The AI Agency Building Custom Agents for You

Sesame.com has basically split the atom with uh the sophistication of their voice agent technology to the point where you cannot distinguish between speaking to a human and speaking to one of their voice agents. We have taken their open-source code, we have forked it and built it into our product. Hi, my name is Dmitri Bonichi and I'm a content creator, agency owner, and AI enthusiast. You're listening to the AI Agents podcast brought to you by Jot Form and featuring our very own CEO and founder, Idkin Tank. This is the show where artificial intelligence meets innovation, productivity, and the tools shaping the future of work. Enjoy the show. Hello everyone and welcome back to another episode of the AI Agents Podcast. In this episode, we have Andrew Westpecker from Powered by. Powered by is a new and innovative AI agent company that we're interviewing today. As you

all know, we like to bring on the new and upcoming AI agent companies out there, and we're really excited to chat with you. So, Andrew, how you doing? Hey, Dimmitri, thanks for having me. Uh, excited to be on the podcast. Yeah. So, I guess just to get things started, um, obviously we could give the elevator pitch for powered by. However, I think it would be better to kind of hear, you know, your background and and and how you got to where you're at. It's it's crazy to me that companies uh, you know, I I just love hearing the stories of companies. So, I'm just very excited to hear about um, your background and where you came from to come to the point of running Powered by. Sure. So, uh, unfortunately I'm the, uh, the OG on this call. I'm, uh, in my 40s, my my

mid4s, unfortunately. Uh, and so I've been in the SAS and enterprise software space for 25 years. I've seen the great, the good, and the not so good in terms of software quality. uh over the last sort of three or five years I've really seen the rise of uh adoption of machine learning technology and more recently uh the LLMs and uh over the last sort of two years I've been specifically investing in series A and seedstage companies. I have my own uh venture fund or you could call it an angel investment fund. Sure. And while I was doing uh due diligence on a couple of investments, I got really deep into AI agents. And I dug myself a rabbit hole into AI agents that I've yet to extricate myself from because of how interesting the technology is. So outside of the the business interest, I took

a personal interest in it and began experimenting with all of the various different tools. uh at that time which was cursor uh wind surf clin and some of the other like GTM sales and GTM tools like AI SDR etc etc and then really began ideiating on sort of building a business uh around AI agents specifically uh AI agents that operate in voice phone SMS text and email uh and started powered by agency uh about 6 months ago, incorporated it. We launched officially uh about 3 weeks ago. Uh we're now working with our first set of customers and so we're off to the races. I think at a at a high level what what we're focused on is the following, which is all of the larger enterprises in the United States, the Fortune 100, the Fortune 500, they already have a team of AI engineers, they

already have uh an AI, you know, R&D budget and a lab that's already tinkering with the latest in in AI agents. Equally they already have deep relationships with the sort of major players in this space consisting of OpenAI, Anthropic, uh Microsoft, Google and the systems integrators like BCG, Deote and Accenture. The challenge is when you look at the rest of the economy. So take take away the Fortune 100, the large banks, uh airline companies, insurance firms. When you look at the sort of the ecosystem of small to medium-siz businesses, uh they often get left behind in terms of the attention uh from the large AI vendors. And those large AI vendors are really interested in large contract values with the the largest banks, the largest retail companies. And so, uh, Howard Bay was formed on the mission to democratize access to the most sophisticated AI

agent technology for small to medium-siz businesses. And so, I think I I I think I want to really explain why this mission is important to us. And that is in terms of the overall impact that small to medium-siz businesses have on the US economy is often really not well understood, but it's astounding. There are 33.2 million SMB organizations in the United States. They generate 43.2% of the US GDP. They employ 46% of all private sector workers. Wow. And the last point, the last point is astounding. Uh they represent 32.6% of all US export value. So the foundation of the US economy, believe it or not, is built on small to medium-sized businesses. And our mission is to democratize access to the latest uh most sophisticated AI agent technology for those organizations um that are being passed by because the largest AI vendors and systems integrators

are are are going after the big fish. Wow. Yeah. I I think a lot of people are not aware of the I think maybe a lot of people are aware of those general stats of oh small businesses make up America like in the American economy but they're not aware of the numbers to back it up. And on top of that, I 100% think that you have a good ethos for your company and a good mission because so I'm a small business owner, but the problem that I notice when I talk to other people is I mean I run a SAS content agency. Like obviously I'm going to be more technically inclined than the majority of uh SMB owners I would say nowadays. And I'm like the only one that I know that's obviously there's the SAS companies I work with that are similarly inclined, but

nobody else in small business is really able to go through and make their own pseudo AI agent through automation platforms and and like knowledge bases and stuff like that. No one knows what that even means. Even even Demetri Yeah, I completely agree. in in our sort of ideation phase of building this business. We spoke to a lot of uh businesses and you know I live in the San Francisco Bay area and so I live in this microcosm where we all speak the same language and everyone's talking about you know this AI model, this new AI agent, this new uh rag architecture etc etc. But when you take an airplane and fly an hour to two hours east of Silicon Valley, that sort of competency and the language that we speak here around the sophistication of these technologies really doesn't exist. And the vast majority of

people in the United States still sort of understand and conceptualize AI as talking to chat GBT. And really over the last year uh we've just seen a tremendous increase in the breadth of AI technology into uh it effectively becoming digital humans, digital workers for variety of use cases that we offer. Um and so we're trying to communicate and evangelize all of the new opportunities that small business owners uh can unlock with AI agent technology. Yeah, I I I I feel that. I see that. I'm very interested in one of the points that you brought up to to start, which I also don't think a lot of people realize, which is I mean many people tend to talk about a gap that is emerging between the uh classes so to speak. Um, and I I think when we're moving into this next phase of the economy

as it's going to be partially run by AI agents, you know, uh, from a workforce standpoint, I don't think people are really aware of what you said at the beginning, which is a lot of these different companies are working with CHACPT or Claude or they have their own own internal teams building out these AI agent infrastructures. that you just simply won't be able to compete with. You know, say you have a huge marketing agency, one of the top uh you know, of those of those companies might be building something out like that. Obviously, the companies themselves that are already big, you know, Google, Meta, uh etc. They're building out their own models. And every huge company is essentially day by day getting farther and farther away from where the SMB is at from from a techn technological standpoint. I think maybe something to point out

here and and then ask a follow-up question from you is people don't even know what like uh technology debt is or like the concept of like technical debt and it seems like your entire goal is to bridge the gap and hopefully make up some of that technical debt that they're unaware of. How do you feel powered by makes that uh decrease and and what makes your platform so so accessible? So that's the uniqueness of our approach to the market. There are a variety of very well-funded uh VCbacked AI agent vendor tools in the market. You'll read one new funding announcement every week on TechCrunch for the next new awesome AI agent vendor. uh the these vendors that I've mentioned are all selling SAS and or developer tools to allow end customers to build AI agents themselves for either customer use or internal use. We don't

do any of that, right? We approach C customers on a completely blue ocean uh sort of bespoke basis. Now our approach is that we've got uh a series of fundamental AI agent foundations that we have built through a combination of open source technology, other vendor tools uh and the underlying LLMs. We sort of package that together and when we go in front of a customer to consult with them on what their potential use cases would be for AI agents, we don't prescribe anything. We consult with them as a you know a management consultant vendor would to you know a large organization. We sort of understand where they want to drive cost efficiencies where they have breakdowns in their current customer engagement process uh sort of where there are inefficiencies or cost overruns in in parts of their business. And then we begin to piece together

where is the lowhanging fruit for an AI agent to be introduced to this organization. whether it's, you know, an AI receptionist, whether it's uh a voice chat uh embedded in their website, whether it's an email uh agent or an SMS text agent or even a a digital avatar, we then work back into uh what we would then build in a bespoke manner for the customer based on their business requirements and some of the challenges that they're dealing with. So we don't expose the developer framework uh nor do we sell the developer framework to the UN to the end customer. We act as an agency business. That's why our name is powered by agency which is we have fundamental pieces of of intellectual property but every new customer engagement uh we treat as a as a unique snowflake. Interesting. Okay. So that that is a unique

situation because I do feel like when it comes to some products obviously you can just log in buy the product and um to some extent obviously there's there's plug-and-play that makes sense here but yeah that's an interesting usually it was funny when you use the word bespoke uh I feel as if that is usually a a you know like a meaningless term you know when people I'm forgetting the the term for meaningless term here but like a throwaway term But that that actually I think in the context of what you're doing, you're truly utilizing the term accurately and and trying to provide that because I I think a lot of people will say, you know, we're you know, we provide unique solutions for XYZ category or thing, but it seems like you're actually going about I mean, for example, you've I'm guessing seen this show

in Silicon Valley, right? Uh-huh. 100%. Oh, and my favorite bit that they do in that show is the consistent uh show, the consistent reference to we're changing the world by integrating whatever. Yeah. Yeah. Therefore, making the world a better place. No. And and that's usually just throwaway term, but so you literally go through and anytime someone wants to become a user of your product, you will try to tailor uh potential AI agent solutions by uh taking a look at their business and seeing what they need. Is that Yeah, I exactly. So there are there are two elements to our model which is one is we have pre-built foundational AI agents that operate in uh voice uh AI receptionist um email text and digital avatars. And then we have a team of four deployed engineers and uh engineers that do the eventual design, build and implementation

uh for VN customer. Where did we get this idea? Palunteer. Palunteer. So, we've we've uh we've borrowed and I hope I don't get sued. We've borrowed their go to market model. Uh Palunteer obviously focuses on extremely large contract values with, you know, the defense agencies and very large enterprises. We've sort of shrunk that down and taken that go to market model uh and applied it to much smaller businesses with much much much smaller contract values. And I think that's also an an important point. Um, we make these AI agents uh affordable for an SMB business where you know you can start um you know getting your feet wet with it. you know, an AI receptionist or a voice chat in your website, um, you know, for for, you know, $299 a month or $599 a month based on the number of interactions that you need

on a monthly basis. So, it's a very different ballgame than a Palunteer type model. Um but we borrowed their sort of we take fundamental IP and blend it with four deployed engineers um and and sort of developer deployment engineers to to render the solution for the customer. And I guess just for the uneducated on the subject as as far far as they are um something I guess I could add to the the reference frame here is I actually used to work in are you familiar with like the no code consulting space? Mhm. It's very interesting to me because what it seems like you're doing is you're almost taking a bit of that component and giving them not only the solution to maybe have an infrastructure for people to do the work, but then you're actually giving them the capability for AI to do the work

because that that's the interesting thing that's occurring I feel like in this Well, that's that's you you kind of spoiled my surprise which is no uh our our first iteration of the business is an agency business, which is just like a web design firm from 25 years ago would go build a custom e-commerce site for a brick-andmortar retailer. We're doing the same thing right now is where we're going to real estate agencies, auto dealerships, uh health clinics, boutique hotel chains, and and we're building uh custom AI agent solutions based on their uh ultimate business requirements and and challenges. Ultimately, what we want to become, and sort of this is the sort of the spoiler, is is we want to become the Wix or Web Flow equivalent for AI agents. Um, I'm in love with this product called lovable.dev. And I would encourage everyone I would

encourage every one of your users to um uh to check it out. It's it's one of those experiences with a product that I've had where it I don't think I've I've experienced a more magical product in the last couple of years than I have with Lovable. And and what it allows uh an idiot like me to do is it it allows me to build a full stack application, a front end, the backend, the database, the data analytics, 100% through AI prompts. It's incredible. We want to take that concept uh and you know after we've built enough book of business in our agency business create a product that allows a lovable or sort of web flow like experience for the non-technical user to build AI agents without any code. Um currently today there are as mentioned a variety of tools for building AI agents but you

still have to have some semblance of developer or AI engineering experience to make it work. Our vision is to ultimately build it for idiots like me to to build sophisticated AI agent technology in natural language English prompt. Interesting. Yeah. I I mean I've heard a lot about level recently by the way. It is something I have tried to fiddle around with a little bit but just haven't gotten around to it. Uh are you in in this sense I I am curious by the way because it is it is always on everybody's mind. You're essentially offering something to small business owners which I actually do think is very beneficial. Are you uh at all in this position? How how do you envision how companies will change their hiring plans moving forward? cuz that's obviously something people are thinking about with this. Um, you know, and you're

you're helping small businesses. I think it's great you're helping small businesses. I'm just curious, how do you feel like this is going to adjust hiring plans for those small businesses? Like what are you able to cuz you mentioned the price point. What do you feel like you're able to provide value-wise in response for that? You know, what they're what they're giving you and how what they're going to get out of it from an even theoretical worker standpoint. Demetri, you've raised an an ethical question and an ethical dilemma around AI agents, which is uh do they replace humans and do they um you know cause unemployment uh much like you know our economy uh 20 30 years ago uh you know outsourced all of our manufacturing base to other countries and led to you know a series of of cities across Ross the US to to

effectively become de-industrialized. Will that same thing happen with uh the meteoric rise of AI agents? I can't answer that question. However, I feel deeply in you know our our company's mission that we don't want to replace humans with AI agents. What we want to do is reallocate your human staff members to higher value activities where an AI agent can perform the menial repetitive uh and lower level lower value tasks where you can then upskill or reallocate your human staff members to focus on more higher value sort of customer uh specific types of activities. uh that is the design as part of our mission to democratize access to AI agents. It is not uh very clearly not to replace humans with AI agents. No. Yeah. And I think that's a that's good to talk through that. Uh the key note that I'll probably point out as

someone who I believe relatively understands the space a little bit is exactly what you said where it's more low-level tasks that don't necessarily add value in the same way that humans can add value. Like I was actually talking with IdaK and the co-host about this in a recent episode, but Steve Jobs said in 1980 that the computer will be a bicycle for the mind. Mhm. And he was essentially making a parallel as to how if you look at the efficiency of running, we're like ranked third or whatever when it comes to uh whatever sect of animals he pointed out. However, when you give us a wheel, you know, for a bicycle, we become, you know, the most efficient. So, we're essentially creators and we're tinkerers. And I don't think people get this because I was I've been in the automation space for like four years

now. And this opened my eyes to uh this, but it's a more expanded version of this. The majority of what we do in work that takes up our day-to-day tasks that isn't necessarily value additive. is just a huge list of if then logic that we have to parse that we have to parse out through uh typing means and through tech and through computer interface and what essentially from my understanding is it's just taking a step further from the API if then logic that exists through you know the automation world I come from and now has knowledge bases to go along with it and then has the ability to determine based off the inputs of text and outputs of various variables, actions for you that go above and beyond the just end points that exist. Well, I I'll talk about there were sort of, you know,

if you're if you're close in the automation space, um, UiPath, automation anywhere, monday.com have built, you know, billions of dollars of market value in robotic process, robotic process automation, right? And so um huge business, huge market category. AI agents are miles beyond basic RPA. And the fundamental reason why is the textbook def definition of an AI agent is that it's a piece of software that thinks, behaves, and learns context over time. RPA tools really are effectively software scripts with a glorified if then tree decision matrix. Exactly. Exactly. We can train our agents uh to do things that are so sophisticated uh because they understand the context of prior conversations and understand the context of of communication and conversation that it's had uh an an in an anonymized manner with other individuals. So, uh, what that means is the more time our AI agents perform their

work, every second they are getting better at their job. You can't do that with RPA because an RPA doesn't have a brain. The AI agent does. Yeah. I think that's the that's the key thing that you're you're nailing that I really like. It's it's very difficult from my perspective for the average person to to conceptualize this and my question just because I I think you're from Silicon Valley. I'm I'm from the world of this space. Obviously, I host a podcast about it with you know it. I am curious to you what is the general understanding of what this is like in the SMB world? Like you you mentioned briefly earlier and I wanted to come back to this point. You mentioned previously earlier you had talked with a lot of SMB owners about this subject because obviously you're running a company on on this now.

What percentage of small business owners do you actually think are acutely aware of what an AI agent can do? Less than 5%. Less than% right. That's crazy. And so we tinker a lot when we work with customers and when we do like demos in initial sales meetings with customers. Um we like to demonstrate our our voice agent capabilities whether it's you know a phone number that you call as an AI receptionist or a uh a voice chat that's embedded inside of their website. And and we have those on our website that you're free to demo and and take a look at. you can talk to an AI agent uh on our website any time. Um but in certain circumstances, we build sort of custom demos for a customer to sort of wow them. And over the last few months, there's uh a company uh that

your your listeners and viewers should take a look at. It's called sesame.com. sesame.com has basically split the atom. uh and I don't mean this with hyperbole. They've basically split the atom with uh the sophistication of their voice agent technology to the point where you cannot distinguish between speaking to a human and speaking to one of their voice agents. We have taken their open-source code, we have forked it and built it into our product and we build these voice agents for an AI receptionist and we test the customer and and ask the customer, "Go call this number." They call the number or have a natural language conversation with that voice agent. We don't tell them that that was an AI and they come back to us um shocked that uh the the the state-of-the-art has advanced to the point where you know you cannot tell the

difference between speaking to a human and speaking to uh an AI agent. And and I'll I'll put I'll put one finer point about this which is kind of a funny point. Do you remember that Will Smith AI video of him eating spaghetti like you know that was 700 that was 700 days ago. Okay. 700 days ago. Okay. Now I it if you look what Midjourney and Sora and what the the other providers are doing in the space with video production where you know Hollywood you can create photorealistic um you know Hollywood grade blockbuster grade uh uh video production with promptbased LLM you know that's this is that video that you're showing there was only 700 days ago. Yeah. Um what I why I'm going to play this forward is with this product Sesame, their open source project, um that was just released uh a couple

of weeks ago. And so, uh, we're at the cusp of where interacting with AI agents over voice, over email, and text, um, is going to cross the what we call the uncanny valley where you're not going to be able to, uh, distinguish between an AI agent and, uh, a human. Uh it's it's a really exciting time to be in this business because of what's happening so quickly uh around innovation and AI agents. Yeah, I really appreciate yeah you bringing up that other product and I'll have to give it a look too. I I I really am excited I think for the general use cases of this. As somebody who is a bit of a public persona, it does freak me out a little bit. um with the fact my voice is basically public domain. Um because I don't know if you've used 11 Labs, but

that's obviously been like kind of the the quote lead in the space for a little while. Um they released that new version. I think it was like 6 months ago. Maybe I'm off on the timing on that, but that that update was incredible. Uh and the these things are continuing to expand and we talk about it with IDKIN all the time. It's it's incredible. You you take even what you mentioned obviously that was hilariously bad. the the Will Smith uh spaghetti thing. people forget I mean 2023 if I'm not wrong as well was the year and it was like November when like Chad GBT first or was it 22 I forget where Chad GBT essentially first came onto the public domain with 3.5 was 23 I believe the fall of 23 right and that is essentially now we've gone through oh my god remember when

chatbt4 came out Yeah, that was like a huge leap, right? And I remember this is what people who are more nerdy about it will notice if they remember the details. Llama just dropped a context window of 10 million tokens for their most recent model. Do you remember what the context window was of chat GBT 3.5 before they released anything past that like that in 49 496 or or wait no it was 20 2048 and then it was 40 4096 a couple weeks later it's went from like a 2000 token window to 10 million in and and that's and and Demetri you you are sort of legitimizing why uh I dug that rabbit hole uh in AI agents because you know I've I've been in hard software tech for 25 years. I've been through three epochs. The dotcom epoch, the SAS epoch, and more recently the

public cloud epoch. And this new AI AI agent epoch is way more uh interesting, innovative, fastm moving. Um, and it's I don't think I've ever in my, you know, almost three decades of professional experience ever been more excited about a piece of technology than I have with AI agents. So, we're starting here at where we're at with your company where you're saying, "Okay, we're going to be able to do these these tasks for you um through through the agents." If you had to you gave a couple use cases. If you had to name the most intriguing one to you, what would it be? Right? Like what what of the use cases is the most intriguing to you and you think most small businesses can benefit from? I'm going to answer it in two ways. I'm I'm going to answer the first one with a subsidiary

business that we built called Virtual SC. And you you can go to getverirtual.sc. What is it? Uh it's designed for SAS organizations that um struggle with the utilization and resource allocation of their human pre-sales systems engineers in sales calls over Zoom or Google Meets or Microsoft Teams. Those are a limited pool of very valuable resources that help uh salespeople uh qualify and close deals as the technical uh subject matter expert for a given product or service. The challenge is there are often times more customer meetings than available hours for those systems engineers. What we've done is we've built a virtual SC. What does that mean? we uh can ingest all of your companies, your SAS companies product docs, uh all of the support cases, all of the implementation documentation and create a singular uh voice email or text agent uh that can act as the

most knowledgeable best-in-class pre-sale systems engineer uh when a human systems engineer isn't available. And the example is, let's say I'm a sales guy or gal and I'm getting a uh a new demo meeting with a really marquee customer account and that's scheduled at 1:30 p.m. tomorrow and my SC isn't available. I have a couple of options. I can cancel the meeting with the customer, which I don't want to do. I can find another SC, which will be hard to do. Uh or I can reschedu. The fourth option is to use virtual SC which joins the Zoom meeting just like Gong or Otter or any other transcription tool. um it's in listenonly mode and when the sales individual the sales rep uh wants to discuss something very technical whether it's you know a configuration issue whether it's how do you implement this how do I migrate

from XY to Z the sales rep can say well James can you tell us about how we can migrate our legacy SQL into your cloud data warehouse he then offers a very you know using those voices that I that I mentioned that are near or at humanlike a very comprehensive and accurate response uh to that technical question even though there isn't a human SC in the room. Um that's something that we've seen a lot of of interest from. Uh the uh another module is what we call outbound AI which we use for any organization that does outbound sales and marketing to uh customers in their lead database. they can use outbound AI to launch outbound sales calls, outbound emails, outbound text campaigns. Uh, you know, so so I'll give you an example. Let's say there's an auto dealership in Florida that has 35 locations and

they have a spring sales event in April where they're offering a 15% discount off of the MSRP of all of their 2025 models. And let's say they have 2500 uh individual contacts in HubSpot or uh Salesforce. Notably, these are optin leads uh and they are have accepted uh the ability for them to be called by the uh the auto dealership. You can use outbound AI uh to call all 2500 of those leads, email all 2500 of those leads or text all 2500 of those leads instantaneously right now. Um and have a you know a dialogue. It's not a blast. It's an actual conversation which is hey this is Frank at the dealership um in Tampa. wanted to let you know that we're having a spring sales event where all of our automobiles are at a a 15% discount off of MSRP. We'd love you to

get get you back into the dealership. Um, can we schedule you a time this Saturday or perhaps next week to come in and take a test drive of one of the vehicles. That sentence and the cadence that I've just mentioned is exactly how the voice call behaves because of uh Yeah. And and now I I just want to I want to I want to caveat with a with a very uh bolded asterisk that you cannot use this to cold call email or cold text customers. It violates uh a congressional act in 1991 called the TCPA act. Y um so it is only allowed for calling, emailing, and texting individuals that have opted in to receive sales and marketing information from your company. That's so we've got the the virtual SC that can stand in for a human sales engineer when they're not available and then

sort of the outbound sales email uh calling for organizations that want to engage customers in mass. Interesting. Okay. And you mentioned earlier about learning uh and what I mean by learning is the fact that these are different than other tools because AI agents are essentially able to learn and improve. How does that necessarily work um from from what the customer experience is, right? Are they giving feedback to the agent to to give it more uh is the agent autonomously learning or how does that entire process work for you guys? So we can do it we can shard it in in in different ways. I'll give you an example of our email agent. So let's say that auto dealership has started a email thread with a customer with their unique email ID which is you know frank.tompsonmail.com. Okay. The subject of that email is visit the

dealership. And so over a period of a couple days, the AI agent who also has its own unique email address and name is having a back and forth thread about uh the automobiles and inventory that qualify for the spring sales event. Now, two weeks later, that customer, Frank, emails the agent with a different thread, with a different subject that says, "When can I schedule time to visit the dealership for a test drive?" The agent has context in memory of the prior conversation it has already had with Frank, right? Wow. And that's well beyond anything that you can do with scripting or with RPA. And the reason for it is that it has the context that it is like a human. It is established a relationship and a set of context of how those two individuals i.e. the agent and the customer have interacted over the

last few weeks. That same concept applies to our voice AI AI agents and our text agents where it learns and and can glean context over prior uh uh you know historical conversations. That's one of the key differentiators for us is that the ability to extend context and AI memory uh across the various different modes that we operate in. Wow. Um, so when when I've worked with AI agents, I I definitely feel that there's this benefit of it of it learning it learning it learning. How do you in the context of these different scenarios then end up sussing out what actually could be uh maybe removed? Say for example, it's not acting in a in a certain way that you want, right? because obviously it getting fed information is one thing, but I'm sure there are instances where it could quote learn maybe a little incorrectly

or some of the knowledge base becomes outdated. How does it recycle out any outdated information? So, uh that's a part of our managed service. So, we're not uh again that's why we're we're an agency business. We're not selling a you know a subscription and we walk away. So let's say we implement uh a voice agent solution for a boutique hotel chain to manage their reservations and and loyalty program. Uh they give us the the sort of the workflow. They give us the knowledge base. They can hook us into their uh their reservation system where we can book reservations uh through through an API or a web hook. Um but let's say they bring in a new system. Let's say they bring in a new uh ERP or or reservation systems and need to change how the AI agent interacts with uh the booking system. That's

just a simple request because the customer is already paying us uh as part of their contract a managed service to optimize the AI agent. But I think you're referring to another point which is how do you ensure that the AI agent is behaving correctly and not overstepping its boundaries of its prompt. Yes, that iteratively also happens with our managed service which is we we start with a prompt uh and and let's say you know for whatever reason um the AI agent does not uh achieve its objective. Let's say for example, it mispronounces someone's name repeatedly, right? Uh and and and doesn't get the feedback. Uh the customer uh is going to say, "Hey, you're you're mispron mispronouncing, you know, my name." it will learn over time through multiple sort of instances of incorrectly uh making that misprononunciation to learn uh the correct way to pronounce

that individual's name. But it's important to guard rail inside of the knowledge base that we give an AI agent and the prompt that it is only able to do these certain activities with these certain objectives with this certain scope of knowledge. So, if you ask one of our voice agents, you know, that's that is representing a hotel or a medical clinic or uh, you know, an auto dealership. Okay. So, now tell me, how far is the sun from the earth? The AI agent doesn't even know the answer to that question. And that's like, you know, well, why wouldn't it? It's it's acting on the most comprehensive set of data in the world which is either you know chatbt you know 45 mini or sonnet 37 it should know precisely and immediately the distance between uh the earth and the sun when you prompt it and

you guard rail it you can it it actually doesn't know that and so you train it to respond to when it gets questions that are outside of the realm realm of its prompt and knowledge base is simply going to say, "I'm sorry. I I actually don't know that the answer to that question. I really am only programmed to to talk about, you know, the hotel property and reservations that we have or uh the Mercedes uh you know, automobiles that we currently have in on offer in our spring sales event." So the the that's sort of one of the the the prime values of working with a partner like us is that we help you implement the guard rail and implement the appropriate prompt and knowledge base so the uh the AI agent is learning correctly but not overstepping any boundaries. You know I find that

so interesting. Uh, I mean, I'm aware of some of these things, but a lot of people get worried about some of the idiosyncrasies of this, like, okay, wait a second. If I tell the AI, I want to do XYZ thing. What if it runs off and does something else? There's actually guardrails you can implement now. Like, I I think what ended up happening at the beginning was, you know, we talk about 3.5 or three even. And I almost misspoke I think earlier. I think it we got three delivered to the public uh in the majority and then 3.5 turbo came out so for regarding chat GPT but so but so many people remembered ridiculous hallucinations and I do not think people realize that was one of the first things that they that that you know we we really worked on in the space was how

do we prevent the nonsense because people are going to want to use this for business and businesses cannot have that. So the average person well I think I I I think the difference is is that the general purpose transformer is acting on the entire world's history of knowledge and communication whereas an AI agent is looking at 1 1 millionth of the world's knowledge and history of communication because it's only operating on what it's given its prompt and its knowledge base. So it's a lot easier to guard rail a voice agent, an email agent or a text agent than it is, you know, a global transformer um that is operating on on the world's knowledge. Our knowledge, you know, powered by AI agents are only acting, behaving, and responding to users based on the guard rail, the knowledge, and the prompt it's given. So, it is

for all intents of purposes, it's it's dumber than a general purpose transformer. Um, but it's better at doing its job and minimizes the amount of hallucinations. Yeah, I think that's something that's interesting in this conversation is most people are presupposing that AI in general. It's a funny pre uh presupposition that it's good at everything. Um, mainly because I think it was pitched at as that sort of way of like, oh, it's AI can do it all when it comes to text responses, whatnot. But to my understanding, AI agents are actually really niche, right? They they're they're very good at specific tasks or is this or or roles. Is this fair enough to say? That's what we recommend. So for we recommend use case specific whether it's an AI receptionist whether it's handling reservations whether it's you know booking appointments and we can what the cool

thing is we can stack our agents so you could have a voice call with an AI agent that can then hand over uh to an email agent that can hand over to a text agent that all have the same name, right? They're all the same person that have the same context of the conversation, but they're operating in different modes. But we fence that into a a boundary of what it's supposed to do, which is your objective is to educate the customer on the hotel properties that are available uh for this hotel chain. and your objective is to book uh the customer's desired room type on their given reservation dates. That's starting small. We don't um we don't recommend uh you know having an agent that does lots of different things uh because that could lead to a wider boundary in the higher propensity higher propensity

for hallucinations. Interesting. Okay. So, I think that covers off on a lot of the concerns that people have. Uh, for me, I I think the best thing about AI agents is primarily the fact that we don't know where it's going to be in a couple years. And I'm really impressed with what you guys are starting um and what you're able to do in in the short time that you've been around. And it seems like you you really have the right ethos for it and everything. So I'm I'm very I'm very glad that we oh you know we we discussed this. I guess last thing would be to ask the question what if you had to add to the functionality that was in your AI agents would you in the future like to have? I know it's a it's a wide question. is a million things

you want, but if you if you could envision if you could have the perfect AI agent right now, what would be what makes it perfect? Uh, this is pretty funny and and I don't know if this is answering your question in the right manner, but we're recording this in April and I haven't filed my taxes yet. And so I would like an AI agent to take all of my documents, uh, get the right deductions, do everything for me, uh, so I can file on time next week. Uh, that's a, uh, that's a use case that we haven't cracked yet. Um, we're not necessarily focusing on that. Um, but that's a funny use case of of something it for me in my personal life. Some company is right now uh they're probably listening to this is right now working on this and next year's tax season.

Um, I'm not going to have to hire H&R Block or, you know, one of these other uh, you know, Turboax, etc. I go to to, you know, uh, agents.com and I just give them all my documents. It does everything for me. Files, pays, uh, gets the right deductions, etc., etc. They get my business. Interesting. Okay. All right. Well, I mean, I, as somebody who has a has a tax guy, but also wishes I could have just had it done via AI, uh, I agree with you. Anybody uh, who is in the small business space or or is thinking about running a small business, please don't complain about your W2 taxes. at in at Thanks at at uh at Easter or whatever again, please. It is we we would appreciate if you understood it is not complicated to to the Turboax click. Uh but anyways, that's

besides the point. I really appreciate you being on here. Please plug away whatever you'd like. Obviously, the product itself and anything else that you may be working on, please plug away as we close this uh close this episode out and you give us uh whatever info that you'd like to share with the crew. Yeah. So um thanks Demetri for for having me. I think this message is for any small business that small to mediumsiz business that is you know wanting to learn more about the power of AI agents and how the way that it it transforms the way businesses work, communicate uh and engage with customers. Uh there are a ton of opportunities to unlock with AI agents across the modes that we discuss which are uh voice, phone, text, email, digital avatars, etc. Um, we're a consultancybased approach. So, if you want to spend

30 minutes with us, learning more about AI agents, uh, with no obligation to hear a sales pitch, we love talking about them. Uh, just go to our website, www.poweredby.ag poweredby.ag uh, and, uh, click the button to book a consultancy. Um, and we'd love to speak with you. Awesome. Well, thank you so much, Andrew, for being on this episode of the AI agents podcast. We appreciate it and we'll see you all in the next one. Bye. Thanks, DJ. Bye.