---

Episode Host: Henry Lopez is a serial entrepreneur, small business coach, and the host of The How of Business podcast show – dedicated to helping you start, run, grow and exit your small business.

---

Resources:

Related Podcast Episodes:

Episode 615: Mordy Hackel – AI and Cybersecurity for Small Business

You can find other episodes of The How of Business podcast, the best podcast for small business, on our Archives page.

Transcript:

The following is a full transcript of this episode. This transcript was produced by an automated system and may contain some typos.

Episode Transcript

This transcript has been lightly edited for readability.

Episode Transcript

This transcript has been lightly edited for readability.

Henry Lopez [0:12]: Welcome to this episode of The How of Business. This is Henry Lopez, and my special guest again today is Colin Rhoades. Colin, welcome back to the show.

Colin Rhoades [0:21]: Hi, Henry. Thanks for having me.

What Is Agentic AI?

Henry Lopez [0:23]: We’re going to chat today about AI specifically. We’re going to share with you what a lot of people are calling agentic AI, and we’ll explain what that means. It’s the next level of AI, really, and we’ll explain how we’re using it, how I’m seeing my clients use it, how Colin is applying it in the environment where he works, and how it’s something I want to challenge all of you listening to start at least experimenting with and leveraging.

Most of you, I hope, have tried ChatGPT, Claude Chat, or Gemini, but there’s a whole other world that maybe some of you have started to work with, which is what they call agentic AI. It really is a shift from just asking a question and getting a better response than you might get from a Google search — this takes it to the next level. This episode is about understanding what that is and how you can apply it in your small business. We’re not going to tackle this from a technical perspective, although I’m sure we’re going to throw out some technical terms, and we’ll try our best to explain it from a business owner’s perspective.

Colin has been on the show before. I’m happy — and honored — to say he’s my son-in-law now. He has become quite the AI guru in the corporate world. He’s the expert where he works on applying AI, applying it to data analytics, and he’s really become a student of AI over the last couple of years. So he’s my go-to person now when I need help applying AI, learning AI. He’s the one who got me on Claude Code, which we’re going to introduce today. He’s with me to add color commentary here as we go through this and help me explain what agentic AI is, and Claude Code in particular. It happens to be the platform we’re using, but this isn’t a commercial for Claude Code — there are other solutions out there. We’re going to share how we’re using Claude Code and why we switched. I switched from ChatGPT to Claude Code, although I still have ChatGPT and use it often, and I use Google Gemini too. But specifically, I’ll explain what I use Claude Code for and share a couple of real examples.

Let’s start with: what the heck is agentic AI? It sounds fancy, it sounds technical. What’s the difference between the AI most of us have been using, like ChatGPT, and agentic AI? I think AI, as most of us have experienced it, is the next level of Google search, if you will. When we’re searching on Google for a specific business, a specific product, or a specific restaurant, we’re still doing those traditional searches — but as you may know, Google is now also serving us an AI answer, powered by Google Gemini. So you ask a question, you get a better answer with ChatGPT or Gemini or whatever you might be using, and then you go and do the work. It gives you instructions, some knowledge, or the results of some research — but then you go do the work. That’s kind of one of the key things about chats, whether it’s ChatGPT, Claude’s chat, or Gemini — is that right, Colin?

Colin Rhoades [3:26]: That’s correct. I think most of us now have experienced AI as a chatbot: something you talk to, you ask it a question, it gives you an answer, and then you still have to go do the work. Now, I think with agentic AI, that relationship kind of changes. Instead of only asking “how do I do this,” you can increasingly say “here’s what I need to accomplish — help me get it done.”

Henry Lopez [3:49]: So it helps you reach a goal, and you tell it what it is you want as the outcome. Now, some of this does overlap. I was using ChatGPT, telling it to do research, going back and forth, and the deliverable might be a Word document I’m going to share with someone else. So that’s kind of the same. Here, though, we’re talking about performing functions — going and doing work.

Colin Rhoades [4:11]: Yeah, and I think the analogy I like to use is a consultant versus an assistant. Chatbot AI is like having a very smart consultant at your fingertips. Agentic AI is more like giving a consultant, or an assistant, or an employee, a desk, access to the tools they need, and an assignment to go do.

Henry Lopez [4:31]: We’ll give you more examples, and it’ll start to make more sense if this is new to you. But let’s just summarize it for now: a chat, like ChatGPT, gives you answers, versus agentic AI, which gets things done. That’s the simplest way I can put it.

Colin Rhoades [4:47]: You’ll interact with it the same way you’ve been accustomed to — speak to it in plain English. It works out the steps, uses the tools it needs to, and you approve anything that matters. At the end of the day, you’re still the human in the loop.

Henry Lopez [5:02]: Right? It’s not like — you know, there’s been news lately about AI breaking loose and hacking things. That’s not what we’re talking about here. We’re not at that level of sophistication, and you have complete control over what it’s doing. It’s in a closed environment. And if you haven’t been using AI, I’d say start with one of the chats — ChatGPT, Gemini, or Claude Chat. Start there. I don’t know that you jump right to code. If you want to, that’s fine, but I think you start with a chat first.

Colin Rhoades [5:34]: Yes, I agree.

Meet Claude Code

Henry Lopez [5:34]: Okay, so Claude Code specifically — the name Claude Code is a bad name, because it seems very technical, like it’s for coders. I think you explained to me that was just its original use case, perhaps. But it’s more than just code generation.

Colin Rhoades [5:52]: Claude Code was built originally for software engineers, and I guess that name has stuck. Really, at the end of the day, that’s the only thing technical about it. When you open it up, it’s going to look exactly like the Claude chat looks. You’re still going to type to it in plain English, same as you would with the AI you’re accustomed to using. The difference now is that it can go and do things, not just give you answers. If a job needs code to get it done, Claude Code can write it for you. You won’t necessarily see that, but that’s what’s going on behind the scenes.

Real-World Example: Automating a Shopify Website

Henry Lopez [6:28]: Great explanation. I think this is a good time to share an example — one where you were involved as well. My daughter Makena, as I’ve shared before, has been launching Final Curtain Finishing. She does needlepoint finishing — I didn’t know what that was six months ago either, so if you don’t know what it is, no big deal. It’s a finishing service, and her online site, where customers place orders, is on Shopify.

Native Shopify really isn’t great at configured products, and hers is a configured service — customers make certain selections to end up with the service they want, what they want done to their needlepoint canvas. So we not only had to use some plugins, or “apps” as they’re called in Shopify, but we needed some custom HTML code. Even my daughter Makena, who has no coding background — I have a coding background from way back when, but she does not, and I don’t think she’d ever really even seen a couple lines of HTML code in her life — was able, using Claude Chat originally, and then Claude Code more extensively, to create custom code that made the website behave exactly how we needed it to. Not just visually — we started with one of Shopify’s free themes and added to it — but for this configuration component, which works very differently from how Shopify natively handles things.

There’s no doubt, Colin, we would have spent thousands of dollars with somebody else. We would have had to hire it out, and it would have been a whole different experience, because we wouldn’t have been able to take the iterative approach we’ve taken. Every time we need something — I just had to add a new product last week — I’m able to use Claude Code. Without it, I would have had to hire somebody again, or pay again, for that custom code. Makena was able to generate that code by giving plain English instructions describing the business function. It generated the code, and then it was just a matter of plugging it in. Now we’re taking that a step further — we’ve connected Shopify to Claude, which we’ll talk about in a moment. Beyond just generating code, Claude Chat could have done the same thing, and it did, right?

Colin Rhoades [8:32]: Right.

Connectors, Memory, and Skills

Henry Lopez [8:33]: But the key thing is these things called connectors, right, Colin? Claude connectors —

Colin Rhoades [8:39]: MCPs, APIs — you hear all these acronyms. Exactly.

Henry Lopez [8:43]: Right. So what Colin did is connect our Claude instances — both mine and Makena’s — to that Shopify account. Now it can read what’s in there: the configurations, the parameters, the fields, the existing code. I just give it plain English instructions, and because it’s plugged in, it can go find what it needs and apply the change. That’s when we start to get into this whole other realm, as you highlighted, Colin — of it doing the work for me, right?

Colin Rhoades [9:14]: Absolutely. This is what an agent means. You now have, at your fingertips, a Shopify admin on your team that’s connected into your Shopify account. It can see everything you could possibly see — all of your settings. Think back just a few months ago, before I was using agentic AI: if I was trying to ask a question about my CRM or my Shopify, I’d have to give chat a link to the Q&A page, ask my questions, send it screenshots. That whole process has been erased now by giving agentic AI access to these platforms. I call it recon — it can now recon your entire backend settings and understand very quickly what problem you’re trying to solve or what question you’re asking.

Henry Lopez [10:06]: Right, exactly. We’ll share a couple more examples of how we’re using it, and how my clients are using agentic AI. But I want to stay right now on the topic of how it’s implemented and how it’s different from a chat. It’s installed locally, in my case, on my laptop. There’s a component that’s still talking to the cloud-based Claude, but it’s running on my laptop, and it’s accessing data that might be in the cloud, like the Shopify account. It can also access files I have locally on my hard drive.

Colin Rhoades [10:40]: Correct. It’s a separate piece of software from Anthropic that you install. It’s almost like an operating system that can now run and do anything on a computer that you could do — if you had those technical skills, if you were a coder.

Henry Lopez [10:54]: So I have it connected, as I said, to Shopify. I have it connected to my Outlook email, and I’ll start connecting it to my CRM — I haven’t yet, but I will. So it allows me, from one place, one console, without having to worry about exports or extracts, to go read that data in real time and help me either answer questions or, more importantly, go do work for me. And that’s the key thing. The other thing that’s very different for me is that it remembers. Can you explain what we mean by it remembers, and how that’s different from how things worked at a high level when I was using ChatGPT?

Colin Rhoades [11:32]: Chat memory is more the AI simply remembering things about you, pulling from its brain. Agentic memory is more akin to it writing down exactly how your business works —

Henry Lopez [11:45]: Or a particular process, a particular function. Right?

Colin Rhoades [11:48]: It’s more like taking detailed notes and filing them away, so that when you open a new session, it can read those notes — they’re right there. It doesn’t have to think back and try to recall exactly what you were discussing three weeks ago.

Henry Lopez [12:04]: The way I’ve expressed it is, it’s almost like it’s keeping a log for each session, so it can pick back up where it left off — as opposed to a chat environment, where it has to go back through everything in that session to remember again. The other thing that’s unique — not only is it better at remembering, but what they’re called are “skills” in Claude Code, which I would have called a program, or an app, or a small app.

I’ll give you an example. One of the things I’ve been doing in chat for probably close to a year now is using it to help with my podcast production. When I record an episode like this one, the first thing I do is get a transcript, then feed that transcript into ChatGPT and say “run that same procedure” — a six- or seven-step procedure that generates a show notes page, a LinkedIn post, and a bunch of other things based on interpreting that transcript. What would happen over time is it would start to add things, or forget things, and I’d say, “well, why didn’t you do all six steps?” “Oh, my bad, I’ll do it now.” So what I was asking it to do is really not what it naturally does best. Now, with Claude Code, I’ve created a skill, and it’ll remember to execute it the exact same way every time.

Colin Rhoades [13:24]: Think of a skill as you telling the AI who you are, how this project works, what the rules are, and the things it should always keep in mind. It tells the AI how it needs to perform a particular job, and here’s the procedure to follow.

Henry Lopez [13:38]: So what I do now — I just did this for the episode I released yesterday — I attach the transcript and say “run the podcast production skill,” and it knows what that means. It goes off, and then what I do is make it better and better. Like any written work AI generates, it’s just a first draft. I take that and edit it. So what I’ve been doing is taking my finished, edited version and giving it back, saying, “I want you to learn from how I’m wording things and apply that to the skill.” I’m making the skill better and better over time, and it’ll remember that as part of the skill. That’s been a game changer for me. That’s not agentic AI, to be clear — I’m just describing some of the reasons and benefits of switching to Claude Code.

Colin Rhoades [14:25]: I think we want to make clear that a skill isn’t some piece of magic — the AI is helping you build the skill for the particular task you need. If you can train an employee to do it, you can train AI to do it, and it’s going to perfect it. Over time, that’s going to compound and become a reusable workflow for you.

Henry Lopez [14:46]: In fact, you mentioned SOPs. If I have an SOP as a starting point — someone doing this manually now, or accessing different systems — that can be the initial instructions I give Claude to begin creating that skill. Yes?

Colin Rhoades [14:58]: If you already have SOPs built out, that is gold for any AI, but especially Claude Code, agentic AI.

A Word on AI Security and Trust

Henry Lopez [15:06]: All right, I want to take a quick moment here to talk about security. That’s not what this episode is about, but of course it’s a concern, and I hear it from a lot of people. As I’ve shared before on this podcast, it’s not that I’m unaware of the concerns, and some of them are legitimate. Some of it is how people are selling services and solutions — the fear factor. I think some of it is blown out of proportion, and some of it is real — who knows. I’m choosing, as I say, to bury my head in the sand about the negative aspects of AI to some extent. Obviously, I’m not being careless with confidential information, especially my clients’ information. I’m not putting anything into the settings that let the model use it to learn — that’s a setting in Claude, as well as in chat, for that matter. I’m keeping everything local, and I’m not putting anything in there that somebody could do damage with.

So I’m taking those precautions, but I’m also not getting hung up on it. Colin and I were trying to think earlier today about when I got started with chat — it’s been at least a couple of years. I decided then and there that I wasn’t going to get hung up on how it can go badly. I’ll let others worry about that. I’m going to focus on how I can use it to make myself more productive, and how I can help my coaching clients enable themselves to improve their businesses. I did have a conversation on episode 615, which I released recently, with Mordy Hackel — and in fact, you connected me with Mordy. He’s a technical expert with a huge background in IT; he provides outsourced IT services for companies. On that episode, he shared some thoughts on the security and fraud aspects, so if that’s something you’re worried about, that’d be a great episode to go listen to.

Colin Rhoades [16:45]: I think small business owners already understand permissions. You’re not going to hire somebody on day one and give them access to the bank account or the keys to every system in your company. I’m not suggesting you treat agentic AI any different — it’s a simple principle. You start with controlled access, and you’re still keeping yourself, or another human, in the loop for consequential actions or decisions. From there, once you become more comfortable, once you understand how the tool works and start to trust it, that’s when you can decide to expand its autonomy. I think the goal isn’t “AI can do everything.” It’s having an agentic AI that you can trust to do one defined task well.

Putting Agentic AI to Work: A CRM Example

Henry Lopez [17:33]: I think it really comes into play for repetitive tasks — things that have to be done on a regular basis. I’d like you to share an example of one you’ve set up in a corporate environment — a good example of agentic AI where, even if it doesn’t save a lot of time on any given day, it adds up. Can you share that example?

Colin Rhoades [17:53]: Yeah, the moment it clicked for me — this was my first week using Claude Code — was when I connected it to my CRM. Before that, I could ask a great question, but there was still an entire manual process sitting between the question and the answer. I’d have to go into the CRM, find the right report, export it, clean the data, maybe combine it with another file, bring that back into AI, and then finally start analyzing it to find the answer. With Claude Code, that agent can actually work within the system, so the bottleneck completely changes. I’m not spending my time gathering information anymore — now I can focus my time on deciding what that information means.

Henry Lopez [18:42]: Brilliant. That’s the key takeaway here in a nutshell — but now, based on that data, perform a function, take an action. That’s where you really start to leverage the power of agentic AI.

Getting Started with Agentic AI

All right, if you’ve stayed with us this long, I think you might be interested to know I’m thinking about starting a series of workshops to help you implement agentic AI in your business. If you’re interested, go to the show notes page for this episode at thehowofbusiness.com — there’ll be a form there you can fill out expressing your interest. There’s no commitment, but I want to gauge interest, and if there’s enough, we’ll put it together. The idea is that by the end of the series, you’ll have installed Claude Code and, over a number of sessions we do together, designed and implemented an agent that’s doing real work within your small business. Go to the show notes page for this episode at thehowofbusiness.com, fill out the form, and we’ll follow up with details. It’ll probably launch toward the end of 2026, or maybe into 2027.

As it relates to the primary topic of this episode — going from the chat world, which is very powerful and which we still use daily, to leveraging agentic AI to do work for us — let’s summarize where we think you should get started. We’re assuming that to start dabbling with agentic AI, you’ve tried or have been working in ChatGPT, Gemini, or some other chat. That’s where I’d recommend people start with AI. Does that make sense, Colin?

Colin Rhoades [20:35]: I agree with that. That’s fair.

Henry Lopez [20:37]: Now you’re ready to have it start doing work for you.

Colin Rhoades [20:40]: I’d ask your listeners to think of something they do every week that they wish they didn’t have to. Think about something repetitive, something you understand well, something with a clear outcome, something relatively low risk, and something currently consuming a lot of your time on a weekly basis. I’d start with the work before you even get to the AI piece. Once you find that repetitive process and document it, you can teach the AI that process.

Henry Lopez [21:11]: Perfect. What I’d add is that it’s often something you’re doing yourself because it’s not something you can necessarily delegate to a VA or someone on your team. Start with a process you know well and that’s at least documented — maybe even just in your head — and build from there. The other thing Colin and I will often do when we’re working on a project is, instead of typing it in, we just talk to it — turn on audio mode, the microphone mode. So that’s what I tell people: if you’ve got the process in your head, just speak it to Claude, and it’ll take it from there.

Colin Rhoades [21:46]: I think that saves a tremendous amount of time — being able to speak into your microphone instead of typing on your keyboard. I’ll add one strong example: use AI to help you learn AI. If we’re on the topic of automation, you’re not going to type into AI, “help me automate my business.” Instead, you could tell it, “every Friday I receive these three reports, I combine these four columns, I check these five things, and I send it to these two people.” That’s a task that’s teachable. And once something becomes teachable, you can delegate it.

Henry Lopez [22:19]: What I’d add, what I’d layer on after you’ve started doing that, is leveraging the intelligence part of it. What if, once it’s gathered all that data, it begins to interpret the data to determine how to route it, or what to do, or whether to route it at all? It can start to interpret and apply that intelligence, like an employee would. Is that fair?

The Democratization of Business Intelligence

Colin Rhoades [22:42]: That’s fair, and I think that’s the beauty of what this technology unlocks for small businesses. Traditionally, large companies have always had this leverage — departments and teams of analysts, developers, IT departments, operations teams, consultants. Small businesses didn’t necessarily have access to all of that. I think that paradigm is now shifting, especially with this next wave into agentic AI. A five-person company now has the bandwidth, the capability, to act like a 100-person company.

Henry Lopez [23:15]: It’s the democratization — just like the internet did for information access, this is the democratization of intelligence, and of course automation as well. There have been tools that could do this kind of thing before, but now, to your point, as a solopreneur or a small business, I have access to the same level of computing power that the Fortune 100 does, to a big extent. That’s the advantage we have. That’s why this is such an incredible time to be in business, and to be alive — because while there’s a lot of potential for this technology to go wrong, what it’s accelerating, what it’s allowing us to do as business owners, is amplified. What the internet did to facilitate starting and running a business, this is on steroids — what I have at my fingertips on my laptop helps me think, research, prepare, operate, and become that much more efficient.

Colin Rhoades [24:13]: I think very soon we’re going to see the first billion-dollar one-person company.

Key Takeaway: Give Yourself — and Your Team — Time Back

Henry Lopez [24:19]: That’s what everybody’s saying, and I was hoping to be one of those people, but I don’t think I’m smart enough. All I want is to continue to become more productive, and to help my clients leverage this technology to become more productive themselves. Agentic AI is not about replacing you or your team, at least not right now, as far as I can see — maybe that’ll happen in the future. It’s about taking low-value work off of your plate, and your team’s plate, so you can focus on the work that matters more. That’s what it’s about, and you don’t need a technical background or to have been a coder to use agentic AI. It’s still a natural language model — you’re talking to it in plain English, giving it commands, telling it what to do. All you need to know is your business, and even that, it can help you figure out. As Colin pointed out, you start with a simple task and build from there. But you have to try it, explore it, learn it — and the best way to learn is to actually apply it, work it, and learn as you go.

If this is of interest to you, go to the show notes page for this episode at thehowofbusiness.com and let us know if you’d be interested in a series of workshops we’re planning that will take you from installing Claude Code all the way through developing your first agent. Colin, as we wrap up here, what’s one key takeaway you want us to consider?

Colin Rhoades [25:46]: The goal, at the end of the day, is not to replace people with AI. It’s to give you and your team the leverage, the tools, and the time back to go do the things that actually need a person — spend time with your customers, apply judgment and creativity, and grow your business. Every business has work that needs to get done, and work that only you, as the owner, can do. Agentic AI is giving you more time for that second one.

Henry Lopez [26:16]: Agreed — we’ll leave you with that. Go to the show notes page again for more information, and to let us know if you’d be interested in joining us for this series of workshops. We’re going to do follow-on episodes on this topic, because there’s so much more to talk about here. Hopefully this was of value to you, and it encourages you to begin using AI if you haven’t already — but likely, if you’ve listened this long, you already have, and you’re ready to take that leap into agentic AI, and, as Colin said, become that much more productive with the limited resources we all have as small business owners.

This is Henry Lopez, and thanks for joining me and Colin on this episode of The How of Business. I wish you the best as you start and grow a successful and profitable small business. I release new episodes every Monday morning, and you can find the show anywhere you listen to podcasts, including The How of Business YouTube channel, and at my website, thehowofbusiness.com. Thanks for listening.