Exploring Agentic AI: Revolutionizing Automation and Decision-Making in Business with Krishna Tammana

[00:00:02] Agentic AI, all the buzz. So let's dive into what is it? What does it mean? How's it different from conversational AI? And can we find the consulting opportunities and the ability to help our customers? Krishna Tammana of HoopShoot joins me today for a conversation his organization focuses on these kinds of technologies on a bonus episode of Business of Tech.

[00:00:28] This episode is supported by Comet Backup. Whether you get hit with ransomware, hardware failure, or human error, there's nothing more heart-stopping than losing business critical data. Backups are your final stand when a threat penetrates your layers of defense. That's where Comet Backup comes in. Comet is an all-in-one backup solution. Whether you need to protect computers, servers, virtual environments, emails, or databases, Comet Backup empowers you to manage backups on your terms.

[00:00:55] Visit CometBackup.com to start your free 30-day trial today. Get $100 free credit when you sign up with the promo code MSPRADIO. Start running backups in 15 minutes or less. Comet Backup, the backup solution that MSPs trust. Well, Krishna, thanks for joining me today. Very nice to meet you. Thanks for having me on.

[00:01:20] Now, I'm going to dive right in because I think the area that we're all kind of talking about is this idea of agentic AI. And I know you're spending a lot of time thinking about this. You've got some conversational AI technologies that apply here. Give me, let's start with the basics. Like, agentic AI is discussed as the next phase. How do you define it? And how does it differ from kind of that traditional conversational AI?

[00:01:42] So, agentic AI, Dave, is just the latest installment in the advancements that we are seeing in AI in general, but more specifically, AI driven by LLMs. So, if you think of traditional AI versus LLMs, the ease with which you can get started, get answers, get these LLMs to give really high quality answers in a very detailed manner has changed leaps and bounds.

[00:02:11] But what it did is only provide you answers. You ask the question, you get an answer. It's very thorough. It's very detailed. Of course, there is a little bit of hallucination that can go on, which can throw people off. But if you tune the models correctly for the domain and the use cases that you're dealing with, you can get really excellent answers from LLMs. Now, what do people really want to do with those answers?

[00:02:37] In most of the business world, what we all want to do is, okay, we have a question. We got the answer. Now we want to take an action based on that answer. Often, not always, but very often. So, agentic AI is a method that allows you to take these excellent answers that are coming from these LLMs and trigger all the activities and actions that you want to take fully autonomously.

[00:03:06] Now, the big deal about this is really that these agentic AI technologies are able to do some reasoning and deduction to come to a conclusion and may take the action on your behalf. So, they actually act like human agents.

[00:03:25] That's really the fundamental technology evolution and the opportunity there, obviously, is to be able to automate things much, much better, much, much faster with a lot more precision. So, that's really where we're headed, Dave. I know we're still in the early stages, but this is just going to fundamentally change how we do everything almost.

[00:03:48] Okay. Well, I want to get your take on kind of the risk balance here because it's one thing to say you're getting information and it's excellent. But by the way, that leaves a space for it is sometimes confidently wrong, right? And there's a difference between then providing that information to a human who can do analysis versus fully autonomous decision-making. Exactly.

[00:04:11] And in particular, the thing that I want to get your sense of is like in the first bucket, while I'm uncomfortable with it being confidently wrong, I know that there's humans in there that are doing analysis with that information and that's where it works best. But if we move to fully autonomous decision-making with something that can be confidently wrong and then execute that, I see that the real problem here is accountability, right?

[00:04:36] Because at least if a human goes forth and is confidently wrong, well, there's some responsibility to that human for the decision they make. But if we set forth agents that are then confidently wrong on autonomous decision-making, well, now we're into some really scary territory. Talk to me about the way that you're approaching this balance and this problem and the way that businesses should be thinking about it. You're 100% right.

[00:05:02] Even without agentic AI, if any system, AI or otherwise, gives you answers and those answers are not accurate but presented to you confidently, you're using it for analysis, you still have the risk. That's something we've been living with Google searches, something we've been living with LLMs, and we will live with agentic AI.

[00:05:25] Of course, the degree of importance and the amount of judgment you can apply on it to mitigate that is very, very high in the previous models. If it is fully autonomous, of course, you have a bigger risk, 100% right. However, there are ways and approaches that you can take to only automate the things that you have high confidence in.

[00:05:50] There are many, many things in the world that we do that are tasks that are obvious. There is not a ton of judgment required or even if it is required, the tuning you can do to get a very high degree of confidence is there already.

[00:06:07] So what we are doing is taking all the use cases and scenarios where the risk is very minimal and then we are automating it. And anywhere where the risk is high or our confidence is low, we present it to an agent, a live agent, a human agent, and present it in a form where if they like the answer, they just have to click yes, go.

[00:06:36] So if they don't like the answer, they can redo the whole thing or they can manually override everything that is to be done. But this notion of having high degree of confidence before you automate anything is absolutely critical. Confidently getting answers that are wrong is very dangerous and we know that. It's business risk at the end of the day. If you automate something incorrectly, I mean, it's sort of like having a bad programmer.

[00:07:06] Somebody who does something wrong, introduces a bug, you put it in production that does bad things. And depending if you're doing a financial transaction, the implications are more severe. If you are doing some simple social media thing, maybe it's not that severe. All of those things apply to agent AI world as well. So that doesn't go away.

[00:07:26] So to me, we have to be very careful in what you choose to automate and make autonomous decisions possible and which ones you simply have to aid the human, which is what we are doing. That's the approach we are taking. So I'm sure you've got a couple of use cases that are natural fits. Give me a little bit of sense of the ones that you think are ideal for this kind of technology. There are a number of areas that we are investing in.

[00:07:55] In fact, just this morning, we released a new website with new agents that are in the market as of this morning. I'll give you a few examples. So we have one very specific agent for real estate.

[00:08:16] And what it does is learn a little bit about what you're trying to do as a consumer and then shortlist all the inventory in the market to exactly suit your needs. So before any human can get involved or needs to get involved, you can take all the time in the world, tune the parameters. I want a kitchen of this size. I want my front door facing the other side.

[00:08:45] You can specify whatever you need and it will go. I didn't. And the area, of course, all other related things related to what we all look for when you want to purchase a home or rent a home. You can look at the inventory in that area, shortlisted correctly, and then you actually can fine tune it any number of times till you are satisfied with something that you think is worth going and looking at.

[00:09:12] So you just saved yourself a whole bunch of time talking to real estate agents, whether you like them or not. All of those things can be mitigated. And by the time you get to a point where you want to go see, you're a highly qualified participant in this.

[00:09:29] So for marketing, this is super useful because by the time somebody says I want to talk to somebody, you are very close to actually pulling the trigger or the odds of conversion is super high. That's just one example. The second one, I'll take you to a slightly different area, financial sector.

[00:09:49] A lot of financial sector companies tell us that collections of loans, et cetera, is a complicated process and there are a lot of humans involved. You know, you're trying to reach somebody, you can't even connect with them. You know, there's so many things that are complicated in that.

[00:10:07] And what we do is have an agent, an autonomous agent that simply makes a connection with whoever owes any money. The first time it may be a simple reminder that you owe this much money. The time is coming up. Click here in the link and it can do the transfer. All the payment can be taken care of from your bank.

[00:10:33] If it is more than that, then the reminder followed by another reminder and setting up a live appointment with somebody, et cetera. So those things are fully automated and you can do this at scale. But the nice thing is this can happen in any channel. It can be on WhatsApp. It can be on voice. It can be on any channel that a customer is available on, has a preference to deal with. So those are a couple of examples.

[00:11:01] But there are any number of these things that are coming to us. Our customers telling us so many of these use cases they're trying to automate. And, of course, a human is involved. But all the pre-work is done by the agent, the autonomous agent. And that just simplifies life quite a bit. And these are just external use cases. But within the corporations, everybody is trying to operate their HR processes, finance processes.

[00:11:30] There are so many things that agent EKI can do. A simple example, if somebody is leaving the company, you have to remove their credentials from so many systems in this world today in any company. All of this can, the moment you remove their or mark their employment as done in your HR system, that can trigger all the other processes. The risk here is very low. There's a lot of preset actions that you can take.

[00:12:00] That doesn't take a, the risk is not high. So you can, autonomous agents can take care of that. Now, this, if I, if I'm being oversimplificating here, and I try to oversimplify the problem, the, this sounds a lot like what we've always talked about with automation around customer service. Right. And, and if we went back in time and we did the way phone tree and automated phone systems would do, they would say much the same thing thematically.

[00:12:29] Now, customers, not particular fans of much of the intermittent, the intermediation here that goes on with this. Talk to me about what, you know, what is doing here that's different, that is not removing the human element and causing more customer friction just by making customers feel like there's more automation.

[00:12:50] I think the, the difference from a simple automation to agentic AI is that there is a lot of intelligence that can be baked into these systems. I'll give you an example. So order management, if I'm a customer at any time of the day and night, I can contact my business, the business, the company I'm doing business with.

[00:13:15] So now I'm not waiting for anybody, the office to open, for example, and saying, I received a product that is defective. I want my money back. By the time anybody in the company even wakes up, my problem is solved. The check is in the mail, or if it's a credit card, maybe it has already processed and refunded the money.

[00:13:37] So from a consumer standpoint, there is tremendous amount of convenience for more complex tasks, which was not possible before with simple automation. Here, the system has verified you. It has seen that your order has been delivered. And this is the product, this is the amount, and it is eligible for refund or not, because it is still within the 15 days or 20 days, whatever.

[00:14:05] It has applied all those rules and, and possibly sent an empty shipping box to you to put the product back in, trigger that transaction, and then refund your credit card. All of those things have happened before anybody could even wake up. Now, the convenience, efficiency, anxiety for somebody who just saw a product that has arrived that's not working is all resolved much more efficiently.

[00:14:33] So that's kind of the opportunity to make these things, make everybody so much efficient, and customer satisfaction is going to be higher. If I don't have to wait for the morning and then make another phone call, stand on the line, and then explain to a human agent, and they'll take it to their boss. And so I need approval from somebody. All of those things can be simplified. It all sounds like ideal, right? Most customers are going to want to do this.

[00:14:57] But I also know that in the real world, most customers aren't ready for something like this to just be immediately bolted onto the side of their business. Give me a little bit of a sense of what the most successful customers have done to sort of prepare their data or their backend systems to be effective for Identic AI tools. And now we're getting to brass tacks of what really, how do you make this come to life?

[00:15:22] So there are a number of things that have to happen, which are not far from what we had to do in the past. The number one thing is, if you have an order system, you need an API available. If it is an order system that you're using that you purchase from another company that's, you know, any number of order systems are there in the market, and those APIs are readily available, then that becomes simple.

[00:15:51] But if you have a homegrown order system, you have to make that API available and securely make it available for somebody like us to reach out. That's just one simple thing. The other thing is your inventory systems, your processing, back office processing systems, all of those things have to have APIs that are readily available. So this is the number one thing that you have to do.

[00:16:19] To take action, you need the ability to trigger these activities through automation and not through a human interaction. Now, though, even if it is, you know, some kind of screen scraping, et cetera, is possible, but that's not ideal. The best thing to do is make sure that you have APIs. That's usually where we end up waiting on our customers. Some customers have these things available and ready. Some don't. Is there stuff around data that has to be prepared?

[00:16:48] Because it would feel like, I mean, I could give you API access, but if I don't know what the agent should have access to or I've given all the data, maybe it starts giving away my secrets, you know, in terms of customer interaction. How do you have to prepare the data sets behind those APIs? I think typically these APIs are fairly limited in what they can and will do. But there are some applications and systems where the API can do everything.

[00:17:16] So you have to make sure that if that is a situation, you have to create another API that is very limited in both permissions and what it can do and what data you can pull and cannot pull. That's a very good question because this is something that, you know, highly flexible systems allow you APIs that can do lots and lots of things.

[00:17:38] But we cannot do that with agentic AI or any automation for that matter that has any risk of real action that can impact business or the customer. So just limiting those. There's a little bit of review that has to happen before you give these things to autonomous AI as to what does the agent need and what. APIs are available that can do just that and nothing more, nothing less.

[00:18:08] So there's a little bit of matching of the needs to the API availability that has to happen. And of course, after that, you have to have ways of collecting data to do the analysis to make sure that everything went exactly like that, exactly as you planned. But that has to be done at scale.

[00:18:27] So that has to be ready in the back end such that every day there is some report going said this is how many orders are processed by agents, autonomous agents. And this is a success rate. This is a failure rate. You know, there are many other situations that happen. Even as a customer, I might start with something that is an automated system in front of an agentic AI. But I might say, no, I really don't like it, but I need to talk to the live agent.

[00:18:57] So business needs to understand how often this is happening, why this is happening, and how to improve this, and the accuracy of how it has processed these things. So there's a lot of analytics that need to come with it. So that's part of what we do is every time there is an agentic AI-based use case that we automate, we also include the analytics for it to make sure that improvement can happen as well as business can understand what's happening.

[00:19:26] And is customer really satisfied or not? You could actually verify that has this been a satisfactory experience for you, and that goes a long way in making sure that what you're doing is actually valuable and it's helping as opposed to hurting. So that's really, really crucial. Well, we have just uncovered what part of the solution provider value is too, so that's where I like to end things. Christian Tamana is the chief technology officer at Gupshup, a leading conversational engagement platform.

[00:19:55] With over two decades of experience, he's held significant roles at companies such as Sybase, E-Trade, Splunk, and Talent. Christian, if people are interested in reaching out and learning more, what's the best way to do that? Just reach out to Gupshup. On the website, we have a demo request. We have emails you can reach out to, or you can call us. Awesome. Well, thanks for joining me today. Thank you very much, Dave. Nice talking to you.

[00:20:20] With every new breach and threat that I cover, it's clear that cybersecurity isn't a luxury anymore. It's a necessity. That's where Huntress comes in. Their fully managed cybersecurity platform is built for every kind of business, not just the 1%. Huntress seamlessly integrates their products and threat hunting team. Their EDR, ITDR, SIM, and security awareness training solutions are purposely built for their elite 24x7 security operations center

[00:20:49] to stop threats before anyone else even spots them. This potent combination of purpose-built cybersecurity and threat hunting expertise is one of the many reasons why G2 users have voted Huntress the number one rated EDR for growing businesses. To see what people-powered cybersecurity looks like, visit Huntress.com slash MSB Radio.

[00:21:12] The Business of Tech is written and produced by me, Dave Sobel, under ethics guidelines posted at businessof.tech. If you've enjoyed the show, make sure you've subscribed or followed on your favorite platform. It's free and helps directly. Give us a review, too. If you want to support the show, visit patreon.com slash MSB Radio, and you'll get access to content early. Or buy our Why Do We Care merch at businessof.tech.

[00:21:41] Have a question you want answered? We take listener questions, send them in, ideally as a voice memo or video to question at MSP Radio.com. I answer listener questions live on our Wednesday live show on YouTube and LinkedIn. If you've got a comment or a thought on a story, put it in the comments if you're on YouTube, or reach out on LinkedIn if you're listening to the podcast. And if you want to advertise on the show, visit MSP Radio.com slash engage.

[00:22:09] Once again, thanks for listening, and I will talk to you again on our next episode. Part of the MSP Radio Network.