188 | AI Agents in Action: How to Build a Business-Ready Agent W/out Writing Code with Pooja Jain

0:00Hello, and welcome to another live episode of the Leveraging AI Podcast. The podcast that shares practical, ethical ways to leverage AI, to improve efficiency, grow your business, and advance your career. This is Isar, Maitis, your host. And we have maybe the top exciting topic of AI in 2025, which is agents, which is going to talk about today. And agents are already here. And I know a lot of people are hearing about this as something that is coming, and it's the next stage, and it's the involvement of LLMs, but agents are already here. And multiple people and companies are already generating significant business value by leveraging agents across multiple aspects of the business.

0:43But for most people and most companies, there's still this elusive concept that people don't understand and definitely do not know how to implement. Now, this gap between the people and the companies who have agents to those who don't is spreading every single day and it's widening. And you want you yourself as a person with capabilities, as well as your companies, to be on the right side of that gap, meaning you want to be on that fast bullet train that accelerating and providing more and more capability to your business versus being standing on the train station and watching that train getting further and further away from you.

1:25especially if you know that your competitors might be riding that train, meaning they'll be able to do more and more than you can for less money and then be a lot more competitive than you can. Now, developing agents sounds like complex and it sounds very technical and it sounds like only big companies like Microsoft with a lot of people who write code can actually create them. But the reality is there are multiple tools out there today that allow you to write really powerful agents with either no code or low code. Today, we're going to focus on no code at all. So one of the most powerful tools out there that are able to do that is called Relevance AI.

2:03And our guest today, Pooja Jain, is a relevance AI expert, and she has been developing agents for multiple companies on relevance for a while now across different aspects and different businesses and so on. Now, in addition to the fact that she knows relevance really, really well, she spent years in developing and implementing AI solutions at Procter & Gamble. So in addition to her recent experience in just building them on relevance, she has enterprise level experience in understanding what is required to actually develop agents that actually work in a business environment. And she knows how to do it in a step-by-step in a tool that anybody can use, which makes her literally the perfect guest to talk about this topic with us.

2:46So this is exactly what we're going to do today. We're going to show you an entire process beginning to end, how to develop an AI agent, what to think about, and how to use relevance in order to do that. But having that knowledge and seeing this will allow you to develop other agents and on other tools because the concepts are exactly the same. Now, as I mentioned, since agents are maybe the most transformational technology we ever created, at least that I know of so far, then this is a really important topic. And hence, I'm really, really excited to welcome Pooja to the show. Pooja, welcome to Leveraging AI.

3:25In the next few years, AI technology will change our world dramatically. Whether you are a business executive trying to catapult your business forward, or just somebody who refuses to be left behind and want to advance your career, this is the show for you. I'm your host, Isar Maitis, a serial entrepreneur and an AI enthusiast. You'll hear invaluable practical tips from innovative business leaders, AI practitioners, and some of the brightest AI minds in our world today on how you can leverage AI in ethical ways to advance your career and grow your business.

4:06Thank you so much, Isar. How are you? I'm doing awesome. I'm really, really excited about this session. I, myself, I'm tinkering with different tools. I'm definitely not close to your level in doing this, so I'm really excited. I'm sure a lot of people are really excited as well. Our top performing episodes in the past few months are all being around agent development. So I'm sure there's a lot of people who are really curious about this topic. It's everywhere in the news if you're following an eye. But still, I think most people now get a feel on how to use large language models and image generation.

4:36I think 99 % of the people don't have a clue how agents work and how to create them. So I think this is going to be really fascinating. Before we get started, a few messages. First of all, thank you to everybody who are joining us live, whether you're joining us on LinkedIn or joining us on Zoom. We really appreciate you being here. I know all of you have other stuff to do on Thursday at noon Eastern time, but feel free to introduce yourself. Tell us what you know, what you don't know about agents and what you want to know from this session as well. That will help me guide the conversation maybe even more.

5:06Tell us where you're from so we know where people are joining from. That's always fun for me to see. There's always people from all over the world, at least on the LinkedIn side of things. There's always a lot of people from interesting places. So introduce yourself, make new friends. In addition, one last thing that I will say is that if you are still watching us live, the next cohort of the AI Business Transformation course starts this coming Monday. So when this becomes a podcast, if you're just listening to this after the fact, you missed that particular cohort. But the cohort starts on May 12th, meaning you still have a few days.

5:37This is this coming Monday. This course is really transformational for people, for companies, for careers, for complete teams. And we've been teaching it, I've been teaching it for the past two years, every single month, sometimes twice a month. So hundreds or maybe thousands of business people have been through it and have really changed their businesses with AI based on the information that they've learned. So if you have not started proper implementation of AI in your business, this is an incredible opportunity to get a huge amount of information, both on the tactical side of what tools to use, as well as on the strategic side on what's the right process to implement AI company-wide.

6:14If this is interesting to you, I'll drop the link in the chat. If you're listening to this after the fact, then there's going to be a link in the show notes to tell you when the next cohort is, which will probably be in August, because we, in between, we usually teach private courses and we're usually fully booked. That's it. With all of that being said, I will give Pooja the microphone. I'm really excited to see what you have prepared for us and get ready, everybody, to learn how simple it actually is to create agents that are extremely powerful. Thank you so much, sir, for having me here today.

6:46And yeah, super excited to share what I am working on and what I have learned about agents so far. Honestly speaking, Agent TKI is evolving. It's evolving on a weekly basis now. We are seeing something new every week coming up. So it is even new for me, although I have been working in the automation space for many years now. But yeah, Agent TKI is something new. And today, what I'm going to show you is first, like, clarifying a bit. So what are AI agents actually and how do they differ from, say, an automation workflow? Because that's where I think most of the people are still struggling with.

7:19And secondly, showing a live demo in relevance AI about this market research AI agent that I have built and what is the concept around it. Should we get started? Yeah, let's dive right in. I really want to learn as well. Perfect. Let me share my screen quickly. For those of you who are listening to this as a podcast, and I know thousands of you are, we will explain everything that we're seeing on the screen so you're not missing anything. But if you do want to watch what we're doing, you can either switch to watch this on YouTube. So go to our YouTube channel, the Multiply AI YouTube channel. There's a link in your show notes.

7:53Or if you are driving right now or mowing the yard or at the gym, wherever you are when you cannot do this, then you can keep on listening to us. And then you can decide later on if you also want to watch the YouTube video. Great. Okay, so firstly, what exactly is an AI agent? Of course, like by now, almost everyone is aware of what is ChatGPT. So where do agents fit in? So imagine when ChatGPT comes in, even now, to get something beneficial out of ChatGPT, you have to give in a proper prompt, you have to add in your business knowledge, and this becomes a repetitive process, right? And ChatGPT is still a chatbot, so it only talks, it does not execute.

8:33And this is why AI agents are really changing the landscape for the businesses because they can execute. So if you want to understand AI agents in very simple terms, think of them as, let's say, if ChatGPT gets knowledge, so your business-specific knowledge, it gets a memory. I mean, a ChalGBT has a memory, but really, you know, something that is a memory specific to your business, to your style of working, as well as two hands that it can start executing actions in your systems. Now, it can be something as simple as writing an email or managing your calendar, which can be chaotic for a lot of people.

9:12It's a big productivity hack if JGBT can start managing your calendar or something as complex as doing a proper research and then going into your CRM and updating the fields there. So this is what AI agents are all about. Like they can understand, of course, because the underlying power is still LLMs. But in addition to that, they can also execute actions in the systems, in your business ecosystem. Yeah, I'll add two more things. When I think on agents versus large language models. So you mentioned one very important thing, which is the ability to take action. The two other things that agents do that LLMs don't do, one is make their own decisions.

9:55So they're more autonomous. They don't have to be, by the way, but they can be more autonomous than just a large language model, meaning you can give them a broader task and they will figure out the steps on their own without you having to define an exact path on how to get there. And the other one is, if you build it correctly, they can be multi-layered, meaning you can have an organization of agents working together. Like we have teams of humans. You have a manager, you have somebody writing content, you have an editor, you have an evaluator, you have a designer, or a similar parallel in any other part of the company.

10:23The same thing, you can build agents and then they can work collaboratively versus just working on their own, which is what LLMs do. So these are the main differences. Very good point, Ishar. And this is also one of the major differentiators when you start thinking about agents from workflow automation. So workflow automations are really like very static. So let's say if this happens, then this should happen. There is no autonomous behavior there. You have to predefine everything and they execute, which is fantastic for a lot of business processes because not every business process needs that level of autonomous behavior.

10:55However, AI agents, they can select their own tools. As I mentioned, they can select their tools, they can select the next step and so on. Where do you really need AI agents? AI agents make sense when your process is super complex, involves multiple steps, and multiple level of decision-making. If this is the case, then you should start thinking about AI agents. Otherwise, if your process is more about, say, repetitive steps that more or less stay the same, you're good to go with the workflow automation. Yeah. I always tell people now started calling Zapier and Make like AI agents. They're like, no, if it's doing just a step-by-step process, you don't need AI in it, or you can bring AI in it into specific steps if you're trying to analyze what's written in an email, as an example.

11:39But the rest is just old school automation that existed for a decade now. Yeah, absolutely. Oh, by the way, Zapier has also introduced their AI agent builder, which is a no code as well. I think that most of these workflow automation builders are now also moving towards agenting AI. Yeah. Great. And then second point that Issa just mentioned about this AI agency. So that is exactly what I would be showing today. So my agent is about a market research AI agent or rather a team of AI agents. And when you start thinking about this, you have to really think of it like a team of humans that are running this.

12:17So let's think of it. There is a manager agent. and manager agent is the one that is coordinating with all the sub-agents in this team. I, as a human, interacts only with the manager agent. So I only give my request or my commands or whatever I want to do only to the manager agent. Manager agent then picks like the next sub-agent. For example, in this specific use case, so I build this for a competitor research. There are several sub-agents like industry analyst, competitor tracker, customer sentiment analyzer, social media auditor, and then a reporting agent that is really gathering information from all these sub-agents, synthesizing it, and sharing it with the manager agent.

12:59If you think about it, it's exactly like how you would work in a team of humans. And that is actually the basics of even when you start setting these up in any platform. So manager agent is really who you should be telling in detail, think of it like a job description or a very detailed SOP. That is literally your prompt that manager agent should understand. And then the manager agent in return from based on this prompt should be able to execute these or rather pick the right sub agent for your task. We will see this in a demo. I think that will make it more clear. Yeah. Anything to add here, Ishak?

13:38No, I think it's great. I really think what you said is the important thing. Think about what humans you would need and even make it more granular because in humans, in many cases, there's one human that does several of this. So think about it more on the task level rather than on the human participant level. And because every agent will be doing one task that will enable it to be very good at that particular task. And then the manager agent, and then you can also add layers like a improver, like somebody that reviews the work and adds comments and so on. There's different layers you can add, but still each agent will be focused on one task and there's will be one or more agents that will help to coordinate the task to make it more effective this time around as well as moving forward.

14:20Yeah, that's a great explanation. I agree to that because when you look at it, I think in an organization, you would not see a team that where one person is just focusing on industry analysis. So it can be equated to, I don't know, a very well-funded startup where you have one person doing one task. Yeah. So now let's go into relevance AI and I will start describing you how relevance AI is. There's an interesting question, which I know the answer and we're going to actually demo that, but it's a good question to ask. The question is, is the human sets up the sub agents and not the manager agent spins them up in real time?

14:56And the answer is yes. And I think it's a yes, but, so I will let Pooja answer the rest. Yes. I will be showing that during the demo. And if it is still not answered, happy to take that at the end. Okay, perfect. Okay. So what you are seeing on my system is really the Relevance AI interface. It looks a bit messy because I've not really organized my workspace here. But Relevance AI is a no-code AI agent builder. And if you go into Relevance AI, they have now tons and tons of templates. So when you start going, you can create a free account if you want to get started. and they have a lot of agent templates, which are fantastic way to start.

15:35So if you're just starting out, I think the best way would be take one of these templates and start editing it for your requirement. That is the easiest, right? Then there are a lot of tools. So these are really the integrations that are already present in relevance. So for example, if you want to extract something from LinkedIn, they already have a tool for it. All you have to do is select it, use it in your agent or in your project and adapt the prompts. Now let's go into this agent that I am working on, which is the competitor research agent. Okay, so firstly, let's see, maybe I will first run this demo.

16:14So we see the output and then I will start explaining it one by one. So this competitor analyst agent, as I explained, is a team of sub-agents that is doing industry analysis, competitor, so it can go onto your competitor's website, track the latest updates it then goes into say trust pilot or any other review website that you have given access to gets the latest reviews i have added an additional layer there so not just get the reviews because let's say if the competitor is doing well in something i want to understand why or if there are any gaps i also want to understand why so it has an additional layer of sentiment analysis tracking i have also trained it to go onto the linkedin or any other social media, for example, to gather what is happening at the social media of the competitor.

17:02And finally, it synthesizes everything and sends me over on Slack. So let's give it a demo. So let's say I just use a demo use case, okay? I'm a sales head for AI-powered CRM for SMEs. So I'm simply describing what my company is doing in very simple terms. Nothing, like we do not need a page-long prompt for this. And I would like to get the competitor insights, say on Pipedrive, because Pipedrive is like the CRM, which is very like a competitor for me, right? So I would just simply run that. Okay. As you can see, I've simply talked with this manager agent, right? So what you saw on my window was the manager agent.

17:47I simply communicated what I want to do, and it has started working in the background. Now, first thing is it is extracting the industry news for me. So this is another sub-agent that it is running in the background. Now let's open another window. So I can show you what is happening in the background. Okay. So I just ran this like before the session. Because sometimes it is extracting a lot of information from the web. It takes time. So let's see what it did. It extracted the industry news, which is specific to Pipedrive, and which is very specific to the AI initiatives of Pipedrive. Because I specifically said that I am doing this AI-powered CRM.

18:31So it just extracted everything along with the source. So those of you who are just listening, there are multiple articles with each and every one with several bullet points and topics and summary. And then there's a final analysis of all of them together. Yes. So this is one agent that did its job. Second agent, which is the Trustpilot review summarizer. What it did? It went on to Trustpilot. It got the latest rating of Fybedrive. And based on that, it gave me like the key positive points. What is happening there? Key pain points. So for example, it found out that based on the reviews, that there is a recurring issue in the limitation of the form of components, particularly regarding constant tracking for subscriptions and so on.

19:16pricing transparency, and the learning curve. So these are already very good insights for me. Now let's go further. It then did a competitor product benchmarking. So what it did, it went to Pipedrive's website, extracted all the latest CRM features that Pipedrive released because that is what is relevant for me, and it summarized that. And finally, it summarized everything, like synthesized all this information from different sub-agents, summarized it and gave me like a report here. For example, the strengths, what are the pain points, current capabilities, and so on. Now, one of the questions that I often get here is how is it different from a workflow automation, right?

19:59So one thing here is, let's say, had this been in a workflow automation, but if I want to rerun this process, it would always run everything from start to end. Whereas now, let's say if I am not happy with the Trustpilot tracking, for example. I can simply say, I would like to rerun or redo the sentiment tracking. And what would happen here is because the manager agent here is responsible for choosing or delegating to the subagent, it would identify that only that particular subagent needs to be reactivated now. So it is only using the Trustpilot review summarizer subagent and not rerunning the whole process.

20:45And that basically is the big difference when it comes to workflow automation, because had this been like a Zapier flow or Makeflow, it would have rerun the whole thing and not just one particular component of it. Yeah, and this connects beautifully to a question that was in the chat that said that so far, like when you just started running this, it looks similar to deep research. How is that different? And I think you answered some components, but I want to dive deeper into that because I think it's important for people to understand. First of all, deep research is an agent. So the biggest difference between deep research and just using Chachapiti for search is exactly that concept, that it understands your question.

21:22And then it says, oh, the person wants to understand this topic. What do I need to do, which is now an autonomous thing, to actually give him the information that he needs? The biggest difference between deep research and this, and there are several different differences. Difference number one is you can, in advance, define all these sub-agents and you will know exactly what it will do. meaning in deep research you can't control what sources it will go to which one it's not going to go to what topics you're interested in and so on so when you build an agent for a specific topic like in this particular case doing competitive research you can build it the way you want it so think about it like a customized version of deep research that's basically what it's doing so that's difference number first number two is as puja said you can go back and forth with specific components of this because they're standalone agents, which is not possible, or it's possible to an extent with deep research, but it will still not be as tailored and specific as the specific agents that you develop.

22:21And number three, which you're going to see in a minute, this will actually go and update your CRM and do other things that obviously deep research will not do. So think about this as a multi-layered, multi-level, more customized version of deep research that can also then go and do stuff like write it to you in Slack or update your CRM. Anything you want to add, Pooja? Because I think it's an important topic for people to understand. Yes. I think that summarizes it beautifully. Another point, human in the loop. Deep research does not have human in the loop, right? You have no control. You cannot really say, okay, there is no concept of asking for permission.

23:00I will show you in a second that as a custom AI agent, you can really train it to ask for permissions in certain areas, like where you want it to not run on its own. Awesome. Another question that is also, I think, very important, and then we can continue, is there was a question about hallucinations. Is there a way to know or to reduce or to verify the information that is coming from these research agents? It is possible to, of course, like you can train your AI agent to ask for the sources. So for example, when I was building this agent, I was really not sure if it is getting the right information from the trust pilot.

23:40So I was checking it again and again by building it. I was always verifying. For example, the stars that it shows me here are correct. The reviews are correct or not. And the date, because I wanted to only go one month, get the reviews one month older only. So that is something I was always manually checking. And I think this is a very important step. You have to be very cautious when you are designing your AI agent. Again, human in the loop. I think this is one of the most or the ultimate guardrails that can be part of your AI agents. Second, the system prompts. They are the best ways to control your AI agents.

24:16I will show you in a second where you can add your system prompts. So whenever you are designing an AI agent, system prompts are really the controllers. That is the information and the kind of positive prompts as well as negative prompts that you add in your AI system prompt. They are the ones that would be controlling everything for you. Awesome. I will add my two cents. When I do deep research on stuff I really care about and I really need to verify the information, I usually run it on three different AI deep research tools. So think about running three different agents to do the research for you.

24:50And then I have a fourth process that actually creates a comparison table. and checks if all three have the same information. When all three have the same information, then it's very unlikely they all made up the same stuff. And so that's most likely accurate information. And when there are outliers, when only one of them finds a piece of information, the same fourth agent go and checks that information is correct. So again, now I get a second verification if the information is really there from that particular source. And all the outliers are not found, are just being thrown to the trash. And I get a summary of that.

25:23So it just depends on how buttoned up you need this information to be. Let's say there's 500 Trustpilot reviews. If 10 of them are made up, 15 % are made up, still not a big deal. You're still getting the 500. But if you're looking on something that you're going to make a very important business decision on, you want it to be 100 % accurate and not 90 % accurate. And then you can add these additional layers and steps in order to dramatically increase the chances that the information is correct. Great point. Right. So let me show you the concept of human in the loop, which is absent on definitely on the workflow automation, because there is no way that you can add like a human approval step in there or even in the LLM chat box.