From ChatGPT to AI Agents: How We're Using AI to Transform Our Business

Most business owners are still using AI like a smarter search engine. They're missing where the real leverage begins.

In this episode of Entry & Exit, Stephen Olmon and Collin Trimble pull back the curtain on their AI journey, from using ChatGPT to clean up emails to building AI agents that automate complex workflows across their business. They explain the mindset shift that changed everything, why treating AI like a digital employee unlocks entirely new possibilities, and how consolidating data inside Salesforce became the foundation for meaningful automation.

They also break down the difference between simple automations and true AI agents, share how they built their own open-source AI stack without a software engineering background, and discuss the practical tools they're using today to create operational leverage.

Whether you're just getting started with AI or looking to move beyond basic prompts, this episode offers a practical look at what's possible today.

In this episode:
→ How Stephen and Collin went from AI skeptics to daily users
→ Why AI should be treated like a digital employee—not a chatbot
→ The importance of centralizing business data before adopting AI
→ The difference between automation and agentic AI
→ Building AI agents without a software engineering background
→ The open-source tools powering their AI workflows
→ How AI is helping automate sales, operations, and data analysis
→ Why the best time to start experimenting with AI is now

Connect:

Stephen Olmon — https://x.com/stephenolmon

Collin Trimble — https://x.com/TXAlarmGuy

Owned and Operated

New Episodes Every Wednesday!

Subscribe For More

Transcript

AI As Digital Employee

SPEAKER_01 0:00

I could give real responsibility than what I used to engage AI with. This could actually change the value of our company someday.

SPEAKER_02 0:06

Well, how do I get around that? And I was like, I could just build you a script right now. I was like, well, what does that mean? And it said, go click through these things and get these API credentials, and I'll build a little MCP server script, and I'll do it.

SPEAKER_01 0:20

Like junior level team member.

SPEAKER_02 0:23

And sometimes it would try to go off the guardrails and do stuff that I did not want it to do. It's like, oh, you want me to delete all these records? Like, I do not want you to delete all these records.

SPEAKER_01 0:32

We didn't have a significant amount of prior experience that would lead you to believe that we could be where we are five, six months later. Yeah. The best time to start was two years ago. The next best time is today. That's right. Welcome

Show Intro And Early AI Use

SPEAKER_01 0:49

to Entry and Exit. My name is Steven Ullman. I also have Colin Trimble with me. And today we're talking about our journey with AI so far. We're going to talk about the beginnings, uh, kind of how we started to do some testing and now like kind of all the crazy stuff that we're doing and kind of share some of that little behind the scenes. A little inside baseball. Why is it inside baseball? I don't know. I like football more. I don't know. Inside football. Yeah, inside football. Inside Texas AM Aggie's football. Yeah. So I digress. Um, so I whenever we talk about this, I feel like you start with me and perplexity, but that's not true. Like we started using Chat GPT in the business well over a year ago.

SPEAKER_02 1:39

Yeah, I know. That's but but we weren't AI pilled. Which by the way, I think it would be really fun as an exercise to like, which we don't have time for, it's like, what should the clickbait title of this episode be? Like, how to get AI Pilled Today? You know, I think it would be really fun to to workshop some titles, even though we have no input on the titles. Our producers tell us what the titles are gonna be.

SPEAKER_01 1:59

So cheers, producers.

SPEAKER_02 2:01

Yeah. So here's uh we get a lot of questions about how did y'all get to this point with AI. And I think it's really important. And I wanna I wanna share this story. We are fortunate for a couple of reasons. Some of it has been luck, and some of it has been because we're brilliant, mostly luck. Um where the power of our AI was amplified because of decisions we made before we really got AI pilled. Um what we were using previously for AI, and we're gonna come back to that, but what we were using for AI previously was what probably everyone else sitting in their car or at their house or at their office listening to this was doing is they had a subscription to ChatGPT and they would ask it questions to help them clean up emails or solve a problem or look at a legal doc or whatever. And that was the extent of it. And maybe some power users created what's called a custom GPT or a project within Claude that you can like give it some ground truth and some instructions, and it'll like answer questions about um how to, you know, work on certain alarms or like tier one tech support. We heard some people doing that, which is actually a great use case, by the way, to start with. So we were doing a lot of that, and that was sort of the extent to what I thought was really available to us and out there. And what what happened was Steven uh had I don't know who you talked to, but you should share like two minutes on like how you really got AI pill because you you got me kind of down this rabbit hole, and I was really hardcore rolling my eyes. Like you were texting me and sending me articles and ex posts and stuff, and I was like, I don't care about any of this at all. Yep. But t tell us like your two-minute version on that.

Getting AI Pilled

SPEAKER_01 3:51

I saw people that I really respected in business publicly sharing that they were blown away by kind of the bleeding edge. Like what the the models really it started in December of 25. There was a whole new set of models across like three or four of the main um like labs or like frontier models, and um that next iteration was such a huge leap forward. And so around Christmas time, I start kind of playing with a couple things, but I I felt like the practical application wasn't there for me yet. And then in January, I started seeing some really clear examples of the difference, and so it's like, oh, there's been a large jump in like the very first iPhone to like the iPhone you have today, like there is a very big difference in processing power and what it's truly capable of. And it seemed like that, like, oh, there's been a large jump. This isn't like ChatGPT when it's like edit my Word document, you know? Yeah, and there are uh like process and workflow elements. And if you're familiar with like Zapier, not just like a little automation connecting two things, but uh a whole new set of capabilities. Yeah, and there were clear examples in a couple businesses where I was like, okay, I remember one person posted something that they accomplished for their real estate business, and I was like, that seems made up. I'm not really sure that's real. It was, and so then I started to use a couple new tools, and within 72 hours was AI pilled. I was like, oh, I actually can treat this, and this is the phrase that you and I use some is like a digital employee. Yeah. Like it started to feel like I could give real responsibility or have things that were like an order of magnitude more difficult or orders of magnitude more difficult than what I used to engage AI with. And so that's when I started to send you examples. I'm like, hey, I think we need to think about this. I think it could actually be real operational leverage, like this could actually change the value of our company someday. Like this is a really serious deal. And then, you know, I I would say in the last 90 days, you've outrun me because you have taken like an even like further uh further deep down into the um most fundamental areas of our business. Yeah.

SPEAKER_02 6:23

Yeah.

Salesforce Connectors Breakthrough

SPEAKER_02 6:24

And I one night was just watching TV with my wife, and I was like pulled up on my laptop, and I was like, all right, Republic City computer, like what the who cares? So I like signed up for a thing, and it was like, oh, we've got these native connectors, and I was like, okay, Salesforce. So I like hooked Salesforce in, and it could like do some cool stuff. Like, really, I was like, wow, this is really cool. Like it can read my data, and it can like edit a couple fields, and it can like that's cool, like that's that's really powerful. Like, wow. And then it was like, I wanted it to make a change in my admin settings, and it was like, no, I can't do that through this connector. I was like, why not? Just I literally asked, and it was like, oh, because this is a pre-built connector and it only exposes certain things. I was like, well, how do I get around that? And it was like, I could just build you a script right now. I was like, well, what does that mean? And it said, go click through these things and get these API credentials, and I'll build a little MCP server script and I'll do it. And I was like, okay, so I did the thing and I got the credentials and I put them in. And then it was, I was like, okay, can you add a user for me? And it was like, yeah. I was like, what? And then I said, Okay, can you can you build agents within Salesforce? Because you can build agent force, which is like agents within Salesforce. And it was like, Yeah, I sure can. And so it started building agents for me, which is like really hard and complex to do in Salesforce, and it like built an agent. And then I was like, Okay, what about this task that I do all the time, which is a massive pain in my ass? Can you update my price book? So we got a new Brivo price book, and I was just like, just take this price book and just can you do it? And I was like, Oh, yeah, here are all the parts that you have that are matching. Here are the ones that you don't have in your price book today that are in this one. Do you want me to go ahead and push this? And it did, and it freaking worked. And I was like, okay, now I'm totally this is insane.

SPEAKER_01 8:21

You start, you start now sending me texts. Yeah. Like, did you know it could do this? Yeah. Dude, last night I was up way later past my bedtime. I was up to like 9 45. It's stuck. I'm kidding.

SPEAKER_02 8:32

Uh yeah. Yeah. And I was, I remember I called you. I called you. Remember at like nine o'clock at night, my wife and I were sitting in bed, like watching TV. And I called you. I was like, dude, oh my God, this thing is so crazy. Um, and so that was my first taste of it. And that's so I want to go back really quickly. One thing that really was beneficial to us, and the reason why this was so powerful is years ago we made the decision to get on Salesforce. And I knew, not really because of AI, I kind of thought AI might play a part in it, but I just knew that having all of my data in a single place was gonna be the most important investment we made in the company. And everyone thought I was insane for how much money we were spending on Salesforce and implementing Salesforce. And I was like, I'm telling you, I don't have a specific reason. I mean, I don't have a specific like KPI that's gonna move, like I just know that all of our data in one place is gonna help me do business better. It's gonna make our employees more. And it was, I mean, immediate impact just from having our data in one place operationally that helped us. But the minute that we looped in an AI into it, it changed the whole game because all of our data for our entire business exists in one single place. And so the context, there's no context jumps from okay, I use this field service application that's sort of connected to this CRM, which is sort of connected to this billing system. And it doesn't really tie together really well. It's kind of duct tape and manual. No, everything is in one. And so the AI was able to make meaningful changes and meaningful impact and and like provide recommendations because it had all of our data. And that was a huge unlock for us. And so what happened is I just started chatting with it to say, can you do this? How would you do this? And I started to learn how to communicate because a lot of times it would say, Oh no, I can't, I can't do that. And I would say, Well, why? And then it would say, Well, yeah, actually, now that you've pushed back on me, you were right, I can do that. And I was like, Oh, shoot. Okay, so this thing sometimes doesn't even know what it's capable of doing. And I had to kind of hold it accountable to that. Uh, and sometimes it would try to go off the guardrails and do stuff that I did not want it to do. It's like, oh, you want me to delete all these records? I'm like, I do not want you to delete all these records. You know, do not do that. Um, and so that that really was the start of it. And I remember I came into Dallas and we were at my hotel, like in the lobby till midnight, just like building crap, just like experimenting, trying stuff, seeing what we could build, what we couldn't build, and then and then the big unlock from there, like the kind of like that was we'll say 101.

Agentic Era And Open Source Stack

SPEAKER_02 11:13

The 201 for me was, and you kind of described this, which is this new agentic era. So, what's an agentic era? It's it's automation plus reasoning, is what it is. So you have Zapier, which is like uh when so a lead comes in your website, automatically send them an email. And and all that, there's no there's no intelligence in it at all. It's just a if this do this. That's all Zapier is. And that is awesome, and it works a lot for a lot of different applications. But where it doesn't do well is when there needs to be some reasoning between step A and B, or between B and C, or when there just needs to be true automation and when there needs to be reasoning and when there needs to be some thoughtfulness behind it. That is what agents do. An agent uh takes an automation chain and it throws in reasoning steps to say, oh, uh uh collections, this customer is saying that it's pretty frustrated because they've tried reaching out to us to pay their bill. Let me go scan the CRM real quick to see if that's true. Oh, sure enough, there was a call right here from the customer saying they wanted to pay their bill and nobody called them back. So yeah, don't send them the next nasty gram email saying that they need to pay their bill.

SPEAKER_01 12:28

Which is where I, you know, again interject with like digital employee. Like the when you have automation plus reasoning, it starts to feel like north of an intern. Yeah. Like it, you know, it starts to feel like kind of a more like junior level team member.

SPEAKER_02 12:46

Yeah. I would say what it feels like to me is if I took a really amazing A player employee, but I only gave them one task instead of giving them five different things on their plate. Right? So like uh I I would not give an agent uh for example, we're building an SDR agent. Well, I wouldn't give an SDR agent all the same responsibilities I would give an SDR. I would give them, if I give them five, I would give him one, and I would build five different SDR agents, but they will be world class. It's also really good, AI is really good at doing things that are unstructured tasks at scale. So that's one really big thing, right? So, like for example, I posted about this on LinkedIn the other day. There's amazing amounts of public data out there that no one is tapping into, except for a few businesses that are selling it to you at an ungodly amount of money. Okay, so there is a lot of free data that you have access to, but it requires you to click into a website and it's a 19, you know, 99 website, and then you have to sign up and it's super hard to navigate and it's slow and it errors out and blah blah blah. But guess what? They have APIs and they have a JSON table. And so for you to go get that data would take forever. You you couldn't, it would take you hours to get a hundred leads out of that thing.

SPEAKER_01 14:11

But for an AI agent, it can get a hundred leads in ten minutes because that that AI agent does not need to sleep between 10 p.m. and 6 a.m. Exactly.

SPEAKER_02 14:24

It's just working, and it also can look at it to say different things, right? So you can say, hey, go pull all this public data, and then I want you to go make sure it's within these zip codes. And then what I want you to do is I want you to make sure that it's a commercial business that is about mid-market size, meaning it can't, you know, it's not just in a strip center. So you need to look on Google Earth or Google Maps to see like, is this like an actual building or what you can start to paint the picture of use reasoning in this criteria to give you uh more information. So uh the agentic steps is where we're at today. So now I'm fully because perplexing computers are amazing, it's just it's extraordinarily expensive to run agents on, but you should start there. It's like the perfect alpha beta platform to like start tinkering. Um, there is not very many uh subscription services that have agent harnesses. There are some, like what is it called? Like N8N or whatever.

SPEAKER_01 15:27

Let's explain what an agent harness is and then also share with like what we're doing from a local machine perspective to kind of explain where we're we're really at today.

unknown 15:36

Yeah.

SPEAKER_02 15:36

So an agent harness is the software that is the agent that you plug the brain into, right? So it would be like, hey, you got this body, this robot that can like do stuff, but it doesn't have a brain. And so so you get to pick the brain. So the brain is the LLM, but the br the LLM can't actually do stuff. It doesn't have the ability to code, it doesn't have the ability to run bash commands, it doesn't have the ability to hit an API. The LLM can't do that, it can just think. So the agent harness is the thing that I can actually go do the thing. And so there is not very many agent harnesses platforms, if you will, out there that are like out of the box. There are a few. We went full open source because we wanted control and we didn't want to spend a million dollars on it. So I bought a Mac Mini and I bought an agent harness, which is a software, it's just a software, it just doesn't have a graphic interface like you know, maybe we're all used to. It's a piece of software that runs like on the shell of the computer, like it runs on the terminal. And then you have to give it a front end and you have to interact with it, and you need to give it a brain. And so we have it looked up or hooked up into multiple services. We use one called Hermes, which is free, open source. It's running on a local machine, dedicated in an isolated container, so it can't jack up a bunch of stuff. And it has plugged into multiple LLM providers via API, and it's literally just go to their website, sign up for an account, add your billing, make sure you put in a billing cap so you don't outrun your token spend. Great advice. You you get a key, and don't share that key with anybody, and don't save it in plain text, and you want to give it to you know, Claude Code, and then we use Claude Code to literally take the open source software, drop it into the Mac Mini. We took the key, gave it to it securely, it hooked that up, then we had it hook it up into what we have is our hub, which is our front end where people interact with it, and now we're actually hooking up all of our agents into Slack because that's where we spend all of our time to work. So um I did that in a matter of months. I used Claude Code to do the actual software implementation for me. I did perplexity doing the research, and then we use Hermes, which is the actual piece of software, and then there's multiple LLM brains in the background, and then we use Slack for all intents and purposes for the way to interact with that agent for notifications to get updates, to say something to it, etc.

SPEAKER_01 18:03

Yeah. Also, I mean some of those agents are posting data to us in Slack on a cadence from you know some sort of cron.

SPEAKER_02 18:10

That's called a cron, by the way, C-H-R-O-N, which basically means you want an agent to do something on it. It's C-R-O-N. There's no H. I don't is it is it? I thought there was an H on it.

SPEAKER_01 18:22

Comment how do you spell cron in the comments if you want to know the truth? We'll find out in heaven. I'll reply. We'll find out in heaven. Or on Google. Yeah, but don't even Google. What is it? What is it, 2014? Yeah. Um we'll find out in heaven. Don't worry.

SPEAKER_02 18:39

Um, so we've made a lot of progress. We've covered a lot of ground today. We just kind of wanted to give everybody a 30,000-foot view of how we got there. Um, take your own journey, go talk to Perplexity or Claude, and go build it yourself.

SPEAKER_01 18:53

Yeah. You are capable. I think like both of us were not software engineers. We uh were interested, but we didn't have a significant amount of prior experience that would lead you to believe that we could be where we are five, six months later. Yeah. So that's right. I think if you are interested and you're curious and you're willing to spend the time, yeah, within a matter of two or three weeks, you will be significantly further along than you re would think today. And so, you know, uh the the best time to start was two years ago. The next best time is today.

SPEAKER_02 19:29

That's right. Um Steven, I got a question. Um, so if folks want to know more about this or they want to engage, how could they do that?

SPEAKER_01 19:41

They could subscribe to our podcast, could subscribe to our YouTube channel, they could also subscribe to our newsletter, um, they could engage with us on LinkedIn or on X, formerly Twitter. And uh you can send us an email. You can go to our website, entryandexit.co, you know, send us a little info app email. Really any of those options or all of them are encouraged. Yeah, they're encouraged. Thanks for listening.