Is hiring a full-time AI developer the next step in scaling a business?
In this episode of Entry & Exit, Stephen Olmon and Collin Trimble share why they hired a dedicated AI developer at Alarm Masters and how they're using AI agents to automate business operations, improve efficiency, and reduce costs.
They break down the three priorities behind their AI strategy: building secure infrastructure, managing AI costs with local language models, and developing custom AI agents that solve real business problems.
From automating billing and accounts receivable to improving customer service, lead generation, and technician support, this episode explores what it takes to move beyond basic AI tools and build automation that delivers measurable results.
Discover how businesses can use AI to scale operations, free up employee time, and create more efficient systems without replacing their teams.
━━━━━━━━━━━━━━━━━━━━
WHAT YOU'LL LEARN:
- Hiring an AI Developer: Why Alarm Masters invested in a full-time developer to accelerate its AI strategy.
- AI Infrastructure & Security: Building reliable automation systems while protecting business and customer data.
- Reducing AI Costs: Managing token expenses and using local AI models to reduce operating costs.
- AI Agents for Business: Automating billing, accounts receivable, customer communication, and sales workflows.
- The Future of Business Automation: Exploring voice AI agents, technician support, and human-in-the-loop systems.
━━━━━━━━━━━━━━━━━━━━
CONNECT WITH ENTRY & EXIT:
Website & Newsletter: https://entryandexit.co
Stephen Olmon: https://x.com/stephenolmon
Collin Trimble: https://x.com/TXAlarmGuy
Owned and Operated
New Episodes Every Wednesday!
Subscribe For More
For building some really fun stuff with AI. We hired a full-time AI developer. Kind of like having a GC and an architect together.
Great metaphor. The problem is that as we did that, it was not necessarily perfect or the most secure.
As you grow and you spend money on building AI agents, and people use those AI agents, you spend money on tokens. And tokens are not free.
My name is Steven Ullman, and I also have Colin Trimble with me. And we really like running this podcast. We're trying to be helpful. Hope that you enjoy it. If it's your first time, it's gonna be your best time. And uh we also run Alarm Masters, which is a fire and security business based in Texas. We've been growing it pretty fast, having a lot of fun. And today we are talking about our new pet hire. Really. Um pretty stoked about it. And uh Khan, why don't you share the background on a little bit of like how we got here, the choice that we made, and then we'll go into a couple of the core things that we're expecting to get out of the hire.
Yeah,
so if you've been on the podcast, you know we're building some really fun stuff with AI. And by building, I mean Steven and I are building. And so we're basically spending any amount of non-operational free time we have on building agents that do super cool stuff that we've talked about on this podcast ad nauseum. Um, and there's things that come along with that um that are require a lot of effort and time. Um, even basic things like debugging and like fixing tickets of things that break, uh, making sure you're doing it securely and in a way that's responsible, and also making sure you're like actually solving real business problems and it's just not a cool piece of technology. Um Correct. And so we realize like, hey, there's this window of time right now where most of the industry is behind with AI. They're using they're they're the best in the industry, they are advancing at the pace of which Claude Code is or Claude is advancing. So if Claude adds a new tool that's super cool, then they use that tool and that you know gives them a little edge in their business. And we are past that. We definitely use Claude, we're doing that, but we're building full autonomous agents that are doing really cool stuff. And so, um, in order for us to do that and to do that well, we want to capture that momentum. And we think that this kind of window is really 18 to 24 months of before all these things become packaged and become tools that every person is gonna use, like a CRM. It's just gonna be like common, like everybody's using a CRM or whatever. Um, and so we hired a full-time AI developer, which it's really funny at first when Steve was okay, I think we should do this. I was like a little protective and a little bit like, well, I don't know if I want to spend the money, and like I can do it, and it feels like a luxury. And then I was like, let's just take the phone call. Because at first I was like, well, like they don't know all the stuff that I can do in terms of like the business challenges, uh, et cetera, et cetera. So we jumped on this call and I was like blown away by, okay, yes, this is gonna take off a ton of stuff off my plate, like debugging and fixing stuff, but also just the infrastructure and the scalability of what we're trying to build is gonna be there, and also the creative solutioning to do some of the stuff better and more sophisticated. And so it was like a no-brainer. So we brought on an AI developer, and we're really trying to capture on this, and so I'm still doing stuff, but I'm really focusing on building the blueprint. So it's kind of like having a GC and an architect together. I'm the architect, and I know a lot about construction, and I know a lot about what I want the end product to look like, and I know it's possible. I know that I'm not gonna ask him to build me an upside-down house because I know that that is not possible. Uh, but I do know what is possible, and I kind of know how it can be framed out. Like this system needs to talk to this system, it needs a hook, it needs a fire, it needs a fire on this, and here's the people, and here's the system. Like, I I can see it, and I just need somebody to go build it, test it, and train it, and then debug it as we as we go. And so um, so I want to kind of dive into the sort of the three categories that this AI dev is gonna work on. We're gonna really spend as much time on the third category, which is the cool stuff they're building, but I think it is important for us to talk about the first two. You want to start off on what number one is?
I would love if someone reaches out and tells us they've done this, but any company in the life safety industry of our size, I don't think anyone has a full-time AI dev. Um, and and if they do, I'd love to talk and compare notes. Um but it's uh I'm I'm pretty excited that that you agreed with me. Yeah, thanks. I did.
Um okay, so the first part is we we built all these things and they're cool and they're actually being used. Some of them are kind of halfway, we haven't fully got them into production, that's fine. Ideating, right? The problem is that as we did that, it was not necessarily perfect or the most secure um or the most sophisticated of all time. And so, you know, day one, it's like, hey, you might want to be aware of these few very concerning things that you're gonna want to resolve. And it was just like, man, we A, it's expertise, and B, it's time, and like highest and best use. We talk about that a lot. Your highest and best use partially could be the vision behind some of those things, but executing is not your highest and best use along those lines. So, um, so yeah, that's the the first bucket is just like debugging, fixing, just kind of getting the infrastructure side um uh kind of a better foundation to operate from.
Yeah, no doubt. Um, I think it's really important to to like folks know that if you're gonna do a lot of this stuff, you've got to do it in a way that's secure both for your business and your customer, because it's touching customer data and touching your data. Um, and so I had like uh I would say that the things I were doing was ahead of most people that have my experience level. Like we had kill switches, we had audit trails, we had secure containment and sandboxes and all these things, right? But like there were still a lot of areas where it was insecure and it needed help. Also, uh, there was a lot of stuff that was just like I didn't have great backups of all of our agents. So like I would try to roll something out and it would break, and I was like, oh crap, I don't have a backup. And it's like, well, I didn't know. And there's like GitHub and all these other crazy developer tools that like back this stuff up. So bucket one is infrastructure, security, efficiency, make it run well, keep the error rate low. That's number one. Right.
Bucket two is um as you grow and you spend or as you grow and you spend money on building AI agents and people use those AI agents, you spend money on tokens. And tokens are not free. Uh, well, tokens from AI labs are not free, and they charge you for those. And on the whole, that expense is a lot less than um what your employees would be charging you in time, generally speaking. So, like if you were to hire just a you know a full-time person to go scrape publicly available web data and it's 40 hours a week and you're paying them $50,000 a year, you know, that's more expensive than the tokens required to go do that thing. But still, there's other ways. And so we've talked about this on the podcast a bunch. We've invested in um our own servers that are basically running local models. And local models are AI LLMs that people have made that are open source. Yay! People are really great, super cool that they do that, really powerful. And you can run those models, and they are probably 75, 65 to 75 percent as powerful as the mainline models, not the like most powerful Fable 5.1 or whatever, but like Sonnet. Yeah, it's probably like 65 to 75% as powerful as that. And most of what we're doing kind of fits into that box. So we also test our new AI develop dev with like, hey, we want to be able to take advantage of all the latest smarts of these AI models that are paid. We also want to be able to use these free models to do a bunch of this low-level stuff. So he's helping us build that out so that we can be responsible with how much money we spend on tokens. So that's bucket number two.
Yes.
We we love maximizing. Yes, we do. Bucket
number three. Bucket number three is building agents. And so this is really fun. I actually think this is maybe one of the coolest things is I was like, man, how do I communicate? Like, I how do I give him a blueprint? Like, what is a blueprint look like? And in what form should that be? Should that be a meeting and he takes notes? Should that be a document I type? So, what did I do? I used Claude, and Claude was like, hey, you should create an artifact. And an artifact is basically just like an interactive website that it just like spins up for you and it exists, and it's visual and text and interactive. And so I basically built every time I have a design idea for an agent, I build it into an artifact and I share it with our AI dev and he can kind of interact with it. And here's the thing: I'm not trying to tell him how to build it. That's not what I'm trying to tell him. I'm trying to tell him what I want, the systems it needs to be connected to, the security risks that are uh possible with it. And then most importantly, what business impact, what is the definition of success, right? So we're working on a billing agent right now. And this is to limit the number of emails and texts and whatever that has to go out to customers that are past due or on an invoice. And a lot of times, these we have great customers, we've got a lot on autobill, but a lot of commercial that want to cut us a check, and maybe they get a new billing person or they change their remit address or they switch from ramp to bill.com or whatever, right? And so it's just steps. So we have a whole team of people that are dedicated to like following up with customers and running down their questions about their invoice. And so we're trying to take all of that and build that. So then I said, hey, what is the definition of success? Well, it would be less number of phone calls and emails from direct agents, uh, faster turn time on like uh faster payment, lower age really on invoice is what it would be. Uh, and then lower outstanding receivables balance and specific groups of customers, right? Or in specific columns of revenue. And so he's building that as an agent for us, and it'll be fully autonomous. And he's going to A, build it responsibly, make sure it's done correctly. He's going to make sure it's safe and secure. He's going to test it, and he's going to actually help us train the team to be able to interact with it because it we do everything we do is human in the loop, meaning we humans are watching any of our AI stuff. And like we don't replace humans with AI here. We use AI to augment our humans. Um, and so he's building that's one example of an agent that he's building.
I'm really excited about some of the things that we're gonna do on the voice side. Uh, whether you know, I think that we can handle more simple tier one type of uh calls. I think we also will have an ability to put a voice to some of our agents. We we give personas um to our agents, and so you can already interact with some of them in Slack directly today. Um, if you're familiar with Victor, like which is a large business that's growing really fast, it's kind of a Slack-based agent. We just kind of have like almost like specialized or contextualized versions of Victor is kind of how I think about it. But I think we actually can get to a point where we put a voice to each persona where someone can just talk directly to each agent uh verbally. Imagine like a technician on site trying to get help or has a technical question. Like there's a lot of interesting use cases there that I'm excited to explore. Um, already doing some really cool stuff. I mean, really to your credit on the sales side and Legion side that we can take further. Yeah, also.
Yeah, and and I actually just built a Claude artifact that is here's where all of our agents are today. Here's what they're originally designed to do, here's what they're doing well, here's what they're not doing well, and here's what phase two for every agent looks like. Um I want to see that. I haven't seen that already. I just finished it. I'll send it to you. It's pretty slick. Thank you. It's very visual. I get very excited about visual artifacts. So, anyway, we're really excited about this hire, and um, we're probably the one of the only ones in the industry doing it. And um, if you are doing it, we would love to compare notes. Um, yeah, if you want to learn more about how to do AI stuff, we have a ton of episodes on AI. You should go listen to them, they're really fast. We also put out a newsletter with notes on those, it's just like short form. So if you're like a TLDR guy, you're like, I don't want to, or gal, and you don't want to like listen to the whole pod, we release a newsletter by these two guys. You're welcome. That's just a condensed version of that where you can go and you can read and see what we're talking about and um kind of get a more condensed version of what we're talking about. So uh and and Steve, what what else could they do if they want to like learn more about this stuff?
They could click the big red button. Uh, they could leave a really nice, kind, encouraging, uplifting comment that I could show my wife later. Yeah. Um, all sorts of she'd be like, who did you pay to post that for you?
Yeah.
Um, it's real, I promise. Um, yeah, you know, just just dive in, you know, jump in. The water's fine. So um thanks for listening. Hope this was interesting, and we'll keep sharing more as we make progress. Have a great week.
.png)
