Maximum Lawyer

Tyson Mutrux (00:00)
Welcome back to Maximum Lawyer. And today I'm to be talking about the second episode in a series of four about AI agents and different frameworks you can use. And today we're going to be talking about the one that's called Paralysation. You might hear it called Paralysation Framework. Just by the name, you might get an idea as to what it is and what it does. But I'm going to make it as simple as possible for you. So I'm going to break this down.

I am for those of you watching, I'm going to show it to you. Otherwise, I'm going to explain it without you needing to actually watch. So if you're listening to this on the podcast, don't worry about it. You don't need to leave here, go to YouTube, but if you're on YouTube, obviously you can see it. So check out both if you want, but don't worry if you're just listening. I'm going to try to make it as simple as possible. This is a very incredibly powerful tool. It's more complicated than prompt chaining.

in that you've got multiple things going on at one time. It could make it a little bit easier to break things. Just know that. You have to be a little bit careful when it comes to planning it out. But it is something that could really speed up your workflows. Think of it as you've got multiple highways. of you're trying to send everyone through one highway, you've got three highways and or...

And so you're kind of spreading it out a little bit. And so you're able to get places a little bit faster. That's the best way I could probably put it. Before I do get into that though, been getting lots and lots of great suggestions. People want me to cover a lot of things. Like I said before, and I'll say every time, I'll cover it if I can. I can't cover absolutely everything, but if I can cover it, definitely will. You can do a few different things.

You can comment on, uh, if you're watching this on YouTube or wherever you get your podcast, if it allows you to leave comments, some do some don't, you can do it there. Um, or you can text me 314-501-9260. That'll be in the show notes. don't have to remember it. Save it to your phone. Shoot me a text. I'm happy to cover, or if you just want to say hi, that's also, that's also great too. I'd love getting, um, get, I get a lot of great information from you. A lot of great texts. So really, really appreciate it. Let's get into this though. You know, what is.

parallelization. So we've been building out a lot of agents in the firm. And I will just tell you when it comes to ease of use and really kind of thinking things through, prompt chaining is super easy. It's probably the easiest one. It's kind of like an assembly line. If you want to know more about prompt chaining, listen to the previous episode, but pretty simple. Parallelization though, on the other hand is...

instead of handling tasks one after the other in a sequential line, parallelization, it allows you to manage multiple tasks at once. And it's doing this all at the same time. So imagine having multiple specialists, whatever they may be doing, working independently on the same project, but focused on different aspects of it.

And then what you do at the end of it is you're combining all of their insights, all of the things that they're doing, all the work, and then you move them together, you add them together, combine them together for a more comprehensive result. That is really what parallelization, what we're talking about. There are several things that I can think of within our office that we could use parallelization for.

One of those is written discovery, but we did instead choose to do prompt chaining. We didn't want to use parallelization, but we could. Demands is another one. We've not built out demand AI agents yet, but it's on the plan. It's part of the plan. We will get to them. That's one where we probably will use parallelization because we're going to be combining multiple things from different sources all at the same time to speed things up.

Um, that is in the works. I also did show one, um, a few weeks ago that I'll show here today. I'm going to show another one that we did not build out, but what I did get, I get it from the NADN community. Cause I wanted to show you a little bit more of a complicated one. It's not really complicated. It's just more complicated than the one that we had showed before. So I'm gonna show that to you as well. But I want to give you sort of an analogy to understand, um,

When it, what we're talking about when comes to parallelization, I come up with another one for understanding automations versus, so automations versus AI agents. And so I used a hammer and a nail.

We've actually since added to that, but, we talked about more of that in the guild, but for when it comes to paralyzation. So imagine you're making a sandwich, you're pouring a drink, you're grabbing a snack. So think about all the things you have to do with that. but you only have two hands, right? And you can only do one thing at a time with each, with each hand. That means, so you're doing that that's sequential, right? So if you, if you were to take one hand, take one hand, you go.

I'm going to grab the mayonnaise. I'm going to grab or if you want peanut butter jelly, can do peanut butter jelly or I'm going grab mayonnaise. I'm going to grab the bread. I'm going to grab the cheese, the meat. All right. That's sequential. So you're going one at a time. But now instead, right? Instead of you doing that, you're doing it kind all yourself with one hand because you're trying to, you know, hold the refrigerator open with the other hand. Now imagine you have three friends. Okay.

And so one of them is making the sandwich. One of them is pouring the drink. And then the other one is grabbing the snack. All right. All at the same time. That's paralyzation. Whereas like you can't as a, so if you think about each of those as different agents, it's really what we're talking about. We're not talking about different automations. We're talking about the, the, if you're kind of thinking about it linear, in a linear fashion, it splits off into three different.

ways or more and different agents are handling it. Okay. That's, that's what we're talking about. And obviously, because you're one person, you can't do that at the same time. You're just one person. Even if you had four hands, you probably still couldn't do all that. So with parallelization, the whole lunch is ready in about a minute or two instead of, you know, five minutes, because you're not, you're not having to run around the kitchen, do everything yourself. and that's what, that's where the real power of

parallelization comes in is where you're getting a lot more done at one time. It does however, it does however require more coordination and you gotta be really, really careful. And so there is, in Harvard Business School, they actually have a quote about parallel processing is the foundation of modern computing and it's quickly becoming the foundation of modern business workflows.

It really is kind of true because you do have everyone really trying to do as much as they can in a short amount of time. It's really interesting seeing how other people are building out workflows and they're really kind of sandwiching everything in. And what is interesting is that sometimes people don't get the logic right. So they don't fire off correctly. you have to make sure you have the logic correct when you're doing all those. Now, why is it...

Something I want talk about is like, why is this something that you might care about in your law firm? Because it's, I understand why some of might be well, why should I even do parallelization? Why should I even be doing AI agents in the first place? Which I talked about the reason why in a previous episode, so I'm not going to get into that. here's the thing. Law firms handle, we handle most of us very complex issues on...

on a daily basis. And even those of you that don't handle, you're not doing litigation, you're doing other things, even though they may not, in the grand scheme of things, may not be super, super complex. know, sometimes we do make our jobs a little bit harder by making them more complex, but I'm not talking about that. I'm really talking about, even if you're not doing like, you know, serious complex mass tort stuff, right? You can use this for just about anything. The simplest...

of workflows, you can combine these for parallelization. You can combine them to really speed up things. And that's the power of it is that you can make things so much faster. Just think about if you could get your matters done, your cases done in a fraction of the time, just by shortening, you're shortening the workflows, what you're doing, but you're kind of spreading it out. You're going a little wider and

Usually when people say the wider and deeper, it's usually that analogy is usually not a good thing. Cause you're talking about knowledge and this situation, we're actually shortening the length of it. we're widen widening how much work's being done at one time. It's really what we're doing. It's really, really beneficial, but you could do this from when it comes to intake, opening files, reviewing discovery documents and compiling the discovery documents, drafting pleadings. There was one that.

I thought of yesterday that I thought would be really cool where emotion comes through this filed by the other side. And you've got one agent that starts that basically is adding it's adding it to the file. It's feeling more like mundane tasks, right? So you're adding, adding it to the file. And then what another one might be doing is it might be taking that and summarizing the arguments from it. And then another one might be taking and checking all the case law, right?

And then after that, can kind of split it off in multiple things too, one is putting it, like, let's say the other side is filed a motion for summer judgment. So you're, taking our responsive documents that we might need with our motion. You're compiling that you're doing the combat, doing the research on it. You can do a lot of different things, which is really kind of interesting when it comes to the parallelization. And it is a lot to juggle. It definitely is. But so make sure you're starting kind of simple and building from there.

And the way I think the way that the reason why some people get a little bit, they get a little confused with this model of doing things or they, they think, my gosh, this is too complicated. Just because we are just naturally used to doing things sequentially because we are, we're a human and we're like, we really only can do one thing at a time if we're going do it well. And that's, that's completely fine. That's a human thing, but

Just remember that is what slows us down. That's what makes us a little bit slower. So if you can really capitalize on this, it's going to give you massive gains because you can do things like analyzing things simultaneously. And I'll give you an example. What we're doing with written discovery where you've got one agent is taking and trying to remember how we're doing this, you're taking one agent is...

taking and checking for spelling and grammar when it comes to the discovery responses. Another one is taking it and looking at it and seeing, looking at the file, analyzing the file, determining, okay, what additional information do we maybe need in our discovery responses? And then the third one is analyzing what's already been created and in our discovery responses, because at this phase of the workflow, it's later on down the chain. And the other one's analyzing, okay, what else based on what we've requested?

Can we also add to this based on what we already know? That's a pretty simple one that you could add into your workflow and it's kind of split on later down in the chain. So lots you can do with it, lots and lots of stuff you can do with it. And that's what's great because that's what parallelization makes possible. And what ends up happening is you got more speed, really deeper analysis, which is something that is...

To me, maybe one of the best advantages of it all, not speed, but the deeper analysis of it. You're going to have better case outcomes. You're going to have faster case outcomes. going to get paid faster, get the cases resolved faster, which is awesome. And which is going to lead to really more efficient law firm, higher profits, which is what is really, really important. So a little stat for you that you may or may not have known, but according to McKinsey, which puts out a lot of great stuff.

Up to 30 % of tasks in legal services can be automated using existing technology. And many of those are actually already in parallel workflows. That's 30 % of tasks could be automated. Okay. Now imagine what you can do with agents now, because it's got the thinking component, right? It's got the, and it's not technically considered thinking. It's got the analysis component, but I always like to call it thinking because that's the best way of

comparing it to humans. And that's the best way. One of the easiest ways of thinking about comparing it to automations is the agents more have like that analysis component, that thinking component, which is what makes them so powerful. All right. Let's get you a real world example. I'm going to pull a couple for you and tell you about those so you can get a...

get an idea as to how to view some of this stuff. So this is one, it's, ⁓ this one's an all-in-one video prep tool where you can start it with just a simple prompt, a chat prompt inside of N8n. And what that one does is you prompt it the kind of video that you want. And these are, I'm not going to go through each of these agents individually. like,

into the backend. There's no point in doing that. If you want, I handle this one in a previous episode, you can go look at that if you want to see it. in this one in particular, you get the initial prompt and then it, whatever the topic is. So let's say you're a state planning attorney and you want to draft something on the importance of having a trust, right? That could be the topic.

In that situation, so one agent is drafting the script writing and then also doing some editing of that script. Just as a side note, I don't necessarily recommend doing it that way. ⁓ Really what you should do is probably create an outline for one, create the script writing for the next one, have an editing agent for the next one. So you kind of break it down to multiple, but for this, don't worry about that. ⁓ We're just going to combine it and ⁓ you don't have worry about it. But this one, got, it starts with a chat prompt.

You have the script writing and editing agent is creating the script. And then you have another agent at the same time creating the title that's specifically for YouTube. And then a third one is doing some key data and statistics analysis. Okay. And that's combining those. can then later on, it's going to merge it later on. So then you then create this master script, adding this additional information. So.

After they've all merged, you are, you're going to end up getting is you're going to have an optimized YouTube title. You're to have an optimized script with data that you can add to your video. then some, any, any other statistics that you want to add that are, ⁓ cited sources, it actually goes out and find sources in the actual data. And then you can use that in your script in your video. So that is, it's one of the, the power.

powerful things. Now you have all that boom pretty quickly and you can fire off a video really, really quickly too. And I've seen a lot of really cool things where people are creating ⁓ AI videos without any humans, which is pretty nuts. That part, that's pretty incredible. So let me show you another one though, because there's some really cool ones that ⁓ I'll see, I'll stop sharing for a second and then...

to show you this other one, which is pretty cool. And this is one that I pulled from the N8n website. And I put it in this workflow that I called our Playground. And on that one, you can see how this one's a little bit more complicated. And again, no need to leave if you're listening and watch YouTube. There's no need.

But this one is split into three different channels and it starts with ⁓ a web hook where there's many different ways of doing web hooks. But let's say that ⁓ you have a new case that's opened in whatever your lead management system is, and that kind of fires off a signal to your software. All right. Let's just call it that. That's the easiest way of putting it. Your web hook inside of N8n picks up the signal and then it knows

I need to then do multiple things. Okay. And this one, ⁓ in particular, this is take, this is one that is set up in N8n for whenever there is not some specific integration. So this actually might be perfect for what you need. And that's part of the reason why I picked it is because this is one built out specifically for there's no direct integration. So you have to build out the integration using, ⁓ API keys and all that mumbo jumbo and, and what books where

So let's use a scenario that it's a very real world scenario. So let's say you're using something like PipeDrive. It sends off the signal. It's picked up by N8n. Now you got to open up the file. So maybe in this situation, it goes into the first one, you're opening up the file and say file by, okay. Then there is no direct integration when it comes to N8n for file lines. it splits off into one, it's creating the file.

In this particular one, you've got split into items, data scraping, then handle pagination. So in this one, let's say you want to go and pull some data from your state courts website. So the first part of this is it's opening up the file inside of Filevine and doing what needs to be done in that. The second part of this is that you're pulling and we can call it, let's say it's a criminal case.

pulling the case information from your state court's to then later on add it to the file. And then the third one, let's say you are, in this one in particular, so this is handle pagination. Sometimes you need to make the same request multiple times to get all the data you need. It's called pagination. So the pagination process goes as follows. Loop through the pages of the input source, which is HTTP, blah, blah, blah, blah, blah, get my stats starts.

Increment the page. So I don't have my glasses on, so it's hard to read this. Increment the page at the end of each loop done with set node named increment page. So you don't even know what any of this means. Basically, let's say it needs to visit another site to get multiple information. And this could be a situation where, say you need to download multiple documents. This could also come from the StakeWords website too.

where let's say you to download the charging document and you need to download some discovery and you need to download whatever, right? This one where it could keep looping and looping and looping until it gets all the documents. That's something you could do here where, you've got, and all this is going on at the exact same time. And then at the end, what you could have it do is merge back together and then have it all combined to the file. And then you could do some analysis of it. And part of that analysis could be is, ⁓

based on the witness statements that we've taken ⁓ that are already in the file, let's say, ⁓ and the information that our client told us and the information we pulled from the charging documents. What are the charges? What are the elements of the crime? And what are our client's chances of success? And I'm just kind of making something up here, but that's the way you could do some of this stuff. Kind of a fun example of something you could do.

To me, that's a really fun one that you could play around with and do some really cool things when it comes to parallelization. And that's one that I pulled straight from the N8n website, the community website. And what's cool is they have lot of really neat ones that are free. They've been built by the community and you could just download them. this one in particular, I copied and pasted it.

into what I call the playground. It's really, really cool. And so that's why I like N8n so much. It's really neat. But there's so many out there. There's a lot of cool programs. It's only going to keep getting better and better. So that's what I have when it comes to an example. if you have anything you... Any questions about any of this, any examples that you want me to send over to you?

Happy to do so, just shoot me a text and I'm happy to do that. there's so many things that you can do right now. Don't try to overcomplicate it though. That's one of the major things I'm seeing people do. Start very simple. Start with the really mundane tasks. I gave you a little, some more complicated ones today. I wouldn't start with these if you're just starting out. Start with things that are a little more basic. to start by automating things first.

And I was talking to Jeremy Denelson and Chris Ragabi, I always say his last name wrong. So sorry, Chris, but Regaby, I think is how you say it. But we're talking today on the Friday accountability call in the guild and kind of how to understand some of this stuff. when it comes to your, not just understand, but how to implement it. Cause it can get, you can hear about all these things, but oh my gosh, I want to do all these big things. Don't start with the big stuff. That's where people get.

They really get steered wrong because they then get lost. get stuck and they end up not doing anything with it. Instead, just start with the mundane tasks that your people don't like to do anymore. Automate those if you can. And then next thing you know, okay, move on to the auto, the AI agent stuff. Then you just kind of build from there. Start at the bottom and not at the top. That's the big mistake. so, cause I know some of you might be saying, Tyson, this all sounds great, but

You how do I actually set this up? That's how you want to do it. You want to go from a very basic standpoint. Take pencil and paper, draw it out before you start building. So draw it out. That way you can see if there's any errors in your logic. So you kind of build it and you will make some mistakes. And this is if you end up building out of yourself. You don't have to build it out yourself, but I do recommend that you get an understanding of this because if you don't understand it...

then you're really not going to understand your business in 10 years. You're really not, because things are changing so rapidly. And if you let someone else go out and build it out, you're just not going to understand what's going to happen. And the other part of this is the people that have the technical expertise in this, that are building these out for us, they don't typically have the background that you do and understand how workflows are supposed to work. So luckily, I've got a Coshiff in the firm that I can rely on.

He's got an understanding of the firm because he's worked with us for so long. If you don't have something like that, you're going to have to work really closely with them to make sure that this is all set up carefully. So get started with this. If you've not set up something with prompt chaining first, start with prompt chaining and then move on to this. This way you can increase your efficiency, get really more comprehensive insights. You're going to able to scale a lot faster when you start implementing these. And so.

You're going to have a lot of fun. lot of it, I say the most important part is just have fun with us. It's, I think it's a lot of fun. I it's really interesting. Don't worry about getting lost on things. You're going to get lost. I think it's part of the process. Um, reminds me of the, whenever I first moved to St. Louis, it was, this was, you know, GPS is we did have them, but they weren't as, as huge of a deal as now. And they definitely weren't on the phones. And I, I went and got a map. That's what I did. I went and got a map.

And I got lost in the city on purpose and then drove around and made my way out of it. And then next thing you know, I knew all the roads and it was pretty nice. Same thing with this. Get lost, find your way out and you'll be okay. It's going to be fun. Okay. So that's all I have for today. Hopefully you got something out of this. I know I was talking to Jeremy earlier and I think some of this stuff, because I did kind of engross myself with it and really.

I dove really, really deep to understand it and immersed myself in it. And so if you feel like I'm skipping over things too quickly, let me know and I'll try to kind of slow things down a little bit. Cause I want to make sure that people have a full understanding of this. So that's all I have until next time. want you to just keep pushing the boundaries. Okay. Keep pushing the boundaries. Take your firm to the maximum. Right. I really want you to do that because.

There's a lot going on these days and I want you to push, push, push, push, push and get above that curve, in front of that curve. But as a reminder, before I close things out, if you want me to cover something on the Saturday show, shoot me a text 314-501-9260. Keep getting lots of great suggestions, so keep them coming. But until next week, remember that consistent action is the blueprint that turns your goals into reality. Take care.