AI Agents Unleashed: Exclusive Preview with Chris Lannon Before the Full Summit Workshop!

Full video transcript

Welcome into today’s live webinar. We are joined today by Chris Lannon from Profit, going to be talking about AI agents, automations, using tools like voice assistants, Make.com for automations to do a number of different things. But to quickly put some context between what we’re going to be talking about today, this is actually going to be a preview of what we’re going to be going into at an even deeper level just next week during the live Scale with AI Summit 2024, which kicks off on November 5th all the way through the 7th. We’ve got a star-studded, stacked lineup talking about things from AI automations, content creation, and different practical ways that you can leverage these innovations to grow and scale your business.

Now today I’m joined by Chris Lannon, as I’ve said, we’ll be talking about some of the projects we’ve worked on in the past, but how you can practically get started with understanding at a foundational level what AI agents are and how you can start to explore and use these in your business. Before we get started, make sure that if you’re watching this, go ahead and subscribe to our channel on YouTube because we’ll be dropping a ton of more videos over the next few weeks, leading up to the summit to post-summit content that is capturing what we’re going to be creating here over the next few weeks. Chris, super excited to have you on and joining us just a week before the summit kicks off, welcome aboard.

Hey Darby, thanks for having me. Absolutely, and I want to quickly, in your own words, give people that high-level background, like one or two minutes of who you are, what you do, how you got into AI, and where your focus is right now with AI and where you’re seeing things as we’re prepping for 2025.

I’ve been a software engineer for the last 25 years, that’s been my career. Then when I started using Jasper, basically about the time when you were teaching people to write books with it, that’s how I was introduced to it, and I was paying ghostwriters basically to write content for me, blog content that I was planning on turning into a book for my Amazon brand — I’m also an Amazon seller on the side, so that’s how I got involved in generative AI, through Jasper, joining your community, and then put my head down and got into the weeds for the last two years just learning the stuff and building with it. My first product that I built, my first project, was Ghost Rider, which was a book-writing tool. You could write an entire book with AI in about 40, it ended up being about 45 minutes, but so that was my first project. Now I’m moving on, I’ve become an automation expert combining that with AI and then building micro products, and you and I have worked on a few things together as well, so that’s basically where we’re at.

Yeah, I know, I love the progression over the past few years where we started working on just some projects here and there — what was exciting, what did we see maybe some potential in, and that really ended up evolving into us collaborating to create the MarketSauce system that many of the members of our community have used. But also through that process, I feel like we’ve developed a much deeper understanding of how to use platforms like Make.com to help tie a lot of these different tools together to create stuff that maybe it’s not just going to ChatGPT to ask it a question about who’s my audience, but you’re pulling in things using Perplexity, you’re scraping Reddit for what people are saying in real time, and tying all these pieces together.

It seems like now we’ve been talking about AGI, artificial general intelligence, for some time, people say it’s coming, it’s coming, it’s in the future, it seems like it’s coming faster than we’d expect, which isn’t really surprising given the speed and pace at which AI progresses. But now agentic AI is the new buzzword so to speak, and there’s some really powerful demonstrations coming out — just a week ago Claude is now able to click and type on your own desktop and perform actions, and I think this is just a taste of what we can see coming. Can you give people that maybe are coming into this with a minimal understanding of what AI agents are, agentic AI, what does that look like now versus what do you see the possibilities of AI agents and how that is going to implicate for our businesses?

Yeah, so up until now it’s been pretty much automations where you integrate, you take repetitive tasks, rule-based tasks and workflows, and you can combine those with AI to generate content, and they’re simpler and less complex systems, and you basically create predictable workflows where the outcome is expected based on a defined set of rules. Now with the newer LLM models, before, when you would develop a task-based output using a large language model, you have to prompt it to coax the steps out of it. Now though, with the newer system-two models, they’ve been trained on the tasks that get to the result instead of just the result itself, they lend better to AI agents which handle more complex tasks and workflows, and they actually do some decision-making and can handle less predictable task frameworks, basically. And then you want to include the human in the loop too there, but yeah, now with one prompt you can pretty much get a whole workflow, and then have, at each step of the way, that LLM make decisions based on those steps.

It’s interesting because to tie it back to where we started with Jasper at the beginning of this, it was a template and maybe a prompt and you’d get an output, and it was basically like predicting based on what you told it you wanted — here’s what I’m generating as an output based on that, and Jasper would fine-tune how their templates worked and people would put different context in. It’s changed obviously a lot over the past two years, but even when we were starting to use Jasper and figure out what these different prompts look like, if you string them together, you took that and ran it all the way to writing an entire book out of it, which is like a series of different actions inside of it — okay, I need an outline, I need context for each chapter, and it’s a series of these prompts that I guess we would call command stacking, people called it different things, but you’re stringing these together and it’s building on its context.

But now, a lot of the AI from what I’m understanding, the way you’re describing it, just to clarify for everyone listening, is that the newer large language models and how they’re being developed, they’re not necessarily needing as much baby-stepping in between steps. Say I want to write a book that’s in this non-fiction style and it needs to be 10 chapters and include an author bio right before, that’d be like maybe 20 different prompts that you would need to string together and go through each one individually. Now with an AI agent you could just tell it that’s what I want, and it’s going to have its system help it with solving the different tasks in between there, writing its own prompts, maybe iterating on those prompts as it’s building towards it, and it’s taking more initiative on its own to accomplish that task. Is that a good way of looking at it? Books are one example, but this could be YouTube videos, this could be creating SOPs for internal processes — is there any other use case that you could see agents being really effective, especially for small business owners that are looking for ways to leverage these agents to help them with scaling smarter and not necessarily hiring what they might have needed a full team of humans for, and agents are going to be able to accomplish a lot of those tasks on their own?

Yeah, with the new GPT-4o framework that just came out recently, you could actually see when you give it a prompt that it’s actually doing some thinking there and some reasoning — it’s simulated, obviously, but you can actually ask it after you get an answer to walk you through the reasoning that it came up with, whereas before you used to have to walk it through the reasoning through your prompting process. So like with the book example, I’d have to go through the predictable steps to write a book — you could actually in GPT-4o, probably with half the steps, you could basically just say I need to write a book on this and it would go through the necessary steps on its own. Of course you’d run into a token limit there, which is why when I created Ghost Rider that was the biggest problem, the output token limit for writing a book — there’s no way you could do it with one prompt, but I think we might be heading there.

So for the book example, you could have an agent, or a swarm of agents — a swarm would be a team of agents that you could build where each agent has a different role or task. You’d have one agent that would write the draft, one that would play the role of an editor or copywriter, one that would play the role of a researcher, and one that would play the role of a publisher or illustrator, so all the steps required in writing a book could be done with a swarm of agents. The publishing part is a little tricky, especially if you’re publishing on Amazon, because that’s hard to automate, you basically have to publish yourself by going through manual steps, but I think they’ll probably change that at some point. With the Claude example you gave, you might be able to train Claude to go into your browser and help you go through those manual steps to go through the publishing process.

Yeah, and you mentioned when we’re talking about Claude and OpenAI, or Claude from Anthropic, right, OpenAI ChatGPT and any number of these tools, how critical it is to keep a human in the loop, but where that human fits into the loop I think is something we’re going to continue to see evolving, because the more you refine the agents you have on your team, your swarm, your AI workforce that are doing these jobs for you, once they’re refined and the system is working, there’s less human oversight that’s needed I feel like, to make sure they’re doing what they’re doing, because over time they just get so refined and it needs less and less quality control once you train it on your process and give it feedback on what’s the correct output or outcome, then they refine their processes so they can learn from themselves from their previous runs basically.

That’s interesting, because now you can basically train an agent to learn from itself, to provide itself constructive feedback based on how you would approach solving a problem as well, and this really is just going to create an infinite level of ways that you can leverage and create these sorts of agents. While maybe we started creating even the prompts for these agents with our own text and our own thoughts, this is the prompt I want, often now I just go to Claude or ChatGPT and say hey, this is the type of prompt I want, can you optimize it for me so that I’ll get the best result, and it’ll do that. You’ve — I’ve seen you create automations that will create other agents on the fly when they’re needed, oh we ran into this issue, okay I need to create an agent that will do this one thing to solve this issue, and then you’ll just erase the agent after it’s done because it’s not needed anymore in that specific process.

Yeah, I think that’s where we’re going to end up, where you have one agent that kind of oversees everything, and you basically prompt that agent with a prompt that says create another agent, or create a team of agents that write a book, for example, and it would know what to do, it would generate the prompts, it would spin up those agents, write the book, and then retire those agents. So you could actually build software on the fly that way, and some people are saying even the business of building software is going to go away because everyone’s going to be able to tell an agent to build software for a single use case or a single instance where they can use it once and just throw it away, or maybe spin it back up to use it the next time. But basically you’d have agents building everything — agents building agents, teams building teams — and so it’s endless what you could do with agents.

With that in mind, can you walk us through maybe some of your process, how you go about training, anything tactical that you use in your current workflow to go about creating some of these automations, and maybe we can use a simple example, but how would you go about setting up a flow like this, without getting too technical on this call — I know we have some stuff reserved for the summit session you’re going to be going into even more detail on — but could we walk through what setting up an automation like this looks like, just practically, how would you go about setting up a flow like this for writing a blog post, for example?

Yeah, so the first step would be basically you’d start with the prompt. I would start in ChatGPT first and test out prompts in ChatGPT to determine what’s going to give me the best output. Then I would go to Make and create a variable that contains that prompt, and then say, to begin with, I would start with an outline for a blog post, so it would be a prompt to generate an outline for a blog post. Then the next module would be having OpenAI write that outline, and then I would take that outline and add that to another prompt that would say, an iteration — if say the outline, I would have it count how many sections are in the outline and then have an iteration set up where it would go in a loop through that outline and write each section of the blog post. At the end of each loop I’m collecting each of those responses and assembling them into a final blog post, and then I could even automatically put the human in the loop at that point to evaluate the blog output and either approve or deny that output. If I approve it, the training part would be throwing that example back into the original prompt as a successful example so that the next time I run it, it’s going to be closer to what I want. So that’s basically the training process in an automation, and then you could go through that process several times where you have several examples and use that as the training examples until you get to the point where 99% of the time you’re getting the blog post you want, or the format that you want.

Is there a specific number of these training examples that you like to start with or build towards as you’re testing out these automations, like you set up three and normally that’s enough, or is it more just the smarter these systems are getting, is more just better in this case as long as the examples are quality?

I think it depends on how you’re evaluating the examples. If say you want the blog post to be formatted properly and you want it to be in your tone and style, it may take more iterations to get to the point where you’re satisfied with the result, but you could also do all that example generation in ChatGPT, or another LLM, or take it over to Claude to generate those examples and pick the ones you want from those to throw in the original automation.

Yeah, I found that to be a nice accelerator to sourcing these examples — if you know I want to create an example of this type of thing, you may or may not have created that yourself, or you may be looking at a client and extracting their content, but just like you said, a lot of times you can ask the LLMs to provide you examples, select the ones you want, and that can help accelerate providing these sorts of trainings and documents. Another question to follow up on that, I think we probably covered some of it but just to make sure we’re being comprehensive on this aspect, because I think there could be some confusion about when do you let AI just do its thing versus when do you really want to make sure you’re feeding it the right information. Are you finding it more effective when you’re adding examples to the prompt specifically, so here’s the prompt and here’s all the examples and it’s part of there, or is it more effective at a certain instance to have its own knowledge base that it’s referencing — I have a Google Doc maybe that has a hundred examples of types of blog posts I wanted to reference — are you building that into the prompt or is it more effective to have this own knowledge base that you’re having as a reference, can you share a little bit more about how you’re setting up this training?

Yeah, so there’s two ways you could do it, and I have better quality results from using threads. If you think of a thread as like a ChatGPT session where you’ve started a conversation with it, you’ve generated several examples, then, so in an automation you can do the same thing with knowledge retrieval in the OpenAI backend, I’m finding that it’s not as accurate as if you use a thread. So in an automation you start with a thread, which would be like a ChatGPT conversation, but you’re managing it in an automation, you feed those examples into the thread so it’s in the context of the conversation, and then you start with your prompting, so you say, consider these examples, I’m going to be prompting you but I want you to consider these examples first at the beginning of the conversation, so it has those examples, and then as you run through the automation, you’re prompting in the same thread, so it has all the context and history of that thread, and that gets you to the result at the end of the thread with all that context. The knowledge retrieval stuff, especially with regard to OpenAI, it doesn’t seem that you can really control what it’s accessing — a lot of times I’ll throw research into the knowledge base, you can do the same thing with threads, but I just find that the knowledge retrieval is more accurate with threads.

Why do you think that is? I think it’s the way that OpenAI has their database set up. I can basically say consider this piece of information, as opposed to when I ask it to retrieve from the knowledge base, I can’t tell it what to consider, so I’m not really sure what’s going on there, but I don’t have any control over what information is being retrieved from the knowledge base when I use the knowledge-based method.

We’ve heard the term RAG in AI and how that relates to all this, can you expand a little bit about what that terminology means and where that applies specifically to what you’re talking about with threads or other ways of retrieving knowledge? Yes, RAG stands for retrieval augmented generation, and it uses vector databases. The retrieval is more related to the words that you’re using in your prompt, so it uses vectors as an index — if you can imagine a vector, it’s a direction in a multi-dimensional space, it’s hard to wrap your brain around, but basically a vector represents a relationship between the words in your prompt and the words in the knowledge base, so that’s how it retrieves what’s related to the words in your prompt. With the thread-based method I think the thread is actually stored in a vector database, but it’s a limited set of data, it’s just limited to that thread, it’s not everything that you dumped into the knowledge base, so there’s less to consider in the thread knowledge base versus the big knowledge base that you’ve dumped all of your data into.

I see, and with that in mind it brings up the point that when you’re creating your own knowledge base, or creating any sort of application, more isn’t necessarily better — you want it to be the right quality of information, not just everything you could ever know, because that could almost muddy the pool at which AI is retrieving this. Do you view threads as their own custom GPTs that serve a very specific purpose, or would you view a custom GPT as having more capabilities or more features that you would utilize specifically with that terminology?

I would see a custom GPT as more of an assistant or an agent, and the thread would be a knowledge base, so the thread is the context of the conversation. Understood, so when we’re talking about threads, custom GPTs, agents, chatbots, it seems like a lot of these might have some similarities and crossover, and maybe depending on how your lingo, how you’re approaching this, there’s some differences in different circles. I’m trying to get a really clear distinction here of where you see just chatbots as they’ve been a thing for years, but what’s the difference between a chatbot and an agent specifically — chatbots can still retrieve knowledge, they can do all this stuff, there are different uses obviously for both of these, but fundamentally is there a key distinction that would classify this as a chatbot versus this is going to be an agent, is it something that has to do with it being more autonomous, or how do you view that distinction specifically?

Yeah, an agent’s more autonomous and it has access to tools, so you can define a set of tools, and you can do this in a GPT as well, you can set up function calling in GPTs, most people don’t know how to do that because it requires some technical knowledge to set up, you have to set up schemas that define API calls, which are internet calls to services that you’ve set up on the outside, or that are already set up, and then once the GPT, which is more of an agent at that point when you give it function definitions, once the GPT knows it has access to those tools in common language format, you can just say — for example, say you want to save an email address to your CRM but you want to do it from a GPT, so you set up a public GPT that you allow everybody access to, they can go in there and build an offer, but before you let them use that, you ask them to share their email address first. That GPT has access to a tool in the function-calling area that allows you to, once the user shares their email address in plain language in the prompting, you can just say save their email address, and it knows to use that tool based on the definition of that tool, and it’ll automatically call the API to push that email address to an outside service, like Make — you can set up a webhook and Make acts like an API, it basically is an API. So for the user, once you have it set up, they just say I want to save this user’s email address once they give it, and the GPT automatically knows it’s to use that tool because you’ve defined it beforehand, and it pushes that email to a Make automation.

I could see that being an extremely simple process to do for very simple automations that will speed up, still something that would be very repetitive, but where does it end, and maybe the answer is who knows, when agents start to take on and do tasks and assign tasks and continue to be more autonomous in that regard. Let’s tie this back into voice — you’ve been working with some cool voice agents, I think voice is a lot of, obviously, how we interact and communicate as humans, but even now just with ChatGPT the voice feature is shockingly fast, to the point where even six months ago the latency might have been — I don’t know if it’s quite there yet, I just asked it to write me a rap song for my birthday and it came up with one on the spot with basically no delay, and that was even a month ago. I don’t know how that rap song would have improved if I’d asked it today, but it seems to me that voice is the way in which a lot of us are going to interact to even start one of these automations going. Can you talk or speak a little bit to what you’ve done with voice to date, right now where you’re experimenting with it, but where do you see continuation of the voice and that technology from your vantage point, because you’re much deeper into the development with these tools — I’m just curious right now, it’s October 29th, 2024, what do you think people should be paying attention to in terms of voice today, something directly applicable to their business versus where do you see things going over the course of 2025 related to voice?

The simplest use case I can think of for voice right now that is something people can implement right away is the ChatGPT advanced voice feature for simulations. Say you’re preparing for a job interview and you’re going to be interviewing with the CEO of a company and they have some public videos or transcripts you can get a hold of, you could dump those transcripts into a thread in ChatGPT, ask it to simulate, to play the role of that person, and then say I want to simulate a job interview with this person, and have the voice walk you through the job interview, and then at the end ask it, as it’s playing the role of that person, how you could improve or what you could say that would make a difference to that person based on their personality or communication style. That’s something everybody can do right now using a GPT or just having a chat session with ChatGPT. The other thing is everybody has to fill out a form when they’re providing information to any kind of online application, you could just completely eliminate that with voice and it makes it a lot easier — I think filling out a form is a pain, I don’t like to do it, I’d rather talk to something extracting ideas, content, submissions of any type, application context.

Just do you want, or just where they’re on the page, without even a phone call, hey just select this and we’ll ask you some questions, you just have to respond, and we’ll take those, and on the back you’re talking about function calling doing all that on the back end, so that it’s sending it to wherever it needs to go and getting filtered through some sort of system. Yeah, and the assistant’s smart enough to extract that information really without any programming knowledge, once it understands what tools it has access to, it’s going to know what to do with that information, all you have to do is just tell it in language. One way we’ve experimented with this, and we’re toying with it right now with the MarketSauce program, is take the job-interview instance, but on the other side could be market research, or employee understanding — another example we saw with one of our clients this year — you want to hire better talent or retain your top-tier talent, how well do you understand the employees that work with you, the people in your company, you can use AI to generate reports that help with getting inside the mind of what’s making them tick, but how are you actually communicating effectively with that person, or potential prospects in your business, or attracting the right talent into your business, are you taking that same approach you were describing with the job interview but simply interacting with, what if I created content about this, what would my prospective employee or recruit or customer or client — how might they respond, what’s a better way I could say this that would probably land and resonate with them more, given the context — and the possibilities are really endless, but we have job interview, market research, employee research, recruiting, I think is a huge field across the board where voice is going to disrupt a lot of things, but ultimately I think a lot of it’s coming back down to how do we communicate more effectively human to human, and then where’s the AI inside of this process to help us with developing that deeper understanding, parsing out the information we need to know on a need-to-know basis, or eliminating just tedious back-and-forths of filling out forms manually, prompting and doing what still might have been an incredibly effective workflow that might have taken an hour a year ago with AI, and now it’s a matter of a couple minutes, which is the reality in a lot of these situations, so I think it’s exciting, I think there’s also implications maybe we just haven’t even begun to consider about what this could lead to.

I’m curious from just a standpoint of AI and ethics, where are maybe the things we should be looking out for in terms of, we should just be aware of where some people may be using this right now — obviously with phone calling, voice automations is a big thing that has a lot of attention on it right now in terms of spam filters, FTC, that sort of thing — what are some things people should be aware of, maybe the flip side of the equation, what are the bad actors maybe using this stuff for that we should be aware of so we can pick it out and stop it from happening to us, if someone is trying to do something nefarious with these tools?

Yeah, even from a data and security standpoint, you say your voice, it’s out there now, what’s this agent doing with the content I just gave it. With voice cloning, ElevenLabs, you can clone people’s voices, and ElevenLabs now has put security processes in place where you have to recite a script in order to train a voice, but there are a lot of open-source things you can run on your computer to train a voice, you could train a voice from a recording. So if you have your voice out there, or your kids, say they’re on YouTube or TikTok, what I would do to prevent people from — I don’t think you’re going to be able to prevent them from cloning a voice, but you should set up a system where you and your kids, if you get a call from your kids and there’s something going on, or you’re suspicious, you can ask them for a passphrase or code word or something that proves it’s them, if you have any question that it’s them. That’s what I would do, I would have some kind of protocol in your family in place that would allow you to thwart some of these instances where people are trying to extort money — they have a kid call and it’s an AI voice pretending to be their kid saying they’re in trouble and need money to get out of jail, or whatever, you could have something like that set up to avoid situations like that. With regard to commercial calling, I think it’s important that they have to opt in to be able to get a call, an initial call from an AI system, so anybody that you’ve done business with, you have them check the terms of service and agree to that before you call them with it, but I don’t think people are going to be — I think if people see the convenience of it, employment pre-screening is one thing I’m thinking of off the top of my head, you tell people, hey, we’re going to pre-screen you before we do a job interview, you’re going to be talking to an AI agent, I don’t think most people are going to be against that, especially if it’s a big time-saver, and as long as they know they’re talking to an AI, I don’t think it’s going to be a problem, it’s going to be the future anyway, everyone needs to get used to it.

Yeah, I would agree it’s going to be more and more commonplace as AI is just more and more in our everyday life, this is just a thing, but what you say about the safe word, that was something my parents had with me when I was a kid — hey, if someone comes to pick you up from school, they have to say this to know they’re actually with us, versus the opposing — but I think there needs to be a level of just general awareness that people are really considering across the board with these types of tools, just to know this stuff’s out there, and these types of voice call, phone call, scam-type things, those were happening before it was all with AI voices, and it does seem that with what we saw here recently with cloning yourself, an AI clone of yourself that can hop on Zoom and interact with people, we’ve seen some interesting stories over the past year of that happening, some weird stuff happening from cloning yourself showing up to these sorts of events. It’s hard to really tell where things are going to land with it, but I think in general having a hyper sense of awareness around who you’re interacting with, and confirming that’s even an AI system you want to be engaging with, is just something everyone should keep top of mind when engaging with this new age of doing business, and with generative AI being so early on as it is, it can be scary where the stuff is right now technologically, but also on the flip side I think it’s also really exciting, and there’s just so much opportunity that’s going to come from this innovation.

Which is again why we’re putting on the Scale with AI Summit this next week, showing real, practical ways that we can — not just see these innovations and look at that shiny object, wow that’s so cool, but what can I actually do with this, what can I actually do to implement it right now. I’d love if you could share just a little sneak peek at what people next week that are tuning into the summit — and of course the summit’s going to be free, and that interview and workshop we’re going to be doing next week will also be available on our YouTube channel if you aren’t able to catch that live session — but can you just walk us through what is the demonstration you’re going to be presenting in that workshop as it relates to AI automations and agents?

Yes, I’ll do an introduction to AI agents and what they are, and then we’ll move into creating a voice AI agent that interacts with your CRM, and we’re going to be using the GoHighLevel CRM, a lot of digital marketers use that platform, and we’ll build an agent that allows you to do autonomous calendar management, so you’ll be able to schedule people, book appointments, do follow-ups and reminders for appointments, you can set up customer service agents that answer questions, do customer support, and one thing that comes to mind is an e-commerce example where you could actually have an agent check your inventory levels or answer questions about a product. We can trigger automations with them, so you could build your own tools and automations that could be triggered, you could essentially tell this agent, hey I want to write a blog post, and it would trigger a Make automation that would write a blog post and post it to WordPress. We talked about simulations, preparing for interviews, you could even do sales-call training simulations where you’re training your sales team on how to talk to an avatar, like you and I were building avatars from the MarketSauce product, you could have a sales team target a specific avatar from that report and have a sales conversation with them. We talked about replacing forms, and then some other use cases are you could do surveys with voice, and then I have some coaching clients I’m building text reminders with so they can keep their clients engaged with setting goals and achieving goals through daily reminders via SMS that help them stay on track for the week, based on — you could base it on a coaching call where you take the transcript, dump it into an automation to analyze the goals for the week and create these text messages that go out daily. So that’s several ways you could use this technology.

That’s very cool, one I’ll add, I heard about it a week ago and thought it was really interesting, in the e-commerce realm is inventory forecasting, supply chain management, being able to pick up on when trends are spiking in sales and identifying when you should initiate new orders and new communications between your suppliers, identifying potential pitfalls, just operating in a more sophisticated manner of managing your supply chain to prevent inventory falling out, or hey maybe we should increase the price, slow down our sales velocity on Amazon, to make sure we don’t run out of stock, which is detrimental to us.

Yeah, so for — I use Shopify to augment my income with my Amazon brand as well, but just the customer service emails, problems with a specific order where they got the wrong thing and want a replacement, you could build automations on the back end that plug into Shopify that allow you to handle all that, have the agent look up an order and basically hand it off to a human, or handle it automatically if it’s simple enough without human intervention.

That’s great, and getting some love in the comments for the coaching-call-to-SMS-reminder example, simple, super achievable, and a great way to create engagement and personalization, and as people have been saying for years now that personalization is the future, right now we’re at that point in time where a lot of the communications between customers and clients can get personalized, but I don’t want to overlook the human in the loop that needs to be in there, and making sure you’re setting yourself up to have those types of communications, because there may be one side of the equation where it’s great to automate and do all that stuff, but if you lose that human touch and element, I think there’s going to be a lot of people, at least in my opinion, that are going to actually gravitate towards interacting and having the ability to talk to a human, versus getting caught in a loop of this AI is just not doing it for me, depending on what business you’re in.

Just keeping that in mind that, at least for me at the end of the day, AI is great for so many different things, but ultimately how are you freeing yourself up, or your team up, to focus on those things that are the highest value, which usually is building relationships, creating value, producing assets, that sort of thing, and just in general a reminder that these tools are here, everybody, and this is stuff people are doing right now actively, and if you aren’t touching AI at all in your business and you’re just getting started, dipping your toe in the water to experiment with things, there’s a lot that could happen. So starting out for someone coming into this who may be overwhelmed with where do I even start with an AI agent in the first place, what would you recommend people focus on first, to make sure they’re getting something of value and something that’s clearly tangible in terms of a business outcome — any best tips or recommendations as we wrap up here, Chris?

I would start with ChatGPT, just learn how to use it, learn how to prompt, try having it play different roles in your company, see how you can use it to automate the simplest tasks you hate to do. You don’t have to really know how to code or anything, most of these things, and the stuff I’m going to be showing you is all no-code — you don’t have to know how to do any coding, you just drag and drop things onto a canvas and set some properties, and even a lot of these platforms have AI in there that’ll help you build a task framework. The thing I’m going to show you, you give it a prompt and it builds the whole task framework for the agent, so you don’t even have to know how to do that, but that’s where I would start — use ChatGPT, just learn how to do it, pay the 20 bucks a month, use the voice, I think a lot of people aren’t taking advantage of the voice thing for simulations, and that’s where I’d start, is just start using ChatGPT.

Absolutely, and I think it’s a great place for anyone to start getting used to this, and a reminder on top of that, just from my experience with ChatGPT, treat it like a friend sitting across the table from you that you can bounce ideas off, and if it’s wrong you can tell it it’s wrong, but in a lot of cases it helps to have some understanding of the field you’re talking with it about, to make sure you’re able to pick up when it may be going off the rails a little bit, or if it’s on track, whatever your field of expertise is — obviously if I’m a doctor in psychology and I’m asking ChatGPT about my field of expertise, I’m going to have a much deeper understanding, and if ChatGPT gives me a hallucination, makes something up that isn’t true, somebody needs to be able to pick that up, and sometimes AI can sound very convincing as well. While a lot of these responses we can get from AI are fantastic ways to speed things up and get information and conversation going, I still think it’s really critical to validate and verify some of this information, whether it be going to confirm that this is how my customers would respond to this type of message — obviously you can practice and train, but getting that data back to validate what’s true and what’s not is very critical, and just be aware that the tools are powerful but we’re still super early on this, and go at it with what I call productive play, a sense of having a little bit of fun when you’re trying out these tools. If you’re just getting started and overwhelmed, just have fun with it, play around, see what you can do and create, that maybe isn’t attached directly to something that has to be a business outcome, but by going through and practicing and experimenting you’ll start to understand how you can flow back and forth with these tools, and with daily consistent practice, even just getting into the flow of how you use ChatGPT day-to-day, you’ll be that much more effective and empowered with AI to achieve the results you want in a much faster fashion.

Super looking forward to your session at the summit next week, Chris, it’ll be recorded, we’ll be posting that summit back to our YouTube channel once we’re all wrapped up and done in the month of November, and you guys can go ahead and register for that at Scale with AI Summit dot com — that’s ScaleWithAISummit.com, register, it’s free, have a load of great information, valuable resources, and fantastic speakers, and in part made possible by our sponsors for the summit, HOI.com and Gamma.app, which we’ll be sharing a lot more information about who our sponsors are throughout the event and throughout the rest of the videos you’ll see here during the summit. Until then, we’ll see you guys on the other side, thank you for tuning in. Chris, as always, thanks for joining us, I’ll see you guys on the other side, keep coming on, cheers y’all, bye.