Martin Kraft (3:39)

Well, I think first of all, it's looking really where. First of all, understand actually all those steps in the process, right. A lot of times, let's say if we are actually talking to, let's say the firm owner, they'll say, oh yeah, this is done like step A, step B, step C. And then we talk to the actual person who is doing the job, maybe like the paralegal or the cleric, right. And they're like, oh, actually there's like step D, step E, step F, you know, step Z. Right. So first of all, really like making sure that we map out the full process that we are working with. Ideally have some kind of diagram, I mean, the relative tools, great tools on there, like Figma, et cetera, where you can diagram out the process, make sure that everybody who is involved in the process kind of agrees to that. Right. Because that can also be sometimes an issue if the firm doesn't have proper SOPs. Right. All those SOPs are not necessarily enforced. Process may look different from person to person. So making sure that everybody's aligned to the stakeholders, then understanding how, let's say the data flows. Where do we get information from? Do we call up the client, ask information, do we have it in the case management system? Do we have it in the file? But let's say it's not uploaded to the case management system and then design a workflow really around that, understand what data needs to be entered into the systems so we can then actually make sense some use of it with automation, with AI, whatever we're utilizing. And then after we have that in place, understanding where can we actually apply automation and what processes can be automated versus not right. And the decision there would ultimately come down to let's understand the full capabilities of the software that we are using already. Right. It's a case management system Clio managed like recently they released automated workflows. You can now apply task list on stage changes right in the in the case life cycle. Understanding we have that native functionality built in or do we need to build something extra using a platform like Zapier, Make.com or something more complex than that.

Tyson Mutrix (5:33)

Yeah. I wonder if you have any advice for people that I'll give you. I'll give you a little story and then maybe you can give me some advice on this. Where I, when I, whenever I first started as an attorney, the, the firm I worked for, they had like their own kind of basic templates. They were somewhat uniform, but I know that each of us attorneys had our own set of documents on our own computer that were kind of our own little templates. And like the receptionist had her own and the paralegals had their own. And so they weren't in like this joint database where everyone was using them. We had take. We had kind of like the general documents that kind of everyone sort of had modified to make their own. So when you go into a firm like that where like they don't have any system systems or processes, it's nothing's all integrated. Do you have any advice for people on how to. When they're starting to build out these processes? Like how do you take all these individual templates and all that and try to blend them all into one unified system?

Martin Kraft (6:30)

I think ultimately here the problem is not necessarily a technical problem per se. It's. It's more of just an operations problem. Right. And a human problem. And making sure that again all of that knowledge is centralized. And ultimately I would say the solution here is quite simple. Just make sure that whoever is doing a particular job, they sit together in a room on the Zoom call and walk through exactly how they do a particular function. After we agree on that, again, we can use maybe some diagramming tools. Whiteboard helps. Then take those documents, take that process that we have just outlined and put it in a simple SOP inside of Google Drive or OneDrive or whatever you're using or your server. Ultimately, I think in terms of knowledge management as well, it's kind of a big thing that a lot of programs out there that try to do say like, okay, we have the best knowledge management system. We are also now working on that internally. We've tried a few of those and really just said, okay, we're going to use Google Drive. We are going to have Google Docs with those processes that has version control. We can share it with everybody. Nobody needs extra accounts. And, and actually Gemini connects directly to Google Drive. So you can just say add Drive. What is the process for setting up a new client account and it will search all of your drive and pull up that answer. And those answers were better than the custom platforms that we were paid, you know, $20 per user for. So I would say simplicity and like first, first of all making sure that everybody's on the same page in terms of just conceptually right on an operational level and, and then not overcomplicating and not over engineering it and just put it out on paper and making sure that at least say the human process is well defined.

Tyson Mutrix (8:10)

I think what you just said is, I hope people don't overlook what you just said because that is such a really simple cheap hack. And I think it might be ultimately it might be better than what we've been. Everyone's been trying to find for years where they want to have this unified system where you can find all the documents, all the SOP and everything. But I remember when I started, all of our SOPs were on a Google Doc. It was like one Google Doc. And if I, if we had continued to build up in Google Drive, I mean we could be doing what you're, you're talking about now where like you take, you just go to Gemini, ask questions of it and we'd be probably be better off at this point. So I, I think that was a, that's a really, really good hack.

Martin Kraft (8:53)

The look I over my career, I have been this for over like five, six years now already in developing the solutions for law firms is we've built very complex systems on Zoho like the most like 50 different steps in between the hundred different flows. And what I realized that ultimately those systems are amazing. But you need to know when is the right time for those systems. A controversial opinion until Your firm is 2025 people. Honestly, technology is not your biggest bottleneck. Your biggest bottleneck is getting people on the seats, making sure they can actually deliver the work. And you have good customer service, you have some standardized sops, but really technology is not the biggest bottleneck because if you have three people on the team. If you increase, let's say process efficiency by 20, 30% while that's still not even one full time equivalent, essentially if you hit team of 100 and you increase efficiency 20%, that's like adding 20 more people to your, to your team. Right? So those, those, those let's say benefits really start increasing as a firm grows. Until you hit reach that point, your problem are human problems, not, not technological problems.

Tyson Mutrix (10:09)

Yeah, that's a good, that's a really good point. When it comes to, we, before the, the show we were, we had a little meeting and we were talking about problems, prompting. Will you talk a little bit about prompting and how just prompting effectively can change the output substantially?

Martin Kraft (10:26)

Well, first of all, I think utilizing AI, I mean considering what I said before, right. Utilizing AI is actually a very quick win that firms can adopt in terms of technology to actually quite significantly increase their efficiency. Because let's say if you learn how to use tools like ChatGPT, Gemini, Claude properly, you don't necessarily need to integrate already all of them into your process and build complex agents or complex workflows that API calls, whatever, right? You can just use the tool effectively. And even if you just do that in select task, you can increase your efficiency by 30, 40%. But most importantly actually increase the quality of the output. Make sure you don't miss any details that could be overlooked. For example by human, let's say a personal injury case. Make sure that your argument is as solid as possible. So by making sure that you're actually communicating well with those models and ultimately it's not that hard once you get the hang of it, you will get a feeling for what information the model needs versus what information is extra and not really beneficial. And it's really striking that line. It's like how do I manage the context that I'm given? This AI model, right? So it can do its task effectively. But I'm also not overburdening it with too much information. Example, if I need to maybe draft a client email, I wouldn't want to put all of my medical records about that case right into the prompt and then ask it about it. Right. Versus obviously if I'm trying to, let's say prepare medical chronology, obviously I would need to provide all those medical records, otherwise it wouldn't be able to do the job. So it's just really making sure that you provide enough information to the model for it to complete the task. But don't overdo it. I mean ultimately that's a fine line.

Tyson Mutrix (12:11)

Yeah. Okay, so I'm gonna. Let's break down that example for a second. So let's say you want a medical chronology created out of broke from your clients records. What is the best way of approaching that then? Because you said you don't want to dump them all into a prompt or a thing. So how would you break it down?

Martin Kraft (12:31)

I mean, if it's very simple case, let's say it's, I don't know, hundred pages of medical records, you can potentially honestly do it in one single prompt and you have quite high confidence that the model will do a good job, right. If you're working with a more complex case, let's say 3,000 pages of medical records, then you need to understand the limitations of the models that you're working with, right? What we've seen for example with Gemini, is that after, I would say like 300, 400 pages per prompt, per chat, its hallucination rate starts increasing. Not only in terms of, let's say, making up information, which is actually somewhat less of an issue, it's more so errors of omission, which a lot of times people don't think about. I have supplied this, let's say thousand page medical records to Gemini, how to actually make sure that he has included all the treatments. Right. And that is the biggest, that's the biggest issues with big records. Because ultimately the model is trying to do its best to summarize, right? But it has a limited as well output window. So it kind of summarizes necessarily everything if it's a very long record. So what I would do there is I would just simply break the record up, right? If it's multiple files, it's obviously easier. If it's one single file, you could just really break it, for example, by dates and then process those records individually in separate chats, right? Making sure you don't pollute the same chat with multiple records. So in separate chats, get those individual summaries and almost you have like this a bit, let's say of a funnel, right? You start with a lot of those, like medical records, very big amount of information. You then condense it to summaries of individual records, right? And then in a separate prompt you then for example, draft the demands or draft technology from that where you just operate in, let's say maybe with 20 pages of data, but that's already summarized data. So the model again can do a good job, right? And that way you distill it down to an actual piece of work that can be confident about and verifiable, that is, that's interesting.

Tyson Mutrix (14:33)

I don't think that's how we're doing it. That is interesting. One of the ways that I've seen some of our case managers do some medical summaries was through NotebookLM and I wonder if you've got any thoughts on that.

Martin Kraft (14:46)

NotebookLM is a great tool, but again, you need to understand the limitations of that. Notebook LM has a higher hallucination rate than, for example, what you'd get in Gemini app if you're using the 2.5 Pro model. Right. So the 2.5 Pro is, let's say the more advanced model. It stays for longer but gives you better results. 2.5 Flash is the 1 that's quicker but gives you potentially more hallucinations, less accurate. And so actually NotebookLM on the background is running 2.5 flash. That means you're not necessarily getting the best intelligence per se, and it's great for discovery. Right. So if, let's say you have, let's say all of your medical records, police reports, etc. On the case, you can put all of them in NotebookLM, maybe potentially share it with your team, have it as an operating hub for that particular case and ask it specific questions for essentially, let's say, supercharge control. F. Right. That is an amazing use case of that. When you try, for example, to generate the timeline and the side part has the reports, you try to generate the timeline of that case with that. A lot of times, based on my testing, it would omit stuff or it would hallucinate stuff. Right. And in that summary as well, it's not necessarily verifiable. Can you just click on a particular fact and you'll see where exactly it got the information from. So for comprehensive chronologies that you can trust, I wouldn't use notebooker Lab. I would follow the process just explained. Just because as well, at each point in those individual summaries, what you can do is you can click on any response or any portion of the text and it will provide a citation to the actual piece of text in the document where it sources that. And what I would do is almost divert, let's say from the concept of, let's say a single case manager profile. That's a bit more advanced, right? Like a single case manager profile. And as the records are coming in, have a specialized person almost working like in a conveyor belt style, which would be processing those medical records, making sure that the medical summaries are correct. And when it comes time to drafting, those medical summaries already would be in your Case management system and. Or legal assigned then can just take those, know that the information there is good and don't have to verify with original records.

Tyson Mutrix (16:59)

It's interesting you say that because that's, that's exactly how we're setting up our medical summary agent that we're building out like we just started yesterday. But the, that was admission to you before that we're having it do it one at a, one record at a time. So as the record comes in, it does it, it create so it can. It starts to kind of build to this overall template that we've built out. So that's, that's really interesting. Something that I really like NotebookLM for is to like prepare for mediations or depos. What I'll do is I'll put information in it and I'll use the podcast function where it'll just, it just has a kind of like a refresher. And the question I have though is, do you think that there's gonna be a time where the, the podcasts that people listen to are just going to be AI podcasts that are created from things like that?

Martin Kraft (17:47)

Well, if you go on YouTube, there are channels with 100k subscribers that are fully AI generated. You may not certainly know it, right? But I guess if you work with AI voice models, you will be able to understand that it is AI. So I mean, there is this theory of that Internet theory which pretty much says that say 90% of the Internet is all bots, right? And humans just represent a small minority of it. I cannot say if that's true or not, but I think definitely not for sure. For sure. 100%, 100%. And right now the concept of you see it, you believe it is no longer relevant. I can create a deep fake video of anybody doing anything at this point in time with sound, with video, whatever.

Tyson Mutrix (18:30)

So I think it's pretty obvious at this point for people listening or watching that you like Gemini, Gemini as your performance preferred model. What's the reason you prefer something like Gemini over something like Claude or, or ChatGPT?

Martin Kraft (18:44)

So a few things here. First of all, let's start I guess with the most basic, also not necessarily always relevant for plenty of personal injury alerts. We only really work with personal injury. Gemini is fully HIPAA compliant, right? In. In Google Workspace. If you get it through Google Workspace, you just accept a BA in three simple steps. And now the platform is fully compliant. ChatGPT. No Claude offers that. Unless you're like super enterprise then. Second, if you're already using Google Workspace it is included. And so again the whole let's say part of change management of making sure your people actually use it becomes easier because you don't have to provision new accounts. They just literally go to their Gmail, click on the nine dots, click Gemini. Ta da. They're there. And then lastly in terms of quality, I guess somewhat more subjective but I do find that it hallucinates less and it can process records better. So it's document processing capabilities, especially for things like medical records is Superior compared to ChatGPT in Claude it just seems to understand let's say especially if it's like a badly scanned PDF or similar. It understands better compared to those models but really it's you know all of them are quite similar but those are the advantages. I feel like Gemini is a bit on top. What we also need to understand is that you know Claude recently released a new model. ChatGPT recently well released a new model which is our outer but okay, that's a separate story. Gemini 2.5, the base model also there has been revisions has been released in March of this year and so it's like already six months old and so Gemini 3 is coming. I'm very bullish on that and I, I hope that it will be better than the competition as well.

Tyson Mutrix (20:33)

We're gonna find out very quickly. I'm change gears a little bit. I don't want to ask you about something.

Martin Kraft (20:38)

I'm going to show you this Tyson, maybe just before as well we continue like I have great actually report. I can also share screen share if you want.

Tyson Mutrix (20:44)

Oh absolutely. Yeah yeah let's do that. Yeah absolutely.

Martin Kraft (20:48)

So sorry. Sorry to go.

Tyson Mutrix (20:49)

No, you're good. But I want to ask you about this article. I talked about this on our live show yesterday but I want to, I to want want to talk to you about it. And this is the anti chatgpt which I the title is stupid. Ignore that but Thompson Reuters multi agent system slashes 20 hour tasks to 10 minutes and they say talk about this new thing where it's going to do research and all that kind of stuff. And this is you know rag was the start but doesn't go deep enough and the reason I wanted to ask you about it when we were chatting beforehand either you or Max, your partner, your co founder he had one of you had mentioned the fact that you know people they got this kind of like FOMO so they'll go and they'll go, you know, pay for a product kind of like that. They you all didn't specifically mention that, but I, what I, what I thought was kind of interesting is that it appeared to be like they just built out an agent. An agent or a workflow that goes and does legal research. This is basically what it does. And what I wanted to ask you about was a couple of things. One was the, the. What are your thoughts on people that go and are about them going in? Not, not them, not necessarily the people themselves, but just the idea of going out and paying for something like this that I think can just be built out by, by someone like you that can build out workflows. So I just. Because it seems like this is probably like an overpriced workflow that you could probably just build yourself for a fraction of the cost.

Martin Kraft (22:14)

I'll go even a step further and say you don't build it, you actually just use the app and use it well. Right. And the reason for that is I looked over the last three months, I've been on an AI research spree. I have been looking at all, let's say, legal AI vendors, specifically in personal injury space, have also built workflows myself with make.com connecting directly, let's say, to the APIs of these platforms. And what I realized is that also the output in some of them was sometimes marginally better, sometimes actually worse than using the app. It didn't justify the cost, right? It didn't justify paying $300 for a case unless it's like a very high stakes litigation case. I understand, right, but let's say like 80% of routine personal injury cases. It didn't justify the cost of paying $300 per case with an annual subscription, which, where you can do the same thing, maybe spend 5, 10 minutes more, but then you're just paying $20 per month per user for your Google Workspace subscription already. Right. What ultimately happens is that with a lot of these tools, human verification is still needed. Whatever legal AI vendor tells that their product is hallucination free, that's complete bs, right? It's based on AI. So ultimately, just like AI itself, itself, generative AI, its architecture is ultimately somewhat creative. Right? It's replicated patterns and it's not deterministic, meaning that you cannot get 100% accuracy. And so would you rather let's say pay $20 per user tool and do manual review or pay $300 per case and still do manual review? What I would also say is an issue, I would say with some of these tools that are like some very specific or let's say integrated directly into the workflow that still need manual verification. Is that instead of giving your team and investing in your team and giving them a meta skill that they can use not only for that particular narrow task, let's say of medical record summarization, but they can go and ask it to explain a concept to a client. They can go and ask it to, I don't know, create an image for a marketing campaign. Campaign. They can go and learn a new concept by themselves. Right. With a tool. Instead of investing in your team and giving them meta skill which will serve all their career, you're paying this provider and providing maybe like, I don't know, 3, 4, 5 sits to the lucky ones in the organization who get to work with it. So I would say there is using those tools, but again, enterprise level. Right. Or if you, let's say you said, okay, I run a 20% firm, I don't want to grow anymore, I just want to make it more efficient. Okay, maybe makes sense as well. Right? But if you're trying to grow, keep it simple. Make sure you actually adopt technologies that drive real value, tangible value, as soon as possible and require minimal maintenance. Right. Because what we also need to understand is even if you hire an engineer to build a more complex system like this, it also needs maintenance. So I think practicalities is probably the, the, the most desirable skills I want to have, let's say low four borders to have when it comes to all of this.

Tyson Mutrix (25:25)

So you built, you and Max have built out Swans. So for those of anyone interest, Swans Co. And we offer sustainable and scalable solutions. And so there's, you know, you build out these systems and everything for people, but we offer sustainable and scalable solutions where marketing consultants fall short. That's, that's from the website and everything, but you definitely do more than that. By 2030, our goal is to empower 10,000 law firms across North America to double the revenue. That is. That's amazing. I think that's a wonderful vision and everything. I do wonder what, what got you into doing this specifically with law firms? That's because, I mean, usually there's some sort of something that triggers someone going into a specific industry. So I wonder what, what about law firms and specifically PI firms?

Martin Kraft (26:15)

Well, originally, I would say my response would be somewhat anticlimactic. It was a bit by an accident. I wanted to do back six years ago, I wanted to do automation. Right. I wanted to. I really got excited by tools like make.com and Zapier where you could build workflows with little to no coding and provide actual business value. Without waiting for, let's say a month of a Python engineer, you know, building that on your back end. And I got very excited by that. I wanted to specialize in a particular industry and I felt just like legal is quite an antiquated industry in terms of processes and there was a lot of potential in there. And so that's why I started Automation for Lawyers originally my solo consulting agency. I worked with multiple clients throughout the time. I actually then joined a law firm in the uk, settlement agreement solicitors, then actually became a partner there later and for them essentially built out their full CRM and case management system, bringing the company from five people to now more than 40. And the company having more than, I believe, like 6,000 now, five star reviews on Triespilot. It was a very specific, let's say, niche of employment law in the uk. But what I think that showed me is that there is a huge opportunity for more entrepreneurial spirit within legal, right? The law firm owner that founded the firm was actually an entrepreneur, right? Entrepreneur that previously ran actually a few consumer brands. They partnered with legal Director, opened up the firm because they saw there's just an opportunity again in the market where not good enough customer service was delivered in that particular niche. And so he decided to form the company and reimagine the process. Like, okay, forget about how law firms are doing doing this. I want to actually understand, like what drives customer value, what's actually valuable, and let's design the process around that. So we designed the full system. The company group later then partnered up with Max to reform out of reach for this one. And the reason that we chose to go for personal injury is I feel like that's the most as well, somewhat entrepreneurial part within the legal ecosystem. It's one that's a bit more competitive and one that's a bit more open to trying out new things, right? And so I feel like that is just the natural first step into making an impact in this industry and almost making sure that we help a few firms a lot, right? To actually drive impact and doing that almost like brute force this industry into providing better customer service, being more customer centric and as well being more entrepreneur by using technology and AI and everything they do.

Tyson Mutrix (28:53)

I think historically, I think you're absolutely right about the PI space. It has been more entrepreneurial. I do think that that's changed substantially over the last decade or so where it's shifted quite a bit. I still think it's. PI is probably still predominantly the most entrepreneurial of the entire legal space, but it Definitely has changed quite a bit, which I think is good. I think it's good for the legal space in general. But I think you're really interesting because you're different from a lot of automation people and AI people. It's funny because they are different. But you're now sort of both. You have to sort of be both. You can't just be automation or AI, you got to be both. But what makes you different though is that you have been inside the inner workings of a law firm and helped build the systems and processes, which makes you substantial, substantially different, I will say, because you have to understand how the legal space works. You really do have to have a understand of the inner workings. And with that, what are some of those basic maybe AI agent workflows or workflows in general or automations in general that you think that law firms should have? Like every law firm should have them.

Martin Kraft (30:08)

Every law firm is a tough question just because I feel like again, at different stages for of your growth, you will need different.

Tyson Mutrix (30:13)

I knew, I knew the moment I said every law firm, you gave me a little pushback.

Martin Kraft (30:18)

But, but okay, it's, it's like, it's like if, if you are a solo team, like what you need, you don't need clio. Like you need to hire your next paralegal, your first paralegal, right. You need that. And you could work on an Excel, that's fine, right. When you're three. Okay. Get a case management system. Right. So. So I can talk about, like I say, a few major things we are working on, if that's helpful.

Tyson Mutrix (30:44)

Yeah, I think so. Let's hear that.

Martin Kraft (30:45)

Okay. So I mean, obviously one that we've been talking about now, especially relevant now is I would say AI adoption. Making sure that your team not building some crazy AI workflows, which there's a place for that again, once you've covered the basics, but actually making sure that each one of your team members adopts this technology, is comfortable with it, and use this on a daily basis. Right. That is really where the power comes. And that's why I feel like where a lot of these legal AI vendors ultimately fall short is because they provide this great tool. They do a few onboarding sessions and then that's it. Right. What we found is even if we do group sessions, we still need to jump one on one or at least department level with our clients and actually walk people through, make them feel comfortable with technology, try out a few things with them to get the ball rolling and then to check in on a Periodic basis. So AI adoption would be one thing. Another big one for growing firms is actually analytics. I think a lot of case management and CRM systems in the legal space don't do an amazing job at providing great analytics. Things like, for example, in your case management systems, on average, how fast are cases progressing from stage to stage? Right. If you're growing your firm, I would love to know that information to understand where the bottlenecks are and other bottlenecks due to, let's say, the client not providing information to the defense being too slow. Or is it actually an internal holdup that I can solve for? Right. That's critical information. I need to be able to make the decision. Things like, for example, especially if you're working in PI, one thing that struck me is that a lot of firms, even somewhat substantially large firms, don't have proper financial forecasting and analytics. Right. So understanding when is cash flow coming in, making sure the data is synced to your case management system where you have, let's say, projected settlement values. Right. Understanding how much is coming in every month in the future to be able to make a hiring decision, for example. Right. Not only that, but also actually understanding how much value is your team generating every month. Because obviously in PI, you start working on the case, you invest essentially for nine months into your team doing the work, but then you only see the result. If you look at your books and you're growing fast, you will think that you are unprofitable, but it's not really the case. Right. Because you're now building value for those future cases. And so having some kind of infrastructure to be able to again, take the information from the case management system process in aggregate. Right. And displayed in a nice analytics dashboard is very important. And obviously all the stuff on the CRM side of things like intake speed to lead, obviously a lot of big things there. And then we would say, let's say, optimizing your CRM and case management capabilities. Right. A lot of firms are already using great tools, but they're not fully maximizing. Right. Maybe you're not using conditional logic within your document templates and you have 200 templates sitting in your case management system, where you could only just have 30, which obviously complicates maintenance and errors, whatever. Or maybe, for example, your case management system actually has the ability to apply task lists automatically when the case goes from one stage to another. Again, that's very useful. You can run analytics on those task completions later. You can make sure your SOP, you don't need that many SOPs, because pretty much everything is in your case management system, CRM side of things. Again, automation that front, making sure that again, data is properly organized and displayed within that. And then we would go into, let's say, more advanced stuff such as creating custom automations with may. Com zapier, those more advanced platform, AI agents, et cetera, et cetera. But as you can see, there are a lot of basic stuff that there is more benefit to getting them right rather than immediately say, jumping into the most advanced flashy workflow. Because if you don't have good data, if you don't have good processes, that will not work.

Tyson Mutrix (34:23)

So I was quickly trying to pull up the data from one of our dashboards because you had mentioned the logjams. And we actually went through this yesterday during our quarterly meeting. And I was very, very surprised. We. I don't know what the type of chart is called, but it's. It's the type. It's those charts where it kind of flows. You take the name of our charts are. It's source versus phase. And so where are all of our clients coming from? And then what phase of the case is it in? In one of our phases, we have case on hold. And it could be for a variety of reasons and it could be, you know, clients. I get us information, I can't get a hold of the client. There's things like that where we, okay, case is on hold until, you know, we figure out whatever's going on. And the biggest source for case on hold is client referrals. So current clients or clients that have referred us cases. That is the biggest source for case on hold. And it's interesting without this chart, I guess we could do the chart another way, but it's. That one was kind of. You've got like the. It kind of connects each other and it's like kind of wavy.

Martin Kraft (35:31)

I don't know the name of that water waterfall.

Tyson Mutrix (35:33)

Is that what it's called is a waterfall chart?

Martin Kraft (35:35)

I. So it's, it's sourced against, let's say, in your case management system.

Tyson Mutrix (35:39)

Yeah, exactly.

Martin Kraft (35:39)

Source against stage.

Tyson Mutrix (35:40)

Yep, yep.

Martin Kraft (35:41)

Probably. Probably just a matrix chart, I guess.

Tyson Mutrix (35:43)

Yeah, it's kind of interesting. But the. So it's interesting though is like, wow. So it, it really has me thinking maybe we need to reassess where, like whether or not we want to devote as much time to try and get client current client referrals. I don't think, I don't think that necessarily we're going to stop trying to do that, but.

Martin Kraft (36:04)

All right.

Tyson Mutrix (36:05)

We're going to, whenever a client refers a case, maybe we just look at it a little bit harder as to whether or not we want to take that case. But it's interesting when it comes to that data though, so I'm glad you mentioned that because it triggered that memory from yesterday. Data is so important.

Martin Kraft (36:19)

Look, if you're running a law firm, law firm is a big business. You need to think like a business owner. Right. As a law firm owner, you're not an attorney, you're a business owner. Right. And so, and so you need good data to make good business decisions, right. Finance sides, things like this. You know, I'll probably go even further. I'll say like, hey, let's analyze for example, average settlement values for the different sources and maybe like those client referrals are the lowest one the list. You're like, okay, you know, maybe we shouldn't take any referrals. Right. If you don't have the data, you cannot make the decision and then you just fly in drug pretty much in your business, you know.

Tyson Mutrix (36:50)

Yeah, people definitely need to know that one. That's one that you definitely need to know. Where are your highest value cases? That's. Yeah, no, no doubt for sure. Is there going to be a point where when we talk about AI, that AI and agents are just, it's just synonymous. It's just like when you're AI is just AI agents. Like is there going to be a point where that, that happens?

Martin Kraft (37:14)

Maybe. I would say, I would say there are a lot of, I think an agent, right. It's like we need to understand what is an agent. Right. An agent, I would say is, let's say the capability of a system to make decisions by itself without supervision. Right. And so even if you look at for example, two point, let's say Gemini 2.5 Pro ChatGPT 5 Extended Thinking. It is a somewhat agentic system already, right. When you ask a good question, it reasons about it, Right. It makes certain conclusions, you know, goes one path versus the other. Right. It has, let's say, chain of thought which you can actually see in the, in the software as well. Right. So it is already somewhat energetic system. Right. Extrapolating that further would be okay, you just say, okay, send an email to this client about X and then goes to your case manager, system pulls up the email, drafts the email, then actually sends it out. You give some more tools to, to the system. So I think those are very related. I think it will ultimately depend on I guess just the social perception of it, the media. But I think it also will be quite synonymous in the end, especially the abilities of the systems become higher and their capabilities become more extensive, primarily through ways of integration. That is meaning actually be able to do, let's say, actual work, send an email, look up certain information, not just look up a website. Right. Make a certain action within your CRM case management system. That will become more synonymous.

Tyson Mutrix (38:43)

What misconceptions do you think lawyers have about agents in general?

Martin Kraft (38:47)

I think most people don't know what agents are. Right. Okay. Let's start with the fact that most people don't truly understand the full capabilities of just like ChatGPT and Gemini and Claude. So I think, again, it's almost. Let's cover the basis first, right? Make sure that we are comfortable using the tools. We know the capabilities of those tools in some of our trainings. We even have, let's see some younger people on the team of our clients that have used ChatGPT for their school work, for example, university, whatever. And then we hear back from them. It's like, oh, I actually didn't know that this tool could do X. Right. Like, they're saying, okay, I could use deep research or like, oh, I could create this. And like, this prompt would generate this result. And so I think, like, there's a lot of legwork that we need to cover in there. Once we go into that, we can speak about agents, I guess, in the future. Right. But it's somewhat more experimental. Unless, again, your team is fully adopting AI already, you have your systems on point, you have your data on point, you have analytics on point. I feel like agents are a distraction.

Tyson Mutrix (39:55)

Yeah. For. But I'm gonna stay on agents, though, for a second though, because I do think that it is. It can be a distraction for the people that are just kind of. They don't have some sort of integrated system with it and all that. But early on when agents were starting to develop and become a thing, it was a lot of what I was seeing was you're kind of building teams of agents where, like, they supervise each other and everything like that. And I wonder if that's. If that's something you would agree to, a ways of setting it up or do you think that that is a flawed way of thinking now?

Martin Kraft (40:33)

Well, that I would probably even call more so, like, agent swarm.

Tyson Mutrix (40:36)

Agent swarm.

Martin Kraft (40:37)

Right, Swarm. Okay. I think that's. That's the term that this video used. If you look at like. And I said YouTube videos, agents war, meaning, like it's essentially a group or a team of agents working together. Right? I would say, how would I define, let's say an agent in a more, I guess, simple sense is just like a system that can make decisions and then can use tools. Right. To execute its goal. And so for that, I wouldn't necessarily say that a team of agents is necessary or a team of those systems is necessary. It could be one of them.

Tyson Mutrix (41:11)

Yeah. Because as we built out more of our agents and everything, it's, it doesn't, it does seem like it's more linear than having this like stacked pyramid of, of a team of agents. Because the way early on it was, okay, so you got this, this team leader and then there's teams underneath it and teams underneath it. And I'm not saying that's wrong or right, but I can just tell you the way we built ours out, it's more of a linear thing based on other triggers, not necessarily based on supervisors and team leaders and the worker bees and all that. It's, it's, the setup is way different.

Martin Kraft (41:48)

It's a workflow. It's a workflow, right. It's AI augmented workflow. It's not necessarily, let's say an agent, an AI assistant, you can ask anything. It is designed to execute a specific mission and it is operating within the boundaries of that. Which I think is the right way to approach as well, ultimately automation at this point in time. Because again, you know, you can, for example, interact with ChatGPT agent. Right. It has the build function there now. But the results you're going to get will be very variable because there's still a lot of. Because AI hallucinates, because AI is not deterministic. There's still a lot of opportunities for error. And so I don't think, just like it's there yet to consistently be delivering value in those very highly agentic systems compared to more narrowly scoped automations that use AI. Right. AI augmented workflows.

Tyson Mutrix (42:34)

I'm very hesitant to ask you this question because I don't want you to say something other than what I'm using. But when it comes to build out agent workflows, what is your preferred system to use?

Martin Kraft (42:47)

There are a few, There are a few on the market. I would say NA10 is quite popular.

Tyson Mutrix (42:51)

Thank you for saying that because I, I was afraid you were gonna say something else. And I love Nadin. I think Nadin's great, but I know there's others out there, but I think it is, to me, it's the most intuitive, but it also has to me the most features.

Martin Kraft (43:07)

Yeah, it's somewhat, let's say more technically focused. And again that's why we really only deploy it for more advanced customers per se or where we're progressing more. One thing there that I kind of dislike is just the fact it doesn't have a proper like cloud hosting option. Like if you use a cloud version by any time it's hosted in German service in Europe. If you use a self hosted version, obviously you can use it whenever. But there is more technicalities than setting it up and maintaining that than just let's say sign up for zapier or.

Tyson Mutrix (43:35)

Make.Com Interesting because I mean I didn't even notice. It's one of those things over here. I just didn't notice at all. That's really interesting.

Martin Kraft (43:46)

Check it out.

Tyson Mutrix (43:47)

I want to go and look and see for sure. Is there anything that you are. Because it seems like technology is, is it's changing so rapidly, rapidly and it seems like there's something new that comes out every single week and you kind of the shiny object syndrome. But is there anything that you've seen out that you are super, super excited about? You're like, oh my gosh, this tool is just amazing.

Martin Kraft (44:11)

I get those moments through marketing copy and then I try it out and then I have disillusionment, you know, especially if we talk about like specific, you know, vendors coming into the market and promise in the world. What I think ultimately the last, you know, doing this for the last year has taught me is that simple systems, boring systems are the ones that actually deliver, deliver value. And as a tech minded person I'm very drawn to finding the latest and greatest. But then when you actually put to practice, when you think about security considerations, I know HIPAA compliance, where's your data stored, et cetera, you're like, oh, okay, this tool is not ready. If I need to recommend the tool to 100% organization, I cannot necessarily come up with a startup that has three employees on LinkedIn and tell them oh yeah, use this. Right. I would be failing my duty as a consultant, right. To tell them that like this tool is quite new in the market. It may fail, your data may not be fully secure with them. Right. So that's why I think the scope of solutions that you use for enterprise or for bigger firms just like really narrows.

Tyson Mutrix (45:28)

Yeah, I'm so glad you said that. I think that's really smart. Boring is, I mean boring is better. It's one of the things I just think boring is better. A lot of times consistency is boring, but that consistency is way better than anything else when it comes to adoption. Because let's say you're going, you're starting with a clean slate that you've got, you've got kind of nothing. And so people have like the right mindset about this if you were to. And I guess you're never really going to be fully integrated with everything because you're always going to probably be adding on and building things to improving and changing things. But to get a fully workable system with all your workflows and I'm not even talking about with AI, I'm just talking about just generally from a tech standpoint, what is generally for like a, let's say a mid sized firm, let's say you got you know, 10 people. What like what is the, the whole timeline when it comes to, from the point you start to the point you end. Is it, is it a year, is it a month? How long generally does that take?

Martin Kraft (46:26)

I think again this is somewhat a hard question to ask because like there isn't had been like a clean slate for a 15 person firm. Like they're already using something, right? And ultimately like even when I worked in house, right in my firm, worked there for about three years and even after those three years of full time work there still was stuff to automate and still stuff to features to develop, right? So I feel like almost like this is somewhat infinite, right? Especially as you expand potentially into different case types as more people join, like your problems start becoming different. I don't know if you go past 50 people, okay, maybe I need an HR system, right? Maybe I need a more expensive, powerful payroll solution, right? It's like the types of problems change. What I would say is you really need to just start simple and take baby steps. This is the most important thing if you're going from a clean slate. Unless you're working with somebody who has already built it a thousand times to advise you on that. If you are a 10% firm, you're starting to technology use the most proofed tools on the market. For example, the systems we like to recommend based on everything that we see generally would be let's say Clio Manage for case management and Lawmatics for the CRM side of things. Just because those are easy enough to be able to configure it and maintained by the firm staff themselves at the same time as providing the necessary functionality for some more advanced automation like email, SMS, task list and Clio manage, etc. And so take baby steps. Don't try to do too many things at once. Start with your biggest issues like hey, document Management is the biggest issue. Let's solve that. Okay, now I see that, for example, my people are doing, like, not necessarily following the process. Okay, let's implement task lists and go step by step. But I don't think there's like a very simple answer to this. To build a functioning system, if you dedicate, I don't know, maybe like 10 hours per week to it, I think you can have something like an MVP already within the first months. You'll just need to add on top of it and potentially then utilize more and more features of the software that you already have. And always check the features of the software you already paid for before trying to, I don't know, like going to make zapier and a. Because you may be surprised. You may be surprised.

Tyson Mutrix (48:44)

Amen to that. That's for sure. I think it's a great answer. I think you answered that perfectly. The last few questions I have for you, it's more of prediction questions. So, yeah, yesterday we were having a little bit of a discussion about neuralink and all that, you know, whether or not you put a. Get a chip, put your head. We were having a kind of fun conversation in the firm. And. But what I said was, here's the. Here's the problem, right? You've got. You're gonna have some attorneys that, when it's available, they will get the neural. They'll get neural link and they will have access to just gobs and gobs of data and information that, let's say you're arguing in court. I think this is a real scenario. When I want the prediction, I'm gonna ask for you. If you think that this is actually a real world scenario where you've got one, a transient neural link, and they can, they're gonna be able to argue all this information that's in their head because they've got the chip against someone that doesn't have it, who's gonna have a substantial disadvantage. And do you see, do you see a situation like that coming in the future where you're gonna have attorneys with neuralink versus attorneys without neural link arguing each other against each other in court?

Martin Kraft (49:48)

I think that would be the, I guess the natural progression of how things are going. Right. If you look at the past few years, it's almost right now, it's like, okay, lawyers using AI versus lawyers not using AI. And Bannon, it's used in their practice, which I think is insane. Yeah, it will be similar, I guess, divide. Just in that particular scenario, I would say, with using it in court. Question is how fast would they be able to process, let's say the data that is coming through neuralink and actually make better decisions? If you actually prepared your case well is like, how much more can you boost your performance by having a neural like, I don't know, I don't know. But it is scary, exciting times. I feel like we'll be coming with this. I don't think I'm putting that thing in my head, but I still want to. It's, you know, like they have like put it essentially, you know, kill switches in cars and whatnot. Right. I don't want to put the kill switch in my head, thank you very much.

Tyson Mutrix (50:53)

I mean it was a scary thought. It really, it was a scary thought. And so I'm with you on that though. A couple more questions. What skills do you think next generation lawyers should prioritize over the next few years?

Martin Kraft (51:07)

Empathy and customer centricity. More so than just like say, I mean, of course, right. But with introduction of AI, I feel like being that guide for the person for your client in distress will be the differentiating factor. Adding more customer centricity to this profession will be the differentiating factor. Ultimately, just as I have seen again in the firm in the uk, which dominated the industry because it focused on the customer. I feel like that is common for other areas and regions of law as well.

Tyson Mutrix (51:43)

Last question. So if we zoom out 20 years from now, do you think that that law is going to be similar to like aviation where you've got pilots that are monitoring the autopilot system and all the systems going on in the gauges, where you've got agents, where the firm, the, you know, paralegals might be monitoring different agents doing different things, attorneys monitoring different agents doing things. Do you think it's going to be, do you think it's going to shift more towards that? Is that, is that what the legal profession is going to look like, like in 20 years? And again, this prediction, we don't know, we don't know how much things are going to change. But I, I, I am curious what you think.

Martin Kraft (52:19)

Okay, so on the most strategic level, unless the judicial system changes, right? Until, and so long as we have human judges having human lawyers and being able to transmit emotions and, and stories will stay relevant. Right? So I don't think that strategic part, unless, you know, the whole judicial system gets replaced by super intelligent AI, which I don't think is happening, I think that will stay the case for more, let's say mundane work, meaning analyzing medical records, drafting responses, etc. That part I feel like will get closer to what you are saying. Just as we are seeing now in coding, I look at people using cloud code for. It's like a AI coding solution pretty much and Copilot pretty much for coders. And they have their computer with three different chats. Like one is writing a new feature, another is running the unit test and the third one is debugging an issue and they just switch in between them quite quickly while waiting they work on the other one. So I feel like that is a possibility. But. But I don't think that will be for say, the more actual strategic work or again, the primary function of the lawyer check will be like that guide.

Tyson Mutrix (53:39)

All right. In 20 years you're going to play this back and see if you're right. Very good, Martin. Thank you so much. If anyone is interested, go to Swans Co if they want to reach out to you personally. Is there, is there some sort of social media channel?

Martin Kraft (53:51)

Hello, hellowands co. Or you can reach out to me on LinkedIn. Martin Kraft. Yep.

Tyson Mutrix (53:58)

Love it. Thank you, Martin, for doing this. I look forward to working with you at the workshop in Nashville, at the conference at Maxon.

Martin Kraft (54:05)

If you want to take us, go.

Tyson Mutrix (54:05)

To maxon.com Martin, thank you so much for doing this. Really appreciate it. Enjoyed it.

Martin Kraft (54:09)

Thank you so much. Tyson. Hey, I've got a question for you. When was the last time you drove over to another law firm near you, sat down over lunch and traded every business tip you've got talked about what's working, what's not and what to do next? No, see, that's what Maxluocan is for. It's real conversations with law firm owners who are actually doing this building, leading, scaling and willing to share what's working right now. You could keep doing it alone, but let's be honest, it's slow, slower, harder and way more expensive than getting in the room and shortcutting the learning curve. As of this recording, we've got 20 seats left to this year's event. Skip the guesswork, go to maxlacon.com and grab your ticket before they're gone.