him on this. none of the stuff so far they talk about is wrong. It's just the way he's using it I'd say is a little bit off. I would make it more process driven as opposed to, you know, him doing it, driven by him. Okay. let's get through to the next one. Starting language for bespoke provisions. Okay, I almost every deal. In almost every deal, there will be some provisions that don't fit your form and require drafting from scratch.

Tyson Mutrux 00:09:57 For simple ones, that's easy, but for complex ones the task can be challenging. Oftentimes the hardest part is just getting started. Once you have a block of text to work with, it becomes easier to see where things need to be added and removed. I can ask Claude to generate a distribution waterfall for an LLC that allocates the two different members according to certain percentages based on the revenue source, and it'll give me something to start with very quickly. Then it can serve as a sounding board as I work through revisions until I arrive at language, yada yada yada. So what's weird to me is that, why wouldn't you just start with doc? Jen okay, doc, so you have your template and then you work from the template, use the AI to work from the template that that was kind of this one's kind of bizarre to me that you would just use the AI to create it from scratch. I that part kind of threw me off. I don't know, maybe I just something I don't understand about that practice area.

Tyson Mutrux 00:10:52 that would just seem odd. It's like, just use a template. So, anyways, in the next one's kind of similar modifying basic forms and M&A deal. There are typically several ancillary documents that are not highly negotiated. So, basically he talks about how basically what I was just talking about taking those forms and just making it fit to the client. I don't know why he wouldn't just do that. number five, which the, the one I just talked about. So. All right. So let's get to the things that he talks about that AI is not good at, okay. Drafting from scratch or drafting complex agreements, which kind of flies in the face of number five. I don't understand it. Number five, just I got to talk to this guy because it just doesn't make any sense. But, so it's not yet to the point that you can hand it an Loi, a letter of intent, and trust it to create a usable first draft of an asset purchase agreement. It may give you a draft that is overly basic in a key terms, one that is overly complex and overkill for your deal, or one that is outright favorable to the other side.

Tyson Mutrux 00:12:03 It's just not there yet. Even when I provide a good from good form to start with, it still struggles. Okay, I agree, this is where okay, so the, some that I've noticed with ChatGPT specifically is that it is speaking in very generic terms. So I'll ask it a question. I'll give you an example what I'm talking about. And then I'll give you an example that some that I got from a client recently where the if I was I can't remember what the question was, but I asked it a question and instead of giving me very specific as there was a very specific answer to it, by the way, it gave me very broad answers like, and so let me kind of let me make up an example so you know what I'm talking about. So if I, if, let's say I want to know what and this is, this wasn't what it was, but let's say I want to know what the specific statute is, about, you know, bad faith laws in Missouri, like, so something like that.

Tyson Mutrux 00:13:13 that's kind of broad, but mine was more specific. But, it it like the answer would be something lying along the line of in Missouri. insurance companies cannot commit bad faith and yada, yada. Like very like. Yeah, no. No joke. I want something more specific. I want the actual statute. Right. And so it'll do these things where even though it can go the next step, You have to prompt it like 2 or 3 times to get to the freaking like the specifics where I feel like Claude's not that way anymore. And I got a, I got an email from a client who had gotten bad information from somewhere, and I'm. I'm suspecting it was ChatGPT that, about the valuation of his settlement. And this was a federal tort claim. So there are very specific requirements when it comes to a federal tour claim. And so his estimate of the valuation of the claim was way higher than what it legally actually could have been. Okay. It legally couldn't have been nearly as high as what he was claiming it was, but you could tell by the language in the email that it was something that he had gotten from AI, most likely ChatGPT it was because he was speaking in very generic terms, and I just responded with the very specific terms about why that was incorrect.

Tyson Mutrux 00:14:32 And, in a settling the case because of that. So it's creating issues on both. For us as attorneys and anything that we're looking at or trying to draft because you can't use the I agree with him. You can't use it to draft it. I would never use it for like an agreement from scratch. I just wouldn't I would take it. At the very least. I would say, here's a form I've used in the past. Make this apply to this situation, which I have done where like it's a very boilerplate document and I say, hey, I this is for another case, make it apply to this case. And I've done that before. And it can do that because it's not changing anything other than like names and a couple other things, but otherwise it's pretty easy. All right, let's get to the next one. Understanding sides of a deal or practical impact. So practical impact is where the big issue right I think we all know that. Like that's where like you are. Just made me think about another client.

Tyson Mutrux 00:15:35 Client email I got, but, they're there. They're like, there's the law. Like in law school. This is like the, I feel like it's teaching us in, like, the AI is like repeating things that maybe we heard in law school but don't necessarily in practice, apply much to things like, okay, Erie Doctrine, big deal and civil procedure. Right. How often do you actually think about the Erie Doctrine? Okay. So it's from a practical standpoint, it's regurgitating things that maybe legal principles, but not with the perspective of your jurisdiction, that you may be in the state that you're in your practice area. It doesn't take things like that into consideration. But he says, I frequently see AI recommend provisions as being favorable to one side, when in fact the opposite is true. It once recommended that a buyer add a confession of judgment clause and a sellers promissory note. One of the worst mistakes a buyer can make in order to fast track future disputes. I'm not sure why the why this is, but it's a constant, consistent error that I see in almost every deal.

Tyson Mutrux 00:16:46 It also tends to hyper fixate on issues that aren't actually important in the real world. Email not being allowed as a method of providing notice. Maybe a little outdated? Sure, but that's not going to impact the deal at all. Yes, that was a real issue that Claude flagged as a major concern. So, I mean, I already kind of talked about this a little bit, but yeah, I agree with that. Negotiating. So, this is the next thing. I think it's insane to try to use AI to negotiate. I don't. I would. At this point, I'm not comfortable at all using AI to help me negotiate on something, but I'll talk about what he says. He says negotiating this one is hard for AI once it wants everything to be fair. And he talks about how it just it doesn't work. Struggles identifying and I and prioritizing where he talks about how AI wants everything to be fair, which I agree with that to where it's it's it's not going to want to be lopsided in your favor.

Tyson Mutrux 00:17:50 That's why you have to you have to prompt it. If you do like I, I have used it in situations where I want something written that's favorable to me. I want it to be take my side, take my position. That's what I wanted to do, and I wanted to help me write that. So you had to prompt it to do that. Otherwise the default is fairness. That's what the the default is generally going to be is fairness. So you have to prompt it if you want something different. even if you say I represent this person, it can. And you because I've had it Do help me with some some deposition questions before because I want to get some new ideas for questions. And it was like it didn't quite understand the fact that, hey, like this, that's the defendant. Like, I need more, I need some more biased language from you. So it's one of those things where you have to prompt it a little bit more to get what you're looking for. All right.

Tyson Mutrux 00:18:48 Due diligence. I know I'll take some peat for this one, is what he says. okay. There's nothing to it contradicts what he said above because he used it for due diligence. But I'll just kind of I'll. I'll take a step back. I'll ignore that for a second. I know I'll take some heat for this one, but AI is not a trustworthy tool for due diligence. The reason is simple even though AI can work through contracts much faster than most lawyers, sometimes it just makes it up. As I mentioned above, I have had Claude read a contract with plainly written anti assignment clause and report that assignment is freely permitted by the contract. This wasn't a one off experience. It randomly decides to pull things out of thin air and ignore the information it reviews. And from what I've read, this is an inherent limitation, not a bug anyone expects to fully solve. I'm not sure I will ever rely on AI for any meaningful due diligence. So, so kind of like talking about generally about this a bit.

Tyson Mutrux 00:19:49 so I think it's good at some basic stuff when it comes to due to due diligence. I don't think it's not as bad as the nuance. Right. It's really bad at nuance. And some of the things he had even mentioned where like, no, that's clearly something that, should not be in there. Like there are certain things, like whenever it comes, like a settlement release, that it probably would not know that we needed it to say something different, where, for example, I bet if I asked it to sign to generate a release for an injury case, I bet it would give me a general release. And I'm guessing that it if I didn't prompt it. Otherwise it would just create a general release. Well, then if my client had claims against multiple people, at least in the state of Missouri, my client's screwed. My client can't go against. So let's say there's three defendants defense one, two, and three. If we settle defendant one, they sign a general release defendants two and three, even though they're not even listed in the release, not even mentioned at all.

Tyson Mutrux 00:21:00 In any capacity. A case is gone. It's over with defendant. And you could. That's a really big mistake. If the if defendants 2 or 3 are or are bigger pockets. So that's the problem. so the nuance it's not really good. So if you're having it do due diligence on some sort of a release or something like that, then? Not good. If you just. I wouldn't trust it. we've tested it out a little bit. I've had similar situations where I'm having it where I've. I actually created a GPT in ChatGPT, and I said I fed it with all the information they needed. Okay. all the things we do, all the workflows. And it was missing. Things were like I would say, hey, like I did a review this identify any issues, and it wasn't identifying some of the major issues that would have been issues. And so it's that's the problem. But I think the bigger, I think that the, the bigger thing is what I talked about is where a lot of the ways he's using it, it's very it's driven by him and it's not not not to say it's not saving him time.

Tyson Mutrux 00:22:13 I'm sure it's saving him a lot of time, a lot of money. He's probably able to operate at a higher capacity. So it's still good. But if he, if he, he could take some of this stuff and make it more process driven and it sounds like I mean, he may be a true solo to the way he's talking. He doesn't have a paralegal. maybe he has an assistant, but he doesn't have a paralegal. No other associates. So at most he might just be, may have a legal assistant. And so that may be part of the problem. But if he were more process driven, some of this stuff could just be done automatically. Files opened up things or documents are added to the file. Things are created automatically. So my just you know, for people in the association, my advice to you is think about this more as a process. when it comes to AI, try to get out of the of the thinking that this is, you know, you're you're operating from a chat only exclusively.

Tyson Mutrux 00:23:09 You will always be operating from a chat. I think, to some extent, just like we've been using Google for two decades. Two and a half decades. So you're still going to be asking it questions. But when it comes to actually doing the fulfillment. Think of it more as a process. How can I build this into the workflow where I'm not having to really touch it? It's just being done. I'll give you an example. I've been I've used this example a lot, but it's just an easy example as to how to use it. we used to have a person manually go and pull the weather data for a case, which is could be really important for a slip and fall. It could be important for a car crash too, but, for a slip and fall, it can be really, really important, at least when it comes to our laws in the state of Missouri. And so getting that weather data, was usually done by a person manually. Now files opened up, AI read that goes and extracts the information, puts it in the file.

Tyson Mutrux 00:24:11 automatic. That's a process driven thing. That's not something where something's type. Someone's typing it in. Go pull me the weather data. It's done. Okay, so think about AI when it comes to, more of a process driven type of a thing. So. All right, that's all I have. I was going to go through, I got a couple of minutes. So give me let me. I'll just go through if there's any interesting comments on this article because I do want to go through that. okay. I kind of skipped through that one. really thoughtful post matches many of my own experiences. Was hesitant to read this as the 1,000th AI balanced take on AI, but this offers good insight. The client communications bit is surprising. How do you make sure the email doesn't sound like AI slop though? Okay, so he says he doesn't have it right it for him. okay. He, the one person asked which paid version of Claude are you using? And seller's counselor says the pro version.

Tyson Mutrux 00:25:24 Okay, boom. Excellent article. I use Claude in almost every area of my practice to get a first run at an issue or cross examination, and then work with that material and refine it to what I need. I still need to be deeply knowledgeable about a case. Claude just speeds up that process. That brings up a really excellent point. if you use AI too much, you need to find another way to learn your file. That's a really important thing. So just, especially when it comes to litigation. I mean, if you could you imagine walking into a courtroom and you had an AI do everything, and you don't, I mean, understand your file. There's already been a lot of studies on this where, if you have AI create things your your ability to remember the things that was being that were being worked on. it's it's very, very low. So it's an MIT study. So, just be very, very careful. Careful, be careful about that. And then also make sure that things are process driven okay.

Tyson Mutrux 00:26:25 Make sure process driven AI as at every angle that you can and you'll be in shape. So all right that's all I have for this week. Hopefully you enjoyed this one. I know it's a little bit different, but I'll be back to my normal way of doing things next week. Have a great week everybody, and we will be seeing you. make sure you check out pexels.com. And if you're interested in the association go to Maximum lawyer.com. See everybody.