Kenneth Edmonds breaks down how agentic AI agents handle your routine tasks, so you can scale lean, run smart, and finally get your time back.
GROWTH PILLAR: AI & Automation
WHO THIS IS FOR: SMB owners / Solopreneurs / Corporate escapees / Leaders building systems
WHAT THEY'LL GAIN: Practical insight on deploying agentic AI, reducing bottlenecks, scaling without hiring, and protecting business data — straight from someone building it in the real world.
Most small businesses hit a wall. They want to grow but can't afford to hire. Kenneth Edmonds, founder of 22nd Century Management, has a different answer — agentic AI.
Unlike a basic chatbot or custom GPT, agentic AI operates independently. It makes decisions, books appointments, handles calls, drafts emails, and manages routine tasks without waiting for your input. Kenneth calls it the next layer of AI — and he's already building it.
In this episode of Knack 4 Business, Kenneth returns for his Season 4 appearance to break down exactly how these autonomous agents work, what makes them different from automation and GPT tools, and how solopreneurs and SMBs can deploy a suite of agents to free up their teams and scale without adding payroll.
Topics covered:
Kenneth is actively looking for beta testers right now.
Reach him directly at ken@22ctymgmt.com or visit 22ndcenturymanagement.com.
Connect with him on LinkedIn or explore his speaking work at 22ctymgmt.biz/speak.
SPONSORS:
Tired of duct-taping your CRM together? Canada Growth Network runs on full GHL infrastructure — CRM, automation, community, and real introductions. $1 CAD your first month, $47/month after. No contract.
Find any K4B episode in seconds. Web Indexer is the smart chatbot built right into the Knack 4 Business site. Search by topic, book a meeting, get answers fast.
Investing in real estate? Property Wizard by Gentry Learning is the podcast built for Canadian real estate investors who want practical, no-fluff guidance.
Want to connect or collaborate? Reach out: info@kreativinsight.com
Join a community built for people who actually do the work. Canada Growth Network — $1 CAD first month, $47/month after. No lock-in.
Listen on your favourite platform: Simplecast
Subscribe to the K4B blog: blog.knack4business.com
Search all K4B episodes by topic: knack4business.com
Want to be a guest on Knack 4 Business? Register here: tinyurl.com/mcdm3hk6
Bernie (00:05)
Question of the audience, are you stuck doing work your business should be doing for you and wondering how to finally scale without burning out or hiring a bigger team? That's exactly what our guest will dig into today.
is here to show you how Agentic AI can take real work off your plate, streamline operations, and help your business run smarter, not harder. If that sounds like the relief you've been waiting for, stick with us and lean in. Here's a conversation you want to hear.
Our guest today is
Kenneth, the visionary founder of 22nd Century Management, returns to the Knack 4 Business visit podcast after his popular season three appearance. Business optimization expert reveals top tips for sustainable growth
Known for his bold story packed career from flying in the space shuttle pilot training program to chasing a train through Mexico. He brings the same fearless energy to transforming how businesses run. His firm is built on simple but powerful principle, under promise and over delivered.
Kenneth blends deep operational insight with practical hands-on leadership to help organizations streamline processes, elevate performance, and build sustainable growth. Respected as both as a speaker and a trusted advisor, he's the person leaders will call on when they're ready to sharpen execution and scale with intention. Absolute pleasure having you back again, Ken. Do you have a favorite quote or saying, sir?
Ken (01:20)
Well, I love that one under promise and over deliver. I think that is the key to being successful in business. You know, it's the basic foundation of customer service.
Wayne (01:28)
us.
Bernie (01:29)
So, agentic AI, us a definition of what it is. I have in my mind what I think it might be, but you know what? We'll get it straight from the horse's mouth. Over to you, Ken.
Wayne (01:31)
Cause the
Ken (01:39)
OK, so the best description of agentic AI is an AI agent that operates independently capable of making decisions on its own. Now in the background, obviously we've given it the parameters and the instructions on how to do it, but it doesn't need our day to day guidance. For example, an AI voice receptionist would be able to fulfill the functions your receptionist would be does right now today, and then it would be able to make the decisions she would make about whether to schedule an appointment as
or to cancel an appointment, to transfer a phone call, those kinds of things, but it doesn't need any human intervention.
Wayne (02:13)
you
Bernie (02:13)
Is that
something that comes with most AI platforms or is this something that you have to buy outside of that?
Ken (02:19)
No existing platform comes with it. We're right in the very early stages of agent AI taking off.
I was up at a conference in Orlando called IT Nation, and we were there for three days talking with and learning from MSPs and people that are foundational in the managed services sector of IT world, I guess is a good say. They're the people that run the networks and help companies make progress.
And we had 14 sessions over three days based on AI and agentic AI. Most people are familiar with chat GPT where you ask it a question. That's the very simplest form. The next form is where you create a custom GPT that's really designed to do what you want to do. For example, you can teach it to talk like you, to think like you do, to ask questions like you would ask them, but it can't operate independently. You have to tell it what you want it to do.
Agentic is the next level, and it is going to be the one that is going to have huge impact on jobs and employment. it's the it does also, as you kind of mentioned earlier, it is a tool that allows businesses to scale. So, for example, a solopreneur with a full suite of agentic AI agents might be able to have a board of directors, a virtual marketing agency, you know, an executive assistant.
those kinds of things to help the solopreneur or the small business person to have the capacity to scale so that he doesn't have to do the mundane things that would tie up his day, we could say.
Wayne (03:50)
And how do you decide what are the mundane things?
Ken (03:54)
A mundane thing would be anything that can be done by a machine that doesn't require a human. So for example, I'll just walk through one example with the executive assistant. So once the agent knows what information that the executive wants to know off the internet, so maybe the executive normally comes in and reads some journals and some online news about maybe it's AI, maybe it's about something specific to their business.
Well, what the AI can do is read all of that before he comes in and generate a summary based on the depth of information he's interested in. And so rather than having to hunt down all of the sources, he has like a one page briefing sheet that highlights the things he was interested in. You can do the same thing with, you know, like emails, a good example, if it knows his email style, just like in the past, a business owner would talk to his secretary.
and say write a letter to Bernie and tell him that I really appreciate being on his show.
And that's all he would tell his secretary, but she would know how to write it the way that Bernie wants it wrote. And she'd type it out, put her initials down at the bottom, take it in for his signature. Well, your AI can do the same thing. In fact, for those, the executive assistants actually modeled off of something that those of your audience that are our age probably would remember. Anybody that saw the movie Space Odyssey, 2001 Space Odyssey, would remember Hal.
the computer on board the ship. And so you could think of this as HAL version 3.
Bernie (05:18)
One of the things I've found, and I was just, I was using AI over the weekend to refine some process. And I said, hi, here are some documents, look at them. Let's do an evolution of the next iteration of what I should be doing. And it says, sure. And asked me a lot, it came back with lot of prompts, et cetera, so I can tweak it. And I say, okay, produce.
Well, I got it was supposed to produce four documents and got one out the gate. And then the other three, it just said, it said, what happened? So we'll still working on it two hours later. And where are we at now? Still working on it. And I'm there's something wrong here. Like it wasn't giving me, it wasn't giving me a feedback loop that I was expecting. And so where I'm going with this question, with this, this line of thought is
You know, here I'm focusing, I'm acting like the AI bot, you know, asking the AI mothership, you know, work through this. How do you guarantee that the agentic model will get a result that you're expecting? And if it can't generate the result, know, flag you and say, listen, this is going to take, you know, 20 minutes, 40 minutes, it's going to be a tomorrow event. How do you sort through that? You know, that's annoying to, uh-huh, okay.
Ken (06:29)
That's an interesting question because it starts with the design process. There are always going to be times when a situational arise that may not fit what you've taught the agent. And so the first thing is, if you can't solve the problem, let the operator, let the user know. That's first piece of it. The second piece is to try to
think through how you ask the prompt as part of it, because prompt engineering is kind of a science to be able to get it to do exactly what you want it to do. And then sometimes you just have to, a lot of times if I get it hung up in a loop, I'll just type a question mark and it'll try to think about that question mark and that breaks the loop. I have that happen sometimes when I'm, because ⁓ I use AI to help me code, because there's some things that I'm not a programmer in. I'm not a Python programmer, but there are lots of
times I need Python code. So I describe the function I want and I let one of the AI agents create the code. And then we test it and we test it and we test it and we try to break it. And if I broke it, then it's okay. How do we go back and make sure we don't break it again? Because that's a key piece of agentic AI is it has to be, because it operates independently, it has to be pretty much bulletproof. Or if it runs into something it can't deal with,
what to do about it and kind of have a way to get out of where it is. Most of the time though, these agents are going to be used for things that are fairly routine. answering a phone booking an appointment, as long as you can tie into the CRM, that's not a lot of complexity in that. They say, want an appointment. You ask them when and they give you a date. So you find the times are available and you give them a choice and make the booking.
So that's a fairly simple application. It's not gonna run into things. With the ⁓ executive assistant, that's gonna be more challenging. And a lot of it will require probably custom code for the individual that's using it. Because you have to know how they write. And some of that is stuff that you can create a routine that would import that and it could learn from it. But there will be things where it will probably have to have a conversation with its executive to say, hey, listen, if I do something that doesn't fit,
please let me know so I can make adjustments. Then it becomes to some extent self-improving. A good example right now is when I create AI agents, I have a really strict prompt that puts some guardrails on what it can do. Because I found out recently that I would give it something that was mostly working, and then it would redo it and then give it back to me and it wasn't working. I had to put a guardrail in.
says you only make surgical changes. You do not completely rewrite the entire program just because you think you can make it better. I mean, it's never malicious in what it's doing. It's just that it sees something it thinks can be better, and it may not know how it fits into all the other pieces. And so it changes something. I've been fighting a little bit of that this morning, in fact.
Bernie (09:16)
good, I'm glad I'm not the only poor SOB out there going, that's not what I asked for. Come on, try again.
Wayne (09:22)
Can I certainly understand when a genetic gets it right, but how does an operator know when it gets it wrong? What are the red flags?
Ken (09:30)
Well, the red flags are it'll schedule an appointment, for example, on the wrong calendar. That would be a red flag that you didn't, it's not right. And, you know, again, that's primarily a programming in the back office because you got to make sure that it knows which calendars belong to which people in the office, for example. Same thing with a good, another good example might be is you'll have in the office people that are maybe out in the field a lot and want to call transferred to their phone.
There will be other people that want it either to go to voicemail or to have a note taken and then the note come to them so that either it doesn't disrupt what they're doing. So again, as you implement it, you have to think about how do I make sure that the agent gets the instructions it needs? And that becomes like part of the onboarding process.
Bernie (10:13)
So if you have an agenda, can I say, no, all of a sudden, listen, you want a herd of them, for lack of better words. One to book my calendar, one to take care of the accounting map. Do you just, do shop an entire suite or you just, you buy a whole herd of them and turn them all loose at the same time or do you do it in staged launches to make sure that?
Wayne (10:25)
Thank you.
Bernie (10:35)
it's running harmoniously, you know, first one runs okay, and now let's plug in the second one, if something fails, go, okay, I the fail point, go back, fix and so forth.
Ken (10:44)
So I would say in that, you know, I would suggest buying a suite, you know, or find somebody that's developing a suite because not a lot, there's not a lot of these out right now. You know, that's one of the reasons I am on this process is because I didn't see a lot of people doing the ones I'm doing. It's interesting that there are other, there are agentic AIs that are used other places in business, like in process control and stuff, but there's very few of them that are designed to supplement the office staff.
And it's not to get rid of the office staff, it's to free up the existing employees to be able to do things that require human creativity and human thought. There are things that are routine and you can just nail it down and here's a process to do it. But there are other things that kind of require, let's say random association. You just start thinking about something, say, hey, we can do this. I'm not big about getting rid of employees, but I am very keen on the concept of giving
existing employees the ability to do more for the company. And what it does is, is in the long term, it lets a company scale without adding bodies. And for a lot of small businesses, that's really huge. There were a lot of times when I had my own business that I couldn't, I couldn't grow because I couldn't hire the people I needed to manage the growth because my cashflow didn't provide for it. They would have made my life a lot better and we could have grown a lot faster. But if you can't pay for their salaries, you're stuck.
And that's the nice thing about agentic AIs is they come with a very low overhead. It's typically going to be a small monthly payment on it to cover the usage of tokens and those kinds of things. And so it doesn't have any benefits. It works 24-7. So it does all of these things that to get a human to do that would be huge. It would be very expensive.
Wayne (12:21)
We always talk about ROIs and low hanging fruit. Can we actually ask an AI what the low hanging fruit they'd like to evolve because it would make the first quickest?
Ken (12:28)
to something.
Yeah, you could. Again, you would have to program to think about it and give it enough data to set for it to look at and say, hey, okay, if we improve this is going to have a major impact. That becomes, goes back to a little bit of even a custom GPT could do that, but we've got to make sure we give it access to the information we need. So a good example, I talked about a virtual board of directors.
Well, that virtual board of directors is not going to be like a CFO unless you give it access to all the company's data. Some people will want to do that, other people won't. But that financial advisor piece of it could do this, is it could look outside at the world around you and look to see what the financial trends are, extrapolate from all the data you can find what the inflation trend looks like, other things that would affect how a business does business.
Bernie (13:17)
is this a new niche market where you're to have an AI consultant that'll help pick your agentic software, your agentic best fit? In other words, one over there is a certain model. Then you have Chatchi BT, have perplexity, have clode, et cetera. And each of them have strengths and pros and cons, right? Strengths and weaknesses. Is that?
where you would come in then as a next subject matter expert going based on what you're doing and what you are, is this something that you would do as a task then to help an agency or a company pick the right product?
Ken (13:52)
I could do that. That's one of the things that I can do because again, like you said, each one has use cases. Claude right now and Jim and I have the absolute biggest context. It's how much data they can hold in one bite, you know, or in one session. And I run Claude out of space five or six times a day. But I use Claude because if I was using some of the others, my context would be a lot smaller and then we would have to go ahead and restart the conversation over and over and over again.
And so that absorbs time and energy. And it's just easier to grab one that has a reasonable operating cost. And then you look at, what's its context? What's it really good at? Because some of them are, like Gemini's new graphics package, very good at graphics. does make some of the best pictures right now. Cloud's got, mean, ChadGPT has got Dali, which does a fairly good job, but still has problem spelling.
which is the spelling is the agent, or is AI's biggest problem when you're doing graphics. Cause you'll tell it what you want it to say and it'll say something entirely different on the sign. know, for auto captions fine. But if you're telling it to create a, like a logo for you and you want it to, to, have your company's name on it, you give it what you want. And the odds of it having your company's name, right. Or pretty small because they're just not very good at it yet.
Bernie (15:13)
So if I'm starting out blank slate, I ain't got no clue about AI. I don't have AI, I might have it, but it's not being deployed with intent. Where should a company start with?
Ken (15:25)
Okay, so that's an interesting question. If you're starting from scratch with AI, my suggestion is to find something that somebody that knows a little bit about AI. And that might be an employee in your company that's doing some other kind of job that you wouldn't expect them to know anything about AI. But especially the modern generation, most of them are learning about AI as fast as they can. Because they're reading all of the signs in the outside world that's saying,
AI is going to take over jobs. know, they see all the people fired from Amazon, all the people fired from a lot of places. Not all of that is due to AI. Some of it's due to robotics, which is closely connected to artificial intelligence. But there are a lot of reasons in this. So they know that the people that, in fact, I saw this thing, I love this. It's not that AI is going to stop you from getting a job, but it's going to put the person who has your skill set and uses AI ahead of you in the line.
because it's just a tool that we're going to have to learn. And I go back and I may have talked about this before, there was a good example was back in the day when there was typewriters and there were typewriter dealers all over and they made good livings taking care of selling typewriters and fixing typewriters. And a lot of them missed a couple of little things happening in the world around them. They missed a couple of guys in Palo Alto in a garage named Hewlett and Packard. They missed ⁓ IBM deciding to put a PC on everybody's desk.
They missed the advent of word processing from Wang. And all of those things separately weren't a big deal. But when they came together in a year or so, maybe two years, typewriter business was gone by about 90%. Because Wang gave them the ability, they didn't have to back up and auto correct. And they could output paper from a picture to out on the screen really quickly and easily. And so we're at that kind of ⁓ inflection point in
I want say in industry and in business where the companies that are slow to adopt AI are going to really outpaced by the companies that are. So right now, early adopters in the industry, because it just lets you go farther faster. And the guy that's sitting there plotting along, kind of walking where everybody else is running is going to be at a disadvantage. Now, it doesn't mean that he can't eventually catch up, but how much revenue and how much profit has he lost?
because he didn't move, because he didn't take action. Now it's like, you know, imagine having the opportunity to buy Microsoft stock the first time it went public. You know, I've lost a lot of money because of that, that failed failure to make that decision. But it's, that kind of thing. It will transform the world in so many ways. And we don't even yet know, you know, where it's going to end. I don't think we're going to see a Terminator kind of world where it's going to come after humans. I don't think it'll do away with
everything that humans do, but a lot of the routine tasks are going to be automated. Actually, let me add this to the set because there is a step between the GPTs and agents. There's what they call automations. Basically, where you string a bunch of decision makers together and it can take something and take a series of actions. But again, it's just one linear series of actions it can take. But one example is if you wanted to
find out information about a dealer, you could make an automation that would go out, or I said a dealer, but a business, you can make an automation that would go out and look at what was on the internet about him, reviews on him, Google Maps stuff on him, what's on, and then if you had a contact there, you could have it scrape his profile on LinkedIn and generate a report. And then you could give that a different name to start with and you'd get a different report. But again, you have to manually give it the name and the report.
you know, what you're looking for. Kind of a bridge between agentic and custom GPTs.
Bernie (19:02)
You're going to make the recipe first. It's not going to make the recipe for you.
Ken (19:05)
Yeah, exactly.
Wayne (19:06)
I'm a big Dilbert fan or was a big Dilbert fan Ken and one of the characters was pointy haired boss and he'd go to a seminar and get jargon and not have a clue. How can you go to a cocktail party and pick up the buzzwords where you know that you know that you know they don't have a clue?
Ken (19:27)
it doesn't take long for somebody that's in the business. Anybody that's doing AI will recognize other AI people. It's almost like it's a foreign language and people can speak a little bit of a foreign language. I'm fluent in Spanish food, for example. I can say taco and enchilada and, know, but that doesn't necessarily mean that I can carry on a conversation that makes sense. I just know how to order food. I actually am a little better than that. So it's that kind of thing.
I would be careful of people that just throw buzzwords at you and aren't willing to explain to you what they're talking about. Especially if you're looking for somebody to help you. Because it's really important that the person that you're talking to is willing to converse with you in a way that you understand. Now I could talk jargon all day long and make it sound impressive, but if you and I can't communicate, how am going to build something that's going to work for you?
Or how am going to help you take the next step? So that's a real critical piece of it is trying to make sure that the person you're going to deal with in this industry is somebody that you can talk to and understand. And they can explain it in simple language if they need to.
Bernie (20:32)
One of the things I've come across is that sometimes you can shove things into the cloud, large language models, know, chat, GPT, et cetera. And your privacy is not part of that equation because, know, not that it's totally hanging out, but you know, if something, if someone queries, know, you're part of that solution. Then I've heard other people talk about everything's inside the box. know, it's all bolted down, et cetera. It's not out in the wild and you can run it.
When you're running an AI agent, are there AI agents that are inside the box, are bolted down, are they running outside? Are the people typically right now that are available in the market, are they running outside? And much like the large language models, everything's outside, right? So you never want to put something totally sensitive in there. You don't want the secret sauce. And maybe not even recounting stats out there. What shape does that take in it?
If you have an agentic AI running for you, is it yours? Like who has ownership of that? Several questions.
Ken (21:30)
And that's a good question. That was one of the things that I approached because I kind of started with the end in mind because I knew I wanted to distribute these into the mass market. And I knew that I needed to have a way to control the data and to make sure that it met GDPR or, you know, whatever country they're in, Privacy Act, those kinds of things, so that it fit in the environment of the company that was using it. So I created basically boxes. We're going to call it, we'll just call them boxes. There's another term for them.
But we'll just call them boxes. And all this box does is take action on data and return data back to the original. It doesn't store any data. So in other words, if it needs to ask the AI a question about something, it goes out and gets the answer back. And that's the end of it. Now, it doesn't mean that that question might not show up if you dig really deep. I have all of the settings on my AI set so that it keeps everything confidential.
I use it in the wild, you could say, but I have the settings on my AI model set so that it's not going to leak my data because I don't let it train on my data. And that's the biggest risk. But I have everything bolted down and to protect somebody from getting hacked with mine, have a term you use called this API, Application Programming Interface. But it's basically just a very precise way to ask questions and get answers.
But other than that, it can't do anything. So it doesn't have any way to even see my code. My code is in a dark box, you could say, and it have any glass walls on it. anything that's attached to that knows is how to ask the question and how to get the answer back. The individual company has to have a key that turns that on. If you don't have the key as part of the API call, then you can't get in. So that provides as good a security as you could get.
It's in fact, this device is something that could be used for HIPAA compliant environments, know, really secure environments, even the government. Now, because they want to keep their data on their network, my device is still deployable, whereas they can provision the data and everything else on their own. They can just give it the, attach it to the resources they want it to have. Whereas, you know, the ones I'm doing here in the US, I provide the real resources everywhere else out of the US.
is going to have to be provisioned by a distributor or a vendor in that company. Because again, I don't want to be responsible for trying to meet GDPR compliance. So I let them tie the stuff on the back end. The customer can decide what access it wants. Because it doesn't have to have access to, for example, to your CRM to take messages. You could program it to do that only. In most cases, the advantage of having it have access to the CRM is
somebody calls, they're automatically gonna get logged into your CRM so that you know who called and have notes about the call, even if they were talking to the AI agent. So it's designed to have that capability, but it doesn't have to have that capability, if that makes sense.
Wayne (24:19)
So a bit of a history lesson, Ken. When did agentic come on the market and where do think it's going next?
Ken (24:28)
Oh, it's, would say probably early this year is when you really started seeing people talk about, uh, agentic AI. Uh, you know, that's about when I encountered it. I was reading some online articles. I kind of monitor the AI world, you know, looking to see what was next. And I read about it I, you know, it was, it was interesting because I have this kind of mind that, um, sees us as something and figures a way to use it. And so I was like, well,
And it started out, was thinking about just copier dealers, because there were a lot of things that they needed that I could do with agentic AI. But then I realized, well, if it did it for copier deals, it could do it for other people. And so I kind of went back and restructured the way I was going to build it so that it is something that I can market internationally and at the same time have security no matter where it is.
Bernie (25:10)
So you assemble all these AIs. say a company is getting ramped up and they're going, okay, I'm going to have these adjunct AI bots. I don't have to have more staff right now. And it's going to help grow business. Is there an up-and-shilling where they're going to have, they're going to run into, hmm, I'm not growing more. Are the flags first off to go, okay, review what's on the go?
to do I get more agentic or do I need more staff? Or do I have to reconfigure how the business is operating? It's not so much that the AI is not supporting me. It's what are the tells that all of a sudden, I've maxed out the capacity, right? Do I wanna grow? Yes. Do I wanna scale? Yes. How do you discern that and what do you do next? Is it like a total review of everything and you kind of ground up from ground zero up?
Ken (25:55)
I'd say the best place to start with is where does stuff get hung up? know, again, you know, the term we could use is a bottleneck. Where's, where's my, where am I getting, where would I need to hire somebody to do something? Cause I can't get it done right now. Might be a good way to think about it. And then the person that's doing that right now, would they be better off? Is this, is this something that could be automated? Is it something we could get an AI agent that would do and that would free their time up?
Or am I better off hiring a person? You know, one of the other things that you could do and you can say maybe in some extent, a gennaker AI is competition for virtual assistants. You know, it was funny at this conference, there were a couple of companies there that had virtual assistants, you know, and they were there basically to let you hire their people, you know, from the Philippines and other places. And, you know, that's, you could say that's probably my competition more than hiring a new body is it's hiring a virtual assistant. I can do it much more efficiently.
But again, there may be things that you need the creativity to do. You again, what you're looking for, what are the pieces that we do the same thing over and over again? You know, you can think about the, you know, how the mail department, you know, the US mail and other mail providers have changed from manually looking at the address and throwing it at a bin to now they have machines that look at it, you know, thousand pieces a minute. You know, and so that's an automation. That's something that can be turned over to a machine and let it do real reliably.
But you know, couldn't ask that machine to paint a picture. Though I would say now with some of the AIs, you can ask them to paint a picture. In fact, I just saw somebody won an agentic AI Miss Universe contest, I think is what it was, you know, created an avatar and won the contest. But, you know, so you think about what are the things that are slowing the business down? What do I need? If I had to hire somebody today, who would I hire? Would be another piece of way to look at it. If I was going to hire somebody, who do I hire?
You know, let's just say you wanted to hire a salesperson. The question that then it becomes is, do I need a person that's going out and walking in front of talking to a customer? Because that requires a human for the most part. Well, we might someday have robots that are smart enough to do it and be a way to handle people. But right now, it really requires a human to do it. Is it somebody I need that just to make phone calls and ask a few questions to see if they might be a prospect?
That's a good fit for agentic AI. You give the AI a list of phone numbers and you tell it the questions you wanted to ask and what you're looking for in the way of information back so that it's not just asking the question and recording the answer, but it may ask follow-up questions. And then that's something it could do. It's getting to the point now. And in fact, it's interesting on the one I'm working on and testing right now.
is she answers the phone says, I'm Susie Q, your AI receptionist. So I don't want to try to hide from them that it's an AI person they're talking to. But in conversation, you might not know that. But I want to make sure that they know it upfront. know, my job is to try to collect some information. Do you mind if I ask you three questions? So you could you could program it to do that. You know, another place where it starts to be automating and you see it a lot in CRMs now,
is mass mailing. Because you can customize a mail out or it's customized to the individual that's going to get it. The same basic message is going to everybody in the mailing list, but you can customize it for the individual and not just putting their name on it. That's a little tiny bit of customization. But it can look back and look at, for example, review their orders over the last year and say, you know, I notice you've been buying quite a bit of this and we just had this special on something I think you might enjoy.
customize the order history and the offer for everybody individually. So that not everybody's getting the same, hey, buy one of these emails.
Wayne (29:28)
And I often think of that Disney movie, one of the animated ones, The Sorcerer's Apprentice, where the man started with a prompt to get the floor washed and the next thing 10,000 AIs are pouring water into the basement and it goes terribly wrong. And we hear a lot about guardrails. If things went reckless,
What would be something that you could see, okay, we've got to this back? What is too far, if anything?
Ken (29:59)
Well, you know, for example, if it made suggestions outside of the questions it was asked, for example, or if it was asking, especially asking questions outside of, you know, so the receptionist, if it's asking somebody how tall they are, what color dress they're wearing today, or what color shirt they're wearing, those would be things that are obviously way outside. You know, and you have to be on the look for that.
Bernie (30:22)
I'm going to suspect someone got in there and just messed with the system going, yeah, this ought to get the customers excited. What are you wearing today? It's not the first prompt. Ask a customer.
Ken (30:32)
Well,
I didn't go quite there. I just said, what color? But, you know, so you see questions like that. That's obviously something that's got out of the the guardrails, you know. And again, that's that would be a programmer's fault. He didn't he wasn't very he got to be very specific. In fact, my just for programming my my software, I have probably a three page prompt now that is the guides, the agent every time we start working together.
And then it has to go back and read like the last four transcripts, you know, of sessions so it knows where we are, what we're doing, what the rules are. And then, you know, if it goes outside of that, you know, something's gone wrong. I go back and fix the guardrail.
Wayne (31:02)
also.
Yeah.
Bernie (31:08)
Did you brush your teeth today? That would be a good guardrail. it'd be. My next question. These are all, in essence, virtual machines running. And if we think back to the reality in our own space right now, if I buy a car, right, it's either power, speed, gas mileage, capacity to do what it has to carry. Are there any of these?
factors around agentic AI and are there most ideal performance based ones? Because if you had say, you know, 10 of these running, these AI is running, but you have, you take a thousand calls an hour, that might influence cycles of time, computing cycles of time and, and, or power consumption. Is that also factor into the, these, ⁓ these guys?
Ken (31:54)
Yeah, did. In fact, it's very much into my into the design position because what I did is I'm using an engine that allowed automatically scale up. if I get to the point that I've got so much traffic going to it that it starts to get to like 50 or 70 percent of its capacity, it'll automatically spool up another server that that way it can, you know, it can split the workload. I use a process called, let's say, let's say an application called Goober. Goober.
Kubernetes engine. And what it is, is it's a self scaling, let's say it's something Google provides so that it's an environment that you can put a program in and it'll just scale up and down as it's needed. Because when you don't need it, you need it to scale down as small as it can get so that you're not waiting for it to spool up. So this case it can spool down literally to just barely being awake and then it can spool up to handle whatever you need in a way of capacity. As far as
Like for phone calls, that would be a limiting factor and it depends on, know, most of mine, I provision with a single phone line. have the ability to add other phone lines and I have ability to attach it to, you know, like a PBX. So that, you know, if it's a phone company and a company that has multiple phone lines, you know, that they can just route a call to it. So they can route the receptionist calls to its extension, so to speak, if that makes sense.
So there's a lot of things like that that'll have an impact, but it's designed to scale as needed because obviously you don't want to say, hey, I don't have time to talk to you today. That's not very good for your receptionist.
Wayne (33:20)
So Ken, you outdo yourself and you've designed enough agents to run your entire day. What do you hope it doesn't run because you like doing it and don't want to give it away?
Ken (33:31)
for me, it'd be the creativity thing. And it's interesting that I give you an example. Right now I have a patent pending on my distribution method. And the way I figured out the distribution and licensing method so that I could sell it in Japan or Iraq, I don't care, you know, because whoever's in that country is the one that's handling it, responsible for all the data. And yet I can turn that one in Iraq off if the person that I sold it to hasn't paid the bill.
So I can turn that individual unit off. that was something that chat, actually, probably Claude and I discussed for quite a while. Because I would say, well, how could we do this? And what about this? And so I kind of asked the questions, and it helped me figure out the answers. so even in the, I mean, the AI agents I'm selling wouldn't do that. if we think about it from the perspective of how I use it,
is I like to create. But there are things that I like to have some help creating because there things that I don't, know, they can research laws, for example, much faster than I can. So I say, hey, if we're trying to make sure that this is copyrightable, what do we have to do? How do we make this unique enough to qualify and have it do a patent search to see if there's anything like it? You know, so that would be the thing that I like is the creativity.
And it might be sometimes it's the one-on-one contact with customers when you're trying to work with a customer to find a solution. A good example of somebody came to me and said, I have this particular problem I'd like to solve and see if I could get that automated or have something do it for me. know, an assistant that could handle that. Then it would be, OK, let's say, what is it you want to do? How do you want to do it? Those are conversations I would not trust to an AI agent.
because I can read facial expressions, I can hear tone of voice. There's just so many nuances that an AI is really not good at. They're getting, some of them are getting a lot better. I saw a sales training agent the other day that reads the nuance of the conversation that the sale rep had like on a phone call or in a practice interview, for example, how the sales agent sounded and how the customer sounded. So that they could go back and review that with a sales trainer and.
to hell, okay, I need to think about how I phrase that or, you my tone of voice or whatever it is. But again, it's like I said, it's the one-on-one stuff and the creativity. You know, like I keep thinking and you know, the last time we talked, I told you I had AD.
Wayne (35:48)
PhD. But
Ken (35:49)
It's interesting that that causes me to, I look at something and say, hey, I could fix that. Now that's where I was when I first saw the AIH and I could use it to fix this. Well, if I could do that, I could do this and I could do this and I could do this. And it just kept going, you know? It's like, okay, I got to do this one piece at a time.
Bernie (36:06)
I've noticed when people will craft prompts and it was in a meeting, this person, he shared it with the group. says, please don't share this prompt outside because it's my prompt, et cetera, so intellectual property. And I get it. And we all concurred. But if you push an AI, genetic AI unit out there, is that something that's...
encrypted? In other words, can you look under the hood and mimic it and then create your own version? Again, this is me not knowing. it secure? Here's my box. It's a black box, but the box does this. then if that's true, how do I make sure it's not doing something surreptitious either? Is there some sort of validation that goes, this will do what it's intended to do, no more, no less.
Ken (36:51)
Well, if anybody was curious about it, can show the inputs and outputs, but it is a black box and you have to know the address of the back box and you have to have that key to get in the back box. And if you don't have both of those pieces of information, well, even if you had both those pieces of information, all you could do is talk to it back and forth, but you still wouldn't know what's in the box and how it's computing it. Because yes, confidentiality is really critical for what I do.
I can't patent the software, it'll be copyrighted as soon as I release it. So that provides me some protection. Again, to the most part, all I'm doing is making sure that nobody can get in and see exactly how I did it. Now, could they reverse engineer it? I wish them well, because I've got about probably 1,000 hours labor in this project so far. And if they can do it, I'll have a head start, is all I can say.
Wayne (37:29)
Sure.
Bernie (37:40)
So how do you make sure that if it's not you, but someone else makes one, you know what, even let's think for a field in foreign lands, we have people that send us something and say, hi, you usually it's a scammer, right? And the send you something, here's this app precedent and you execute it. Next thing you know, you got some dark, dark nefarious stuff happening at ransomware. I mean, that's extreme or spyware, right? How do you, how do you guarantee that the
The bot is pure in its intent. In other words, there's no, that's not what I expected.
Ken (38:14)
Again, that goes back to testing for intent, but in reality, like I said, there's no way into the box. It's sealed. It's like one of those big things on a side of compressor, know, black and it's all sealed up and it's got goop all around it. And you don't even want to try to go in there and see what's in it because it might not be good for you.
Bernie (38:31)
Yeah, no, not so much. I don't want to look in. How do I make sure? How do I validate it's true? In other words, if I if let's go to livestock, I use that as an example, a pedigree dog or pedigree horse, right? You know, I'm sure this has happened on several occasions. Someone says, Hi, this is the Arabian horse. And it turns out to be not so much Arabian. And at least it's a horse. That's that's about the only thing you got on on the mark.
What kind of looks like a horse, but you know, you know, no.
Ken (39:02)
In the sense of if you're buying something, buying one, you you could feel free. would feel free to ask them to, let you run a sample and see what goes in and out of it. And I'd be glad to, can, I can make that access available so they could see what it does. I won't leave it that way because I don't want anybody else in it. And they won't be able to see what the individual components, but again, they could see the output of what that box is sending to like chat GPT, for example. Okay. No. And in my case, that would probably be.
Wayne (39:02)
Okay, so.
Ken (39:29)
That'd be the only place I'm sending data out is I send it to chat GPT for, you sometimes I'll have it. I'll need some information. Most of it again is already created in the box. So doesn't even need to go to chat and be GPT very much. it would do about the only thing actually that's going to need from chat GPT is tokens. Cause every time somebody chats with it, it uses up tokens. So it has to have the access to get the tokens, but that's really all it's sending out even.
Yeah, so everything happens in the box and the, licenses in the box. The only thing that's going into the box is the, audio feed, you know, and that'll go in and out. So you could monitor very easily, but you can just pick up your phone and monitor the conversation.
Bernie (40:03)
Okay.
Is there a governing agency or governing like you have one company called Trustpilot, know, they're companies out there to go, yes, this is good as gold. Is there any standard out there that says?
Ken (40:23)
Not yet. It's way too new. And there's too many approaches to it. know, so I am sure that it'll be one of the things that come up down the road, you know, for right now I can do is I am definitely going to provide insurance on, know, on what I do, you know, and that's to protect me. And, and, know, something happened to a client that cost them, obviously I'm going to be liable for it, but there's just, it's, it's way too new. It's kind of like, um, kind of like the internet was in the very early days.
Bernie (40:50)
The Nigerian prince comes around and says hi.
Ken (40:53)
Yeah. But one of the things, kind of touch on that though. One of the things is, is if you have an agent at AI scanning your emails for you, it would be able to look at the email and know if it was scam or not. Because almost always if it's scam, the return address that it's going to send a reply back to is not who it says. If it says AT &T, it might be sending it to testing.att.
three other things, but they spoof the address and it looks like it's going to AT &T and it's not. In fact, you know, that's just a quick tip for the public is, is you always look to see what it's going to send us as a return address before you hit the reply button. Because make sure it looks like it smells like the right company. Because scammers actually can't get to the right company and then get to have an email usually. But they can sure make it look like it's the right company.
Wayne (41:39)
Can you mention photocopy, sales and repair? And you've mentioned other countries. So two questions. One, what would be the best way for someone to get a hold of you if they want to have a conversation? And what is, how long is the process to find out whether there's enough interest to go forward?
Ken (41:56)
Well, to get hold of me, Ken, at 22ndcenturymanagement.com, 2 2 C T Y G T dot com. That's the best way to get a hold of me because I check my inbox pretty regularly. In fact, it's sitting open on my desk. It's just a short conversation to see what a person's interested in, you know, and see what, you know, in fact, right now I'm actually looking for people that want to be, ⁓ let's call them beta testers.
Because I'm making really special deals for people that want to test them as they come out. Because everything's got to be tested in the real world. I can make it work perfectly in my world because I control my world. But I got to get it out in the wild a little bit so that we can see what breaks. I tell them, here, try to break it. And then let me know. If broke it, I'll fix it.
Bernie (42:37)
It's always a challenge. It's not a challenge to break stuff is what I'm getting at. Kenneth, I want to say thank you so much for being our guest today and ⁓ bringing a new insight to agenting AI. To my host, Wayne Pratt, being my co-host, our co-host, the show's co-host, to you the Knack 4 Business listeners, because you know what? This is really important stuff to understand because it's the next layer of AI activity that's going out there.
And if what Ken has been talking about really appeals to you, you want to be a try it out, test it out, reach out to Ken. If you want to understand more about agentic AI,
You want to trial out some examples of it functioning inside your office space and having a good feedback. There you go. Because this will help reduce your bottlenecks in your space.