Maximum Lawyer

Watch the YouTube version of this episode HERE


In this Maximum Lawyer episode, Tyson Mutrux sits down with longtime friend of the show, automation and AI expert Kelsey Bratcher of Hired Gun Solutions to talk about the future of intake, phone answering, and legal tech. The conversation starts from a simple Facebook post about hiring an AI answering service and turns into a brutally honest breakdown of why most voice AI products are expensive, over‑engineered, and still not better than a well‑built DIY solution.


Kelsey explains how tools like Retell let law firm owners build their own AI receptionist for a fraction of the cost of vendor setups, using the same documentation they already give Smith, Ruby, or Lex Reception. You’ll hear practical guidance on scope, latency, and call flow, why “simple beats fancy,” and how to use AI plus APIs to replace low‑value data‑entry work while protecting the human parts of client communication. Tyson and Kelsey also zoom out to the bigger picture: migrations between case management systems, why legal tech pricing is broken, and how AI will reshape non‑lawyer roles inside law firms.


What you’ll learn:
  • Why most AI answering services are overpriced and still average at intake.
  • How to use Retell to build your own AI receptionist with your existing scripts.
  • The importance of keeping scope narrow and latency low so calls feel natural.
  • When AI can replace virtual receptionists and when you still need a human.
  • Smart call‑flow tweaks (using caller ID, fewer confirmations) that boost conversion.
  • Where outbound AI is risky and when it works for expected, simple calls.
  • How Kelsey uses AI to build one‑off tools and migrations in hours, not days.
  • Why legal tech pricing is broken and which non‑lawyer roles are most exposed.

Highlights
  • 00:01 – How a Facebook post on AI receptionists sparked this episode.
  • 01:28 – Kelsey’s “build your own on Retell” philosophy and cost breakdown.
  • 04:27 – Retell explained: connect language models and voice APIs without coding.
  • 06:34 – Do AI receptionists lose leads? Why scope and consistency matter.
  • 10:34 – Latency: the real make‑or‑break factor for voice AI.
  • 12:45 – Fixing annoying call flows: stop over‑confirming names and emails.
  • 14:12 – Why Kelsey avoids outbound AI for provider calls and sensitive data.
  • 17:18 – Using AI to build tools that talk to APIs instead of “agents clicking around.”
  • 22:25 – Five‑minute Codex app that fixed a multi‑hour data‑entry mistake.
  • 26:07 – Inside a multiplaintiff lawsuit tool tied into major CRMs.
  • 29:19 – Faster, cheaper case‑management migrations using AI‑built scripts.
  • 31:33 – Moving away from Zapier/Make in favor of AI‑built micro‑apps.
  • 33:32 – “Legal tech is going to get rock and rolled” and why.
  • 36:29 – The flat‑fee AI trap and token consumption.
  • 39:18 – Which non‑lawyer roles AI eats first and which survive.
  • 44:37 – First 30‑day steps to test voice AI in your firm.

Connect with Kelsey
Website 
Instagram
Facebook


🎟️ Get your MaxLawCon tickets: maxlawcon.com
🔍 Vet your vendors: beccaslist.co


Maximum Lawyer helps law firm owners build businesses, not jobs.

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 (00:01)
All right, we are back with Kelsey Bratcher, longtime friend of the show, and he's the owner of Hired Gun Solution, where I think that this is gonna minimize your role. We're but you're an expert on automation, funnel systems, and processes, and now also AI. Kelsey, how you doing, buddy?
Kelsey Bratcher (00:19)
Mm-hmm.
I'm doing pretty good. I'm doing good. It's been a I think it's been a a while, Tyson. I think I've been I haven't been on since Jim left, I think is the last time that I was on.

Tyson Mutrux (00:29)
Yeah,
I think it's been a while. I know you and I, we've we've stayed in touch. We like text back and forth, which is sort of how this a bit of how this kind of started, where we wanted to have you on because there's there was a comment that you had left. So there's a post. I'm gonna read some of this for people so they have some context, and then we're gonna get into some deeper stuff when it comes to AI and what you might recommend, what you don't recommend. But so here was the post. It was from Jonathan Franco, and this was from April.

Kelsey Bratcher (00:46)
Sure.

Tyson Mutrux (00:56)
We are looking to h at hiring an AI answering service. Anyone have a recommendation for a company you love that has worked out really well? Have you found you were l losing leads compared to live human agents? That was a question. And then there was a comment, which I'm not gonna read the full comment. ⁓ and then you you posted a comment to a comment, I think, or no actually just a comment to the post, it looks like. “Build your own on Retell.” And it was funny because, ⁓
Becca saw this and instantly sent me a screenshot and said, “Hey, you think we should have Kelsey on?” I was like, “Yeah, absolutely. Let's let's have Kelsey on to talk about this.” So that's the setup. ⁓ I guess let's just start with in that moment you see it. I guess what was tell me your mind. What's going into your mind at that time and really now, whenever you see someone post about something like that?

Kelsey Bratcher (01:28)
Mm-hmm.
So the the main what what starts going through my mind with a lot of these stuff that's happening, the conversation around AI is there's a lot of people that want to buy, especially as it pertains to phone answering, right? And it's very expensive, right? I've had it I've had my own clients try it out. I've talked to other business owners that are messing around with various phone
related AI stuff and having different degrees of success depending on the the circumstances. But to get one set up for you that's supposed to be in an industry that is relatively standardized aside from maybe some mild customizations, ⁓ it's does the result is basically like you're paying, I think, five to ten thousand dollars to get one set up. And then you end up with this pretty sizable amount that would be, I would say it's competitively priced
to like your traditional virtual receptionist services, right? But it's really not delivering you anything better. And and it's I I guess it's like, it's just not worth it in my mind, considering how much easier it is to do this yourself, right? And when I say I used Retell as the example at the time
Because it's the one that I have the most familiarity with, and I would say has the lower, it's a lower technical barrier if you wanted to do it yourself. They also have services that you could pay for to get assistance with implementation that are pretty decent. But it's nothing like the level of effort that's required for everything, if you were to say everything in my firm, right? “I wanna I wanna build as this virtual receptionist,” that's ambitious. But like if you just wanted to replace your virtual receptionist.
This is easy and like relatively low tech. I mean there it is technical, it can be technical, but
It's all about like how well that like literally the same information that you would give to Smith or that you would give to any of these other Ruby receptionists, Lex Reception, all these other virtual answering services, that same kind of content can just be provided to Retell, doctored up a little bit, and then you have a phone answering tool that can answer questions and take messages, which is the minimum requirement, right?
And that's kind of where I'm starting from is like and you can get this done for like pay as you go. So literally setting it up and testing it for like less than fifteen bucks.
Tyson Mutrux (04:14)
Yeah, so will you do me a favor and 'cause there are some people that are not super familiar with some of these things. So explain when it comes to like just l let's say you're not even a basic AI user at this point. What how would you describe what AI what what Retell is?

Kelsey Bratcher (04:27)
Sure.
So, what Retell is, is it's a it's a a software service you can pay for that connects to various AI models. So you can choose if you prefer Chat GPT or GPT‑55 or whatever version that you want. There's a variety of different language models you can choose. And then you also have tie‑ins to a variety of voice ⁓
AI or voice APIs that you can use like ⁓ what's it ElevenLabs, there's a couple first‑party ones. I can't remember it kind of changes all the time. So but you have access to these and so it's
It lets you build a chat, make a prompt, right? Or a couple of prompts, depending on how sophisticated you want to get. And then this thing will integrate with already has the connection set up for you to use these APIs without having to like do all the programming and stuff that might be required. So the main things you're worrying about, well, while you do have to set it up, if you were to hire one of these companies, they're gonna ask you for.
all your process documentation and then convert it into AI prompt and then just toggle some settings for you that you could easily do yourself is kind of what I'm where I was coming from with that.

Tyson Mutrux (05:46)
Gotcha. So and for anyone that's listening to this, I did put up on the screen, I've got Retell's website up. And so it it does have have more detail detailed information if you want more information about it. I like you so voice AI is honestly something we've not, it's on our roadmap, but it's not really something we've dug a whole lot into. Another people like Jeremy Danielson, who's like he's built out his own, you know, call answering service where I mean he doesn't sell it, it's just his own thing where he—

Kelsey Bratcher (06:14)
Yeah.

Tyson Mutrux (06:15)
Like he he's had pretty good results from it. I I mean, I know a lot of people have been pretty h hesitant about adding AI to like the receptionist part of things. D I've have you seen much drop off when it comes to like leads and all that, where like the leads don't want to continue down the path because it's AI versus like a human?

Kelsey Bratcher (06:34)
It depends on the circumstances really. So like, ⁓ I'll give you an example. If you where where it's gonna fill in the gaps, I think it is better than a virtual receptionist. So like that's in my opinion, across the board of what I've seen.

Tyson Mutrux (06:50)
Why do you
say that? Because that's something people are going to be very skeptical. I I think I here's the thing. I believe you, but I just I know my initial reaction when you said that is like, no w no way. But like, why do you say that?

Kelsey Bratcher (06:59)
Yeah.
The reason I say that is if you have if your scope is constrained, me so the important thing is what is your scope, right?
You're not making a, I'm gonna make something, I'm gonna make it so that it's as if I answered the phone as the business owner and I know everything there is and I can answer and help anyone that calls with anything, right? That's not what we're trying to do. We're trying to minimally replace what somebody that would be a virtual receptionist would have about your business, know about the answers, the questions that you have, and they the AI will be able to be more consistent than what that person will be.
A hundred percent of I mean when I say that person, every single time you call in a virtual reception, you're getting a different agent, right?
And that agent is going to have like a little snapshot, 10‑15 seconds to familiarize themselves with your with your brand, your tone, the types of issues that they need to handle. ⁓ whether it's a sensitive issue or like an angry issue, like it that person needs to be extremely dynamic and they don't pay them enough to be that way. So like your AI can basically do the same effective job as that person without really any issue. With this, with the exception of if you start getting crazy with like all your requirements, which you... Would have the same problem with if you did a virtual receptionist as well, right? Like it's the number one complaint that I always hear is people are they're not good, who's better? And then someone has the same recommends their service, and then that someone else comes in and says, “I don't like them because of this reason.” It's the same problem, and that problem isn't going to exist in the same way with an AI doing this answering service, right? ⁓ like you're just simple stuff, like, ⁓ “What are you calling about?” ⁓ “Your personal injury situation, let me take
message forwarded along,” you can get crazier than that, right? You can I don't want to rant you, but like ⁓ start from the minimum, right? And the minimum is answering the phone properly, collecting information, taking a message, and then forwarding it along, right?
That is the minimum requirement of these services and that's where you can start is what I'm that's kind of where I'm saying. And like for doing that, it's just as good as anybody else. Now, do I think you should replace if you have your own reception or phone answering service in‑house, it will not be as good as that because that person can be coached, updated to the minute about what's going on in the firm. And so I don't necessarily say replace your entire suite with it unless you're feeling ballsy, but
Definitely a virtual reception service could be replaced with ⁓ this. And like I've seen it work, and the quality isn't I would say it's more consistent the consistency is the same. The one thing you have to watch out for that I think is important is is the more crap you try to get it to do, the you g increase the latency.
So it the conversation starts to become more robotic as opposed to free‑flowing. And then the other tip that's non‑technical is that when you ask it to confirm things like spelling, email addresses, ⁓ phone numbers, you need to be respectful of what the person is doing. Because like I'm not gonna confirm the way that my name is spelled more than once.
And I'm certainly not going to do it for robot, right? And I have a name when I my name's Kelsey Bratcher, and it's never spelled correctly. Both my first name and last name are never spelled correctly by people or AI, right? So those are the two biggest issues: is latency, and those can be that can be controlled through it'll tell you like any sorry, I go, I feel like I'm ranting, but when you use Retell, it's going to tell you what the latency is.

Tyson Mutrux (10:34)
I love it. It's good.

Kelsey Bratcher (10:39)
approximately for the settings that you've configured, right? So like in examples, if you use like a hardcore language model for answering simple questions about your business, that increases the latency. Right. Or if you're gonna have it do certain calculations like
filtering out background noise, these things you can toggle and kind of adjust for what you think is tolerable for latency. And that's really going to be the big thing is that when somebody answers the phone and it takes three seconds. You know when you get the robocall and you answer the phone and you're like, “Hello, hello,” and it's like 10 seconds, then all of a sudden somebody's like, “Hello,” and you're like, “I'm hanging up now because this person wasn't ready to actually talk to me.” This is the same effect that happens when somebody calls calls the AI, calls into an AI and then—

Tyson Mutrux (11:10)
Yeah.

Kelsey Bratcher (11:26)
starts talking to him, that's kind of one of the things that I've noticed.

Tyson Mutrux (11:29)
So probably then like the simpler the better so that you can decrease the latency. Cause I would say one of the most important factors is the latency, right? Like 'cause because of that one factor. It's not that it can do a billion different things, it's that it's it's quick.

Kelsey Bratcher (11:37)
Yeah.
Yeah, a hundred percent. And then when you're having it do things like n like I said, negotiating the spelling of a name, it's better to just not ask them how they spell their name than it is to go through all the bullshit of like step by step “K‑E‑L‑S‑E‑Y” because it's compressed, it's phone quality, right? So unless you're they're spelling it phonetically.
It's not gonna happen, right? It's gonna be a high likelihood of it happening, but it's not going to be guaranteed. And you're not gonna get that that's that's something you're not gonna get with a person doing it either, right? So really what is the most important thing when you when you're getting is making sure that the phone number that they can call you back on is correct, right? And that's usually provided by caller ID. So you're more or less using the caller ID.... as like the source number and then you just say, “Hey, this is the number that's is that what's best for you?” and then people just say yes instead of asking for the phone number and then confirm doing all this confirmation shit. And that's kind of like ⁓ those are like the I mean those are simple stuff. They're non‑technical things, but you experience them as a call if you called into this thing you would get pissed if you had to confirm your name and then your email address and then your spelling and then like all these other things. And that's kind of
That's more of like a good call flow rather than any technical issue that you'd have to overcome with the one of these tools.

Tyson Mutrux (13:05)
For listeners, I gotta say tech shit is a technical term, by the way. That is—

Kelsey Bratcher (13:08)
I I mean
I'm a cusser, so I apologize.

Tyson Mutrux (13:11)
You know me, I am too. I completely get it. I are you doing anything without going? Because I saw something with I got the opportunity of seeing what EvenUp Law is, or I think it's EvenUp Law is what they're doing. And it's really interesting. Like they're checking like provider balances and things like that on PI cases, doing outbound calling, which I I have my both reservations but also doubts about because I and they're saying that.
It's AI speaking to AI, which I do believe that to some extent. But with providers, I am very I'm I'm skeptical about the ability of AI to call a provider. Let's say it's a local chiropractor's office, say, “Hey, we're checking to get a balance of this.” You got the latency issues, everything else. They're going to be giving the social security number, the date of birth, all of that. I I am very I am skeptical about whether providers are going to release that information to an AI. But that being said,
Have you done much with outbound AI? 'Cause I I'm very intrigued by it.
Kelsey Bratcher (14:12)
I have not done anything like that with the outbound because I can see a lot of potential issues with like all that like sensitive PII being exchanged through now a third‑party channel that you don't necessarily have control over. and that like that's like HIPAA, right? Like actual HIPAA stuff. I have had some luck with when calls are expected, right? And the purpose of the call is known.
Then it does work well, right? Like if you filled out a form on the website and you checked a box that says, “Please call me immediately to schedule an appointment,” and then the AI calls them to schedule an appointment. That person's expecting the call, so they'll answer. The second thing is that they know why it's calling and they pro you could actually communicate that it's a scheduling AI to help you book your appointment. And then you narrow, like again, you're narrowing the scope of what that thing needs to be able to do rather than like
In the scenario you're describing, I think it would be relatively not impossible, but I think it would be difficult to make it be as dynamic as it would need to be to handle it effectively, right? so whenever you can make the call ⁓ the the nature of the call expected
and that the the timing of it is expected, that's when you'll have success. But I've heard I've had a lot of people like everyone's probably gotten a cold email like, “We have our AI prospecting system that's gonna call all your leads for you,” and like that's probably a waste of time. I mean
Just because like nobody wants to talk. It's like the the attitude is like, I don't want to read a blog post on your website that AI wrote, right? Like that's not I'm not gonna do that. I don't read people's websites anymore because of that shit. Like I'm gonna use ⁓ whatever AI, I'm gonna use Claude or I'm gonna use Chat GPT or or whatever to do the equivalent because I if I'm gonna read AI garbage, it's because I queried it and not because somebody else tried to pass it off or be cute with it, right?

Tyson Mutrux (16:12)
Yeah, I 'cause so you you've mentioned ⁓ actually keeping things simple, starting simple a few times. That's always been my advice to people. But here is help me with this. Help me help me with balancing out this issue. I like to go into have like an overall plan for all of our systems. And so the problem is sometimes where you start with a bunch of small things with AI. And this I I was talking to somebody just just ⁓ yester what's today's Wednesday on Monday.... Where it was during ⁓ our little huddle that we have in the the association where we we talk about what our focus is for the week and everything. And he was saying how and I'm not gonna say who it was because he doesn't give me the permission, but he was saying he had created one AI to do this thing, and then his other AI was canceling it out because he had forgot he had created the earlier AI. And so I I think that it's sound advice to start simple with these things, but
Kelsey Bratcher (16:52)
Sure.

Tyson Mutrux (17:07)
when you have like the overall plan, how do you balance out your overall plan for all your systems with also like being simple with all of these multiple things that the AI is doing?

Kelsey Bratcher (17:18)
That's it. So for me, I find myself not I use it as a way of like
I don't use agents to do a lot of stuff, right? Meaning, like, I'm not gonna go and get Co‑pilot or whatever and put it on my computer and like tell it to go do stuff for me because it's an inefficient way to do it, right? Instead, you're gonna build like what I've been doing is think of it more like software, right? Especially because rather than having an AI with a skill that can go log into my Gmail account and look at my emails and then choose what to do and organize them for me correctly, there's an API for Gmail, right? And the agent can use.
use the API when it needs to do stuff. So all the heavy lifting is done through these existing APIs rather than this thing figuring out how to do it the hard way. Right. And I'm using a kind of a I'm using a real example that I've heard people talk about where the AI like opens their browser and then clicks on shit as opposed to like just connecting to it through the command line and then using an API to do all the same exact thing. So rather than having the agent do all the work, the agent
references a tool that it that you could have built for it, right? Or that already exists as an interface. But like if you're going to be building stuff like that, you need you're gonna start off solving a simple problem, but then like you said, you've got five simple problems and they're starting to clash. So you need to start thinking about things in a bigger way, which leads to making software realistically.

Tyson Mutrux (18:48)
We I I have c I have a question about that. Before we get into that though, would you explain the difference between API and MCP with the with the ⁓ the 'cause I think that that could be a really confusing thing for people.

Kelsey Bratcher (18:50)
Yeah.
So APIs have existed forever.
⁓ and what they are is it's a an interface that most software companies in the current reality provide as a way for their customers, partners, and developers to build and integrate their software and or build add‑ons for the product that you're using, right? And everybody in here probably has somebody, some their practice management software has an API of some kind that integrates with something. ⁓ and that's basically like the the beginnings of it, right? And this interface exists.
Oftentimes MCP servers are what give a common language to these different language models to interact with those APIs. So it already stores the information. So when you're being given an MCP server, ⁓ the AI is interacting with the MCP server to basically talk to these APIs for you. And so it's kind of like a middleman for those types of things. I find like with the stuff that I'm doing, I don't use MCP servers as often as you would expect because I have like I use Codex over Claude.
Claude Code. It's my preference because I save a shitload of money by just having their max account and I can I've never run out of tokens. We'll just say that. Right? and you burn like 500 million tokens a day plus using this product, and you try to do that on in the equivalent on Claude and Claude, and you and you blow out your thresholds, you start charging more money. This has been my experience.
But your command line on your computer can act as the same thing, right? It's a as if you were to take a what a command line interface is, is if you open a terminal window on your computer, this might be a little technical Mac users, it's called Terminal, right? And it lets you put in commands that will do things, right? Well, it's hard for you as a user to remember all the commands and parameters and the ways that these things need to work, but it works exceptionally well with AI because it knows the language and the structure of all this stuff and what
needs to do and so it's it's easier it's faster it's on your local machine you don't have to have something hosted somewhere or running on your machine constantly because your terminal window is there at all times and your most of these programs or Codex and Claude Code can interact with your terminal to do things on your machine that then give it information and that's how ⁓ most of this stuff can work especially if you're using if you're using these things as an agent right the other thing that I find myself...
Doing is making little apps that you can run on your own computer, right? So instead of having like an artifact in like Claude when you're doing something to make like a little cool little interface or a form or whatever people use it for, you can just have it build the software that you can then open an HTML file on your computer or go to a web hosting, you can go to an address, localhost, and then it's running on your computer. And
you don't have to do any you have to have some technical awareness of what these things are and I'm probably s speaking a little bit of Greek to everybody but this is free. Like this not none of this shit costs money. It's very quick. Like I'll give you a a really good example. The other day I fucked something up and I swapped an ID for a contact for an ID for like what's a ⁓ let's just call it a a project, right? So the project ID was a contact ID and vice versa and because there were numbers

Tyson Mutrux (22:20)
Yeah.

Kelsey Bratcher (22:25)
It was able to actually reference context of that ID that should have been the project and and like there was like a day where this was happening. So all this shit was all these notes were all on the wrong stuff. So I used Codex. I said, “Hey, I have this problem. I'm using product this tool. I swapped these two things. I had the foresight to drop the correct IDs in the note body itself. So then I had download the get all the the range for the two days, find every note that has this pattern in it, extract the IDs, and then
Make me a little piece of software that shows me all the contacts and projects that it should, that it thinks that there it currently has and what it should be, and then give me the ability to test it and do like a dry run and then up and then once I was satisfied, it would just go fix it for me.” It took five minutes to make that. If I had tried to do it myself, it would have taken if I would have I wouldn't have been able to make that by myself, first of all. But if I had to fix all that shit by hand, it would have taken me f seven or eight hours. Because we're talking like probably a couple hundred notes that needed to be made.
moved to something else. And like this is just the software a one time use piece of software that I created that ran on my computer locally and took care of the problem. It it connected to the API of the tool that we were using and it found all the records and fixed them all for me and then gave me a little report that said everything's great and everything was everything was fine, right? ⁓ and that's kind of what I'm kind of the direction that I go with how I've been using A AI is to make these little tools. And they're getting
bloated some I mean I'm building a practice management software now for somebody and it's going to be intense but like you have to keep track of all your stuff somewhere and that's kind of where like yeah maybe everyone's heard of these brain apps or like you make your AI brain. if you're using Codex it kind of already has one so you don't really have to worry about that as much but like if you're making software you should probably learn how to use Git which is

Kelsey Bratcher (24:23)
GitHub and like when you have like let's say a a feature idea that you want to create, right? You tell Codex, “Hey,” or or Claude, “Hey, go create an issue in GitHub for this,” and like it's like “I need to change the buttons from red to blue.” And then it will build a little use case for it and then keep track of all the shit, your backlog, if that makes sense. And
I guess what I'm r I feel like I'm rambling, but one thing I will say is that it already knows this software development language. And if you want to understand it, then you can just ask whatever AI that you're using how to do the thing. And it can explain it to you at the very basicest of levels, all the way to like you wanted to change career paths, if that makes sense....

Tyson Mutrux (25:11)
Yeah, and I for anyone that's ⁓ listening, I'm I'm actually pulled up GitHub on ⁓ the screen. So anybody has has questions about what GitP GitHub is, it's it's I mean it's a must if you're gonna be doing this. But ⁓ I I will tell you, I I get a lot of pleasure out of hearing you say that you switched the client ID and the project ID because I remember years ago at the Zapathon, I I'd built out this Zap that I thought was amazing, and you had to show me that I I was using the client ID and I was supposed to be using the project ID.
So it just makes me gives me a lot of satisfaction to know that even you screw that up sometimes. So that's that's fantastic. Okay, so I w I I I have I'm gonna back this up just a little bit. When it comes to building a software on your computer, are you using Codex to build that 'cause I have heard that Codex is better than Claude Code when it comes to code. So are you using Codex to build that software?

Kelsey Bratcher (25:44)
Yeah, yeah.
Sure.
Yeah, so I'll pull up Codex and like I can actually if I can share my screen I can show you something I'm running on right like it's actively doing shit right now.

Tyson Mutrux (26:07)
Yeah, yeah.
'Cause I think this
would be ⁓ for people to see would be amazing as to how like from a practical standpoint had to 'cause you did speak a lot of technical jargon there, but I think it is helpful for people to see it in real ⁓ real life.

Kelsey Bratcher (26:15)
Let's see.
so this is ⁓ an example of what Codex is, right? So right now I'm building a multiplaintive lawsuit tool for ⁓ Jim Hacking. I still work with Jim on the regular. You can see that this is like hosted on my computer right here. But this is doing all the work, right? This is all the shit that it's doing right now. ⁓ I'm using goals, which is ⁓
Basically you can go, I mean, this is my prompt. This is gonna be a little embarrassing. But like it was doing. “So I need you to stop fucking around the prototype I provided you.” So like just to kind of get I mean that's how casual I am with it a little bit, probably too much, but you can see stuff's happening in real time. ⁓ on this, this is on my computer, but it's doing this basically going through we had a what happened was we needed to do a multiplane of lawsuit for immigration related stuff.

Tyson Mutrux (26:56)
Yeah.

Kelsey Bratcher (27:17)
⁓ time was of the essence, so we needed to create like an intake form that would then do a bunch of stuff that Jim had outlined for what would be required to do this mul multi‑plaintiff lawsuit. So we built the tool and integrated it with Salesforce and HubSpot and
A PandaDoc for document signing, it was pretty elaborate, but it was like not intended to be like a long lifespan tool because we don't know we were gonna do more multiplane of lawsuits, and we also needed it in like a week. So we originally built this and then this is the second version of it that's being rebuilt, and so I actually used Claude Design, which is their design software.
⁓ it's similar to Figma. I can show you that in a sec, but basically that takes care of the user interface portion of it. So like as a normal person, you can do this without having any technical ability. And then that creates like the basic framework for how it should look. And you can convert that into like you can put it into Codex Project or Cup Claude Code or whatever you want to use, and then it'll build that software for you.
with some guidance, right? And you can use agents and other shit like that. And however sophisticated or crazy you wanna be, ⁓ you can get. And so that whole—

Tyson Mutrux (28:34)
Yeah, so we used so this is a a really practical one too, where we were migrating from Zoho to our case management system that we built. And Kashv built like a one time use software that it it migrated all of our files over and it was I mean it it documents and everything. Like so like if you if you like if you're on I don't know if Filevine is still, you know, their migration takes months still or I I can't imagine it does, but if it does, you could—

Kelsey Bratcher (28:52)
Yeah.

Tyson Mutrux (29:02)
If any c case management system, not not to pick on Filevine, but the any case management system, if it's if they're telling you it's gonna be months to migrate, then they're they're crazy 'cause it you can build this stuff and get the migration done in a matter of days at most. ⁓ it's it can happen pretty quickly. It's kinda crazy....

Kelsey Bratcher (29:19)
You could you could do it in real time now, if you were to be if you were depending on your objective, like let's hypothetically say you know the main ones Clio, Filevine, what is there, CasePeer, ⁓ they have to have an API. So I'm gonna just use Filevine and Clio. Let's say you're going from Filevine to Clio or from Clio to Filevine, you have a bunch of fields and contacts and
stuff like that that you need to move back and forth. while you're still using it, you can make it you could theoretically make a thing where like in Clio if you change someone's email address, they would update our tool and then that would move to Filevine so that you can kind of sync them. And then as you're ready to migrate over, you can just press go and then dump all the data into like all your files could be downloaded on your computer and then uploaded to Filevine or vice versa, right? This is

Tyson Mutrux (30:11)
That's exactly what we did. Yep.

Kelsey Bratcher (30:13)
There's more involved, it's not as easy as that, but like it's not as hard as it used to be, which would have been hiring somebody to write a script. Like if you wanted to leave Filevine, for example, they give you access to an S3 bucket, which is basically Amazon's solution for storing files, which is what they use as Amazon. And ⁓ you have to make a script that will go and download every single file and recreate the folder hierarchy by product by project ID.
Otherwise, you're not gonna retain the folder structure that you had before. And so ⁓ that would take days, right? Waiting for someone to make the script, testing it, and then you're downloading files for days, depending on how you have three, four, or five terabytes or more data on your Filevine account or or Clio or anywhere, it's gonna take a long time to transfer all that f all that shit. And so
It could be a person doing it or you could have AI babysit it for you. With a little guidance, you could create these tools, these one‑time use tools, and basically have that done for less than you would have hired a data migration service for, which is I mean, for something like that, twenty‑thirty thousand dollars wouldn't be un I'm not you're not gonna get quoted that, but it you it wouldn't be surprised to hear that.

Tyson Mutrux (31:27)
Not not at all. I I wonder how much you're using like the N8Ns of the world, the Makes of the world, things like that these days.

Kelsey Bratcher (31:33)
As little as I possibly can now. Because the issue is that they're kind of like acting as a middleman where you can make agents and these tools. But I can just do that on my computer. And if I need it to be online, then I can just host it on
Very inexpensive services. Like there's a if you just go if you were to say, like, “Hey, I made a program using Codex or Claude Code, where can I host it?” It'll give you six recommendations. You have like Railway, Render, ⁓ Vercel. ⁓ you could host on Amazon AWS if you feeling like you want to get into SysOps, Google Cloud, you can do Microsoft Azure, you could do
what is their other one? ⁓ DigitalOcean is another one. And like these are all like hosting services that give you the ability to host databases and web applications and all kinds of other stuff. And depending on what the thing makes, we'll make that available to you. And you're gonna get this at a very, very discounted rate. So like a lot of the stuff like a typical Zapier user, you're gonna be paying like I don't know, maybe up to a hundred bucks a month for 4,000, 5,000 tasks.
And then you're gonna pay a premium on top of that to use it with AI, which then also has token cost, right? So it's like, why would I go through all that trouble if I could just have the AI make that little serverless application float in my own thing that I could control and use, but basically for free almost. I mean, immeasurably inexpensive is how much it costs. So I I refrain I use them because I have a bunch of shit that I built for people over the years that need to be maintained.

Tyson Mutrux (33:02)
Can I read?...

Kelsey Bratcher (33:09)
And if I could push a button to just swap all that out to my stuff, that I would do that immediately because it it's actually when I get, ⁓ “shit, there's something broken in Make,” I have to go fix that because AI can't do that quite as well as I'd like to now. And that can take time that I don't want to spend doing that anymore, because if I had built it with AI, I wouldn't have given it a prompt, “Hey fuck you, fix it.”

Tyson Mutrux (33:32)
Well, that's the cool thing about like building things is like you can you can edit them too. You can change them if they're not doing like if things change you can you can edit and make it something different. But can I can I read ⁓ a text exchange that you I actually is your your text to me. Can I read it? I and if we if if you don't like it Okay. I I because I think this it's kind of funny. ⁓ but I it's gets to the heart of kind of what we're talking about here.
Kelsey Bratcher (33:53)
Yeah, yeah, you do whatever you want.

Tyson Mutrux (34:02)
So ⁓ “legal tech is going”—this is your your text to me. “Legal tech is going to get rock and rolled over the next two years. The apps are total dog shit. They are overpriced and wildly abuse their customers.” Like this is one of the major reasons why we started BeccasList.co so people could find vendors that worked for them because legal tech does a lot of vendors do try to take advantage of of law firm owners all the time. It's kind of crazy. But ⁓ I I guess my question is for you is like I am very surprised.
By all of the money that has flooded into legal tech over the last couple of years, i it does not make any sense to me because of what we're talking about. Is there any rational explanation that you can think of as to why so much money continues to flood into the legal space right now for legal tech?

Kelsey Bratcher (34:47)
I don't know why money would continue to flood into it ⁓ unless somebody's trying to become the de facto player. But like if you look at like what Clio and Filevine and they're spending a shitload of money trying to make their little AI tools, like and give them cute names like ⁓ like what is it, Lois for Filevine and I I can't remember what they've called it's been four or five different things. But like the kind of stuff that you can do with AI that these companies have provided you with, this is like baby level stuff and you're paying
extreme amounts for it comparatively, right? But if you go into like why someone would make a new product, right? I guess if you I'll say this the legacy products are being trying to adapt to the new world of AI, right? The there's new players flooding into it because they're using AI to build their apps, they're using AI like first, right? They're thinking about, okay, well if I made a tool that ingested documents,
And it like created metadata store for the AI to be able to use as like a to create something like a demand letter, right? And reference whatever bills and medical records that you might want to reference. Like the only thing that is ⁓ let's just pretend like there's no HIPAA, right? Just for a second. They're not doing anything that you wouldn't just be able to do with a Claude account.
Right. With a clever use of of like Claude or or GPT‑5 or whatever. Like again, there's data privacy concerns, but like they're probably like there's just not anything special that they're doing and they're gonna mark it up. The other thing that's interesting, if if you remember back in the day, there was a certain company that gave you text message, one phone number per project, right?

Tyson Mutrux (36:28)
Yes.
Yeah.

Kelsey Bratcher (36:29)
That you
could use to text the client and then they would text in, it would always go to the same place. You got a thousand cases, you had a thousand phone numbers tied to your to your account. ⁓ we all know how that ended because text messaging is a consumable metered service, and you can't charge a flat fee for a metered service because somebody's gonna blow you out. And a lot of these legal services are charging you a flat fee for you know $350 per project or $350 dollars per case.
You're going to you can consume way more tokens than two hundred and fifty dollars if you were to just accidentally prompt it, right?
⁓ so you're basically paying a flat fee for the lifetime of a case, $250. They think they're cute and they're building some margin in there to some extent because they've created some formula. But the reality of the situation is that they haven't and they should be selling it as a metered service rather than a flat fee. Because even if they have their own infrastructure, like let's say they built a fucking data center themselves, it still costs electricity....

Tyson Mutrux (37:30)
Which they have not, but if
they did, but—

Kelsey Bratcher (37:32)
Yeah, it still
costs electricity. They of course they haven't, dude. They're a scrappy startup trying to get into legal tech. They don't have real money. Like they might have a couple mil to get started, but they're not data center money.
And that so that data center is gonna cost money. Any hardware that they purchase is going to cost money. So there's no way to sell a flat fee on that because they will not be able to predict the future. So anytime I see pricing like that, I'm like, what the f there's no fucking way. Like I don't care how much margin they packed into it. I'm more insulted that they tried to pack margin rather than just sell it as a metered service, which is what it should be sold as.

Tyson Mutrux (38:08)
Yeah, and I think I do think the average person doesn't look at it that way. They they're not but with the history with what you're talking about, the company you're talking about, like they had to they had to kind of r kind of unravel all of that and they had to get away because it wasn't a what from a long term standpoint, it just was not sustainable. So, ⁓ and those of you that have been on that were on that service, you know exactly what we're talking about. But the ⁓ so let's let's ju kind of start to land this plane 'cause I
I I feel like you and I could just talk for days about all of these different things. ⁓ but I let me get into a little stickier thing. I'm curious when it comes and this is a little bit out of your wheelhouse, but it's it's because of it's because of what's in your wheelhouse. I think that non‑lawyer employees have a serious problem coming and it's that they can be replaced by the vast majority of the tools out there now.
What do you do you have any concerns about that? Do you think that we're at the point where the tools are good enough where they can start to replace full time employees? I guess what's your perspective on that?

Kelsey Bratcher (39:18)
So I do this stuff myself. I'm I do more or less glorified freelancing and consulting, right? That's how I've done it. I've never really hired anybody because, well, in the past I've had trauma as a result of that, right? That I have not been able to work through. So I am not ever going to hire anybody ever again. Because anything that I'm going to be doing, I'm going to do in such a way that I can use AI to do it. Right. because the way I look at it is I don't need to, right? There's no reason to do that if you can get ahead of it. Now
Are you gonna have who are you who's gonna interface with your clients? Right? That's a qu I don't think that's gonna be replaced by AI anytime soon because you have a lot of people that don't wanna talk to AI. They don't wanna interact with them, they don't wanna get sent, you know, but but like who replies to your text messages? Why does it have to be a person? Especially if you were to engineer a system that has access to the same shit a person could have access to to get the answer that's pretty close to what it is.
And if it's outside of your garbage like let's just say you have a a text phone number where a client texts in and asks about the status of their case, right? Why does it require a person to do that? So—

Tyson Mutrux (40:22)
I'll tell you why
you know what I'll tell you right now why? Because people people don't the attorneys don't want the clients to know actually what's going on with their case 'cause they may not have touched it for touch it for.

Kelsey Bratcher (40:31)
Okay, well fair. Yeah, I mean that might
be the thing, but I mean you could guard it, you could guard it up so that it is very vague, very you know, elusive answers, and then if they really pressed, then you would say, “All right, well give us a call and we'll talk to you.” But like going through doc like I but there's like if you had a document dump, right? And again, privacy aside, let's pretend for a second that it's not an issue.

Tyson Mutrux (40:37)
Right....

Kelsey Bratcher (40:52)
If you got a document dump of a thousand files, would you want to pay someone hourly to do go through all 1,000 of those files by hand and like classify them, understand what's extract detail from them that's important, and then also be able to trust that they did it perfectly? The answer, you wouldn't, right? So that person doesn't need to exist. That job doesn't need to exist because it can be built into an AI that can or you can build a product or service that will

Tyson Mutrux (41:09)
Nope, not at all.

Kelsey Bratcher (41:20)
take documents and understand them and and compile them together in a useful way, right? ⁓ and that doesn't require a huge amount of effort to do that. So like I guess what I'm trying to say is I do can I have concerns about anybody doing knowledge work, basically. But like for sure things like data entry or like what in in the past like people that use virtual assistants for doing certain types of grinding that's unpleasant, that's inexpensive labor, that is not something that's necessary anymore.
Because there's nothing any anything that anybody's doing remotely is at risk right now. I mean to—

Tyson Mutrux (41:55)
I agree. When it comes to
let's kind of circle it back and then we'll wrap things up. I wanna make sure people ⁓ can reach out to you as well. What is ⁓ what what's the best way for people to reach out to you? Hired Gun Solutions, Kelsey Bratcher, how do they get in touch with you if they if they wanna work—

Kelsey Bratcher (42:10)
I would say
for me on Facebook is probably the best way right now. I mean my to be transparent, my website isn't even up, right? So it's it's it's not something that I see the need for anymore. ⁓ because like it's
I don't well for my business I don't necessarily need it, right? ⁓ you can l LinkedIn would be another one. Kelsey Kelsey Bratcher on LinkedIn. That's basically where people are getting in touch with me now. I'm a I am kinda guarded. I don't have
a huge amount of availability at the moment. I mean it I'm in a transitionary position as well because I find myself building more and more of these tools that aren't going to require what I used to do for for a business, right? So I've had to shift and evolve how I'm doing things where like I can envision a world in the near future where ⁓ there's just software being built for you behind the scenes by somebody that understands what the software needs to be and do. And
it's catered to your specific situation. Now you could do this yourself, but like if you needed help with it, that's kind of this kind of how I've been I mean I'm building like ten different apps right now for doing all kinds of different things. And they're like I s like the example I showed you that Hacking
Jim Hacking stuff we're doing is multi‑plaintiff lawsuit tool that's gonna handle the whole thing from like signing people up all the way to like delivering the information that he needs to create the complaint or whatever he's gonna submit, right? and like we're doing that to get ready in like a week or two from now. If like that's like the the time horizon on this.

Tyson Mutrux (43:46)
I don't I don't have Jeff's ⁓ Jim's p ⁓ text pull up, but it was funny. he wanna he and I were texting a while ago about it's you've opened his eyes to AI. Cause like he was he was very resistant early on to it. And he's it's really whatever you're working with him on, it's it's really opened his eyes. It's kind of crazy. I I think we're we might be entering this era where the CTO's the new CEO, where the the CTO is the one that's kind of like the really
The one that it holds all the cards, all the power. I think that that's that may be where we're where we're heading. I don't know. But let's just to wrap things up, if when it comes to because we started when talk when it comes to voice AI, if people were gonna start with voice AI, you're talking about starting small. What if they gonna start in like the next 30 days, I guess I guess what's the starting point? How would they do it?

Kelsey Bratcher (44:37)
Well, if you're not wanting to purchase one of these services because it's either not within your budget or because you have the same concerns that I have, which is it's not as hard as I think, right? ⁓ I would say just get a go to get Retell AI and try it out. Mess around. It doesn't cost you anything.
Like you can get access to the tool. Sorry, I take that back. It does cost you a very little bit of money. But like ⁓ when I say you put a credit card on file and it will charge you for those tokens, but we'll just call it twenty bucks, and you can get enough to be able to feel your way around it and make a decision about it whether or not it's good enough for you or not. Now you can extend this thing pretty badass, like as an example, like if you were inclined, you can make it so that when somebody calls in and you wanted to go look up that person in your practice management software by their phone number, it could do that.
retrieve information that the then the voice bot could use to some extent right to answer questions or to be able to know that “I don't need to ask this person if they're a returning or if they're an existing client or not because I just queried the practice management software and they have an active matter or project or whatever.” So you skip all that bullshit right and like you're not going to get that with a virtual receptionist and you're probably not going to really be able to get that from a normal person because this could happen before they the AI bot even answers the
phone, it'll have information from the caller ID. And then like right in the middle of the greeting, it's querying and getting that information to be used in the call. And like you can be guarded with it. I mean I'm I'm explaining like the next level of what it could be, but this isn't out of reach. Like this is all within like a normal, a normal realm of like what I think is possible for a person that's inclined to do so.

Tyson Mutrux (46:25)
man. As always, Kelsey Bratcher, the man, the myth legend, you you delivered. I really appreciate you doing this, man. Love it.

Kelsey Bratcher (46:31)
Yeah, no worries. No worries.