Loading summary
A
Simply put, an AI agent can actually go out and do executable tasks and complete jobs. You don't need to know how to code shit. They do it for you. Obviously. Don't paste your fricking keys into, like, the chat box.
B
He had me do that. But don't do that. It cost me $5,000 and I got my cloud account taken down for like three days. I like to build HR agents.
A
Do they snitch?
B
Yes.
A
Oh.
B
Every three days that they work, they get assessed. Are they hitting their KPIs? And when they don't, maybe kill them and start over again. It's a beauty with agents. Welcome to Follow the Yellow Brick Road, the show where online businesses learn how to turn clicks into customers and growth into real scale. I'm your co host, Emma Rainville, the wizard of ops, helping companies transform chaos into systems that actually run.
A
And I'm Mitch Barham, the wizard of ads, the guy who knows how to turn paid traffic into predictable revenue. Together, we break down what really drives profitable online businesses. Traffic and funnels to operations, scaling, and everything behind the curtain. Because getting customers is only half the battle.
B
The real magic happens when ads and operations work together. So if you want smarter traffic, stronger systems, and a clear path to scaling your business, you're in the right place. Let's follow the yellow brick road. All right, Mitch, Today I want to do something different. I pulled the most asked questions about AI agents and I thought I would just ask them and we could do like rapid fire for this podcast.
A
Sweet. I'm only going to give yes and no answers.
B
How do I build an AI agent from scratch?
A
That's a long answer, short answer. You have to give it your basically a framework. So to start, it has to know about your company, your brand, yourself, tone, all that stuff, right? So you got to feed that. So it has your framework for like your orchestration layer. Then determine what do you want it to do, what do you what, what jobs, what roles, what tasks, all that stuff, and know how those things are completed. So you can train it with like your SOPs, examples of the work, and basically structure it out like you would train a actual human being. It's a very short answer, me trying to not take forever to explain it and give out.
B
Can I answer it?
A
Go for it.
B
If I was going to build an agent from scratch and I had no idea what I was doing, I would go into Claude and I would say, please tell me step by step, like, I have never used a computer before how I should build an AI agent using A long term framework that I can build off of to add as needed. And then I would let it build out step by step what I need to do. And any step that I didn't understand what it was telling me to do, I would ask it questions, I would go and download the app, I would open up CLAUDE code and I would go back and forth with Claude on how to build in Claude code and build in Claude code. But I would use, I would use the statement, how do I build an agent for long term use? And understanding that I'm going to want to connect agents in the future and build on this, that's what I would do. Mitch's answer was better but not executable. I think for the average person that would be asking, how do I build an agent from scratch?
A
Yours is much more executable.
B
I was like, oh, operator.
A
Yeah.
B
What is an AI agent and how does it work?
A
It's magic. It's a little gnome forest. It makes cookies. Agent.
B
Okay. I think that was the perfect answer for such a stupid question. What is the difference between an AI agent and a chatbot?
A
Oh, I mean, simply put, an AI agent can actually go out and do executable tasks and complete jobs. And a chatbot is chat GPT going to the web and being like, tell me about this or what is this?
B
Is that Boris's voice?
A
Yes.
B
Oh, wow.
A
Yeah, I'm picking up how to do Boris's voice.
B
Can I build an AI agent without coding 100.
A
You don't need to know how to code at all. You literally build them with natural language, like you said to the first question. Literally turn on your microphone on your computer and have a conversation whether you want to do it in Claude, chat, GPT, Open Club, whatever you want to do, you just have a conversation with it. You don't need to know how to code shit, they do it for you.
B
So I didn't make this up. Literally says this. What is the best framework for building AI agents?
A
The workforce stack. So that's three layers. Very simply, that's our framework and it's. The operator is layer one, layer two is the orchestration layer and layer three is the execution replacement layer. Makes everything very simple.
B
Very simple. How do I build an AI agent with Python?
A
You ask whatever platform you're using to do that and actually you don't even need to do it because you can just talk to it and tell it to build it in Python, then it will build it in Python. You don't need to know how to code.
B
How do I build An AI agent with Google Gemini.
A
You go into Google Gemini. I mean, everything is natural language. It's pretty wild. Like, you literally, with Gemini, any platform, you are just going to literally ask it and tell it what you want to do.
B
Can I build AI agents in OpenAI 100%. Just like you can in Claude. And just like you can in Claude, it has codex versus code and it has agents versus Cowork, and both have managed agents. How do I connect an AI agent to my business data that's building the orchestration layer. How do I give AI access to tools or APIs safely?
A
Obviously, don't paste your freaking keys into, like, the chat box.
B
He had me do that. But don't do that. It cost me $5,000 and it got my clot account taken down for like three days.
A
I think she got hers back thanks to Richard. What did he do?
B
I don't know. But I told him about it and it magically came back.
A
Going to get mine back.
B
I think he just went inside the thing and messaged them. How do I train an AI agent on my own documents?
A
On your own documents? Train an AI agent on your own documents? Well, you literally can feed it your documents.
B
Where?
A
In either the desktop app Vs code. You can put it in a project, you can keep it in a file. Really. You can build a rag system which then houses all these documents. Kind of like a brain, but not really a brain. And then it's trained on everything and anything that you need it to do.
B
We really should teach people how to build an AI brain.
A
Interrupt this podcast to ask you to do a huge favor for us.
B
Huge.
A
We drop so much knowledge to help you guys, but you're not going to be notified unless you hit that. Subscribe in the notification bell to be notified the minute we drop a new episode to help you grow and scale your business.
B
Because you never know when we're going to be taken off YouTube.
A
You have no idea. Like, I could all of a sudden be like Liam Neeson with a particular set of skills.
B
Also, we read every single comment, so
A
leave a question or comment, we will reply to it and maybe we'll even talk about it in a upcoming episode.
B
Probably. If you TR us, we will definitely talk about you in an upcoming episode. Thanks. How do I build an AI customer
A
service agent connected to your customer service platform? You're going to feed it all of the common questions, all the different answers. You should have an SOP on this. And so you're going to basically feed it all that info of common questions, frequently asked Questions how to respond to people. Really the best way is to feed it a ton of your old, not old, but know, responses so it can answer like you. And then also feed it ton of. A ton of inquiries. And this is all gonna be built in natural language, by the way. Like you're not coding anything like I. If you're using like, I don't know, Zendesk. And then, yeah, attach it via like an API key or an mcp.
B
I think that's so much. Build an agent that scrapes your customer service platform like Zendesk or whatever, and have it pull all of the answers that the customer service agents gave. All of the protocols for closing tickets based off of how tickets have been closed for the last 1650 days. Build a product knowledge center, email macros, FAQs and quick study guide. Then take what that agent builds and make it the brain for your customer service agents. And then make macro agents, email answer, chat answer. You can even have like, if you have customer service voicemails come in and then you answer via email through voicemail, which a lot of people are doing now, and text, and you build an agent for each and then you train them off of that file. It's a lot less work.
A
That sounded like a lot of work.
B
No, the agent did all the work. Okay, you, you had the agent, you built an agent to go, this is where cowork is great, because you would build that agent in cowork to go scrape and build all that information. And then now the information lives in cloud code, which you feed it to Claude code as part of your orchestration layer. It also works beautifully for your marketing agents in your orchestration layer.
A
Hmm.
B
Not enough people are pulling their customer service data and feeding it to marketing for future promos or pain points. Well, the reason why you want it in future promos, because why are people refunding. Let's address it now. What expectations are we not hitting that we can hit in future marketing, Right?
A
Yeah, Yeah, I like it.
B
How do you monitor an agent, I. E. Manage an agent over time?
A
Well, in my opinion, you're going to have other agents manage those agents. And then also if you want the real nerdy answer, like there's an actual feedback loop inside of it with like, I'm not going to call it the right name right now. Think of it like an error log file, but it's literally not an error log file. It's just the only thing my brain could think of right now. But it's like a feedback loop where it learns over time and it actually logs its own errors. And then another agent can monitor that and then send you alerts if you need to in Slack or email.
B
I like to build HR agents.
A
Do they snitch?
B
Yes. Oh, you know that I have a police. I have an execution agent. I also have an HR agent, which I tried in Claude. I'm moving it back to ClickUp just because I like to be in ClickUp. But the HR agent reviews and summarizes agents based off of their KPIs and accuracy. So how many times did it go back on their deliverables? And then it gives me a report every three days on all active agents. I don't use every agent every day.
A
Right.
B
All active agents every three days. On how those agents do on a one through five scoring system. One being they hit KPIs and they had nothing coming back. And five meaning they didn't hit KPIs and they didn't hit their goal of deliverables. And then I got to look at the agent because I broke it. I didn't build it. Right.
A
Interesting.
B
But instead of doing like quarterly reviews every three days, because why not? I don't have to do any of the work.
A
Yeah.
B
If it's getting one or two, I don't have to look at anything. If it gets three twice, I need to look at it. If it gets four or five, I need to look at it.
A
No, that makes sense. It's just interesting. It's not really a difference in how our brains work. But I go to the super nerdy route and then yours is much more operator.
B
I'm in love with simplicity. I really think that the more I can find KPIs like Chick Fil A. Talk about chick Fil A all the time. Chick Fil A has one KPI. Do you know that? What's their KPI?
A
I don't know. But you've told me before, but I forgot it.
B
Okay. KPI is. Was this. It's a yes or no. Was this the best?
A
That's right.
B
Fast food. Custom fast food experience you've had even if you were here yesterday.
A
Right.
B
And if the answer is no, then I know that I need to look at the micro KPIs because your answer was yes. I know that the kitchen's clean. I know that the food is fresh and prepared properly. I know that the condiments are stacked. I know that the dining room's clean. I know that the bathroom's clean. I know that my staff treated you well. I know that.
A
So happy.
B
They're fucking Teenagers. How did you do that?
A
Yeah. Teach us.
B
How did you teach me? Your way. But yeah. So I, I like to find the one KPI that tells me where I need to look and ignore the things I don't need to. There's so much noise.
A
True.
B
And I find so much value in knowing what not to pay attention to.
A
Smart.
B
Last question. How do I keep an AI agent safe, accurate and secure.
A
Keeping an AI agent safe, accurate and secure. So safe and I think secure kind of go hand in hand. But like I'd love to hear obviously what you have to say as well. But my nerdiness I go right to like you obviously are giving it guardrails. You're not giving anybody else access to any information that you don't want out there. So like you're not putting your API keys out in public. You're not sharing the files. It's very strict. Guardrails as well. And like when you give it access to things, you're limiting its scope of access. Accurate. Again it's down to guardrails. Like what is its specific job and what is the end results look like. Right. And then I'm sure for you there's going to be some level of scorecard KPI that comes back.
B
There is a scorecard KPI. The first one is safe. What did I learn? Don't put KPI keys in the chat box and then tell Claude to go put it in the folder. Put it in the folder and then give access to the folder instead. Make sure that your GitHub is private and not public. That's super important. Didn't know that. Found that out the hard way. It wasn't my hard way. It was someone was telling me that they had public facing API keys and they got screwed. So I found it out the hard way. But it wasn't my lesson. How to keep it accurate. I'm actually going to come back to that. How to keep it secure. That employee handbook. That employee handbook has an NDA that it has to agree to before it activates itself. That employee handbook has a lot of guardrails including our value system. So we have a value system within our company. We also have a policy on political events and what we will report or hold an opinion on. We have a policy on how we address things in the world and the workplace and in our clients work place that are possibly offensive to other people and how we address those things. This way, if you get a chat with the HR agent and you say one of my staff is requiring me so one of your human Staff goes to your HR agent and says, one of my staff is requiring me to say they them. I don't want to say that. What do I do? It's going to answer appropriately, not what it finds on the interwebs.
A
You can find some wild stuff out there.
B
Exactly. Personally, for me at Shockwave, I want to operate within my legal compounds and I don't necessarily subscribe to grammatical errors in writing or in addressing other humans. So we have that as a policy. You get to do what you want. But our agents follow the same policies. That's part of the security. We don't store passwords and spreadsheets. We store them in LastPass. That's a KPI for our agents as well. So if they need to create a password in the corporate course of their job duty, they create the password, it goes into LastPass and then signed to a human operator to change the password and reshare it. So we have guardrails like that within our employee handbook, which is a massive handbook, but it allows our agents to work autonomously through the chain without stopping and waiting for human intervention because it knows what to do with a lot of the different nuances. And then in order to keep them accurate. That goes back to our HR that I was talking about earlier that assesses. We have an HR assessment agent. I called her friend because my chick, my HR person in Scalewind is Fran and I just absolutely love her. One of the best HR people I've ever worked with. So I have her assessing our agents, our active agents. Every three days that they work, they get assessed, they get a one to a five. That tells me the deliverables coming back not done correctly. Are they hitting their KPIs? And when they don't, then we retrain or provide more training or assess training and maybe kill them and start over again. It's a beauty. With agents, we can do that quite easily. So that's how I keep them safe, accurate and secure. And that is the end of our questions.
A
Love it.
B
That's it for today's episode of Follow the Yellow Brick Road, where the wizard of Ads and the wizard of Ops break down what it really takes to build and scale scale a profitable online business.
A
If you found this episode useful, make sure to go to follow the yellow brickroad podcast.com to check out the Hidden Control Chamber for all kinds of awesome freebies, guides, checklists, everything that we do to grow businesses.
B
So hop over there and grab all those free resources in the Hidden Control Chamber. And remember, getting traffic is one thing. Turning it into customers that build the systems to support real growth. That's where the real magic happens. Until next time, keep following the yellow brick road.
Hosts: Emma Rainville & Mitch Barham
Date: June 2, 2026
In this episode, Emma (Wizard of Ops) and Mitch (Wizard of Ads) tackle the burning question among founders and business operators: What’s the real difference between a chatbot and an AI agent? Through rapid-fire Q&A, they break down the crucial distinctions, practical frameworks for building AI agents, safety and training considerations, and actionable advice to leverage AI in streamlining customer service and business operations. The duo uses their lively, accessible banter to demystify technical concepts, making AI accessible for both techies and non-coders alike.
(01:30 - 03:28)
Emma’s Approach:
Mitch’s Advice for Beginners:
(04:11 - 04:40)
(04:40)
(06:18 - 06:42)
(07:23 - 09:31)
(09:52 - 12:55)
(13:09 - 17:28)
Guardrails:
Accuracy Controls:
Cultural and Policy Alignment:
Emma and Mitch keep the atmosphere practical, no-nonsense, and—frankly—fun, using relatable stories and real-world business scenarios. The biggest message: AI agents aren’t just chatbots—they’re your new workforce, if you train, guide, and secure them right. You don’t need to code—just know your business, your goals, and ask the right questions.
Final tip:
“Getting traffic is one thing. Turning it into customers that build the systems to support real growth. That's where the real magic happens.” (Mitch, 17:51)