Start small before you sign a five-year, multi-agent deal | Anatolii Iakimets, Director of Product Marketing, Kibo | Ep. 8

[00:00:00] Welcome to Data vs. Commerce, where we explore the messy middle between database and doorstep. I'm Matt Johnson. And I'm Floyd Blakey. Let's dig in. Okay, I am here with a very interesting character. His name is Anatoly Iakimets. He is the Director of Product Marketing at Kibo, and he tells me that his last name is harder to pronounce than his first one. So I'm curious, how did I do? Did I get it right? It is pretty, pretty good. Spot on.

[00:00:28] It's pretty good. Okay, that's great. I am really excited for today's topic because we're going to talk about AI agents, but in a way that probably not a lot of folks have talked about them before. And we're going to dissect them like it's high school biology. But first, Anatoly, can you just tell us all who you are, what's your background, and what are you doing these days at Kibo?

[00:00:49] Anatoly, I lead product marketing at Kibo with focus on B2C commerce. And I've been in this space for over 12 years, predominantly in commerce and telecommunications. And my first adventure with artificial intelligence was around 2015-16, when it was still the age of machine learning stuff and neural networks, and everybody was super hyped about it.

[00:01:15] So I have a history with AI, and it's near and dear to my heart. So I'm glad to be here. Amazing. And that's the hot topic. I don't know if you've heard everyone's talking about AI, maybe too much. But if we dig into how that's really playing out in the ecommerce world, everybody is talking about, of course, agentic commerce has been the subject of like every press release from Google for months now.

[00:01:42] And what that means is just deploying agents into your commerce workflow, right? Does it get more complicated than that?

[00:01:49] Yeah, I think everybody is trying to figure out what they can do with agents. And there are a couple of areas. One is pretty obvious where everybody's trying to make their customer experience on the front end cheaper and more efficient in terms of customer service agents. And you see commerce companies doing it, customer service companies doing it, basically trying to reduce time to resolution and resolution cost.

[00:02:16] And at the same time, when you go on the more on the backend or back office and operation side, how do you actually make your employees more productive? This is where everybody's still trying to figure it out. Because for commerce, it is not as obvious as for some of the other areas like for example, coding or software engineering.

[00:02:38] So what is an agent? I'm mostly not playing dumb. I think I've got a pretty good sense of it. But people talk about, you know, they might talk about Claude, like he's their best friend. They might just say AI to refer to LLM and various tools. Some people might distinguish between AI as a concept and a large language model as a tool. But what is an agent? Can you break that down for us super clearly?

[00:03:04] Yeah. And feel free to stop me if I talk for too long. But in a nutshell, if you think about an agent, there are two distinct parts to it. The LLM and what people call harness or scaffolding sometimes.

[00:03:18] The LLM is a straightforward part, whether it's Entropic or OpenAI or any other frontier lab like Google, for example. Or it can be an open weight model like Kimiki 2.7 or any other model. Like for example, there are a bunch of other models. DeepSeq, for example.

[00:03:37] And then there's this other part, which is the harnesses mentioned scaffolding, which is all the code that make the agent work. And it's basically the software as you would think about it in a sense like any other application. And it has webhooks, it has events, it has what is called the heartbeat, which allows the agent to run, like for example, every 30 minutes evaluating what are the outstanding tasks.

[00:04:07] It has scheduled jobs, it has ability to pull data from sources, it has tools to do search. So for example, to give you like an example, what the harness means for you as a user in terms of experience is when you go to an agent, and you ask it to write a blog post, you will see that it splits your request into multiple tasks.

[00:04:31] Like, like, like, oh, I need to go and search for it, I need to pull data from the search, from the source. And then I need to, for example, to go back, after I write the draft, I need to go back and check whether the draft actually matches the prompt. And sometimes people say, well, it doesn't work well, it doesn't answer the question. So I need to do one more iteration. So all those things that kind of direct the LLM what to do is the harness or the agent.

[00:05:00] Okay, so if I were to just slice an agent in half, about half of that is the model on the model is the thing that we're all really familiar with, right, we all have our own preferences, whether it's, you know, Claude, ChatGPT, and then the harness is the actual stuff around it, the instructions that tell it what to do. Yep.

[00:05:19] So for most agents, then it's probably true that the model is kind of interchangeable. Like if I've got an agent, and I've decided that, you know, best friendship with Claude is over now I'm best friends with ChatGPT. You know, that's simplifying it considerably, but you can drop different models into these agents, whatever they're doing in your workflow. And that it might make a difference, but the harness stays the same code around it. Yeah.

[00:05:46] And basically, if you think about it, the models are becoming the commodity because in the sense of like, it's a funny situation where from one perspective, you have all the brightest minds in the world, all the people with tons of PhDs working on these models to make them better, smarter. But at the same time, the gap between models shrink. And then at the same time, also, you as a user, you can pick whatever model and they have similar costs. And for you, it's almost a commodity.

[00:06:14] The switching cost for models is not that high for you as a user. Okay, people have preferences, but it's kind of becoming almost like a commodity. And even the Sam Olpont talked about it, how models will become the intelligence on a tap, while the harness basically becomes the differentiator. The agent becomes the key differentiator if you're providing the agent because the model can be anything and how well your model runs will depend on the harness on the software that you write for the model.

[00:06:43] Okay, so as e-commerce leaders are thinking about introducing agents into the commerce workflows, you're saying that the harness around the model is the critical decision point in selecting that agent. So it sounds to me like this is just another software decision, like the same ones we've been making for over a decade, right?

[00:07:05] Like we need to choose what set of instructions, whether a human wrote them or another AI, are going to tell the model what to do. So in thinking of it that way, where a lot of these agentic commerce decisions are kind of like platform decisions in a trench coat.

[00:07:27] What are the, I mean, guess, I guess give me an example of a type of agent that an e-commerce leader might select and what they might be thinking about when they're trying to make that selection in the best way possible. Yeah, absolutely. You're absolutely right. And this is also an interesting situation because also like if you think about it, how the agents in a commerce space differ from agents in software development, it's actually very different.

[00:07:52] Because in software development, you're not necessarily concerned with what, like how the code is written exactly. Like, okay, as long as it works, of course you have requirements, but as long as you meet the requirements, you don't care how the code is written. And the same way people who use agents to write text, they not necessarily care about the specificity of the text as long as the idea is conveyed. And this is good because this is the inherent nature of the models. They are not deterministic.

[00:08:20] Every time you ask a question, you get a slightly different answer. And it's fine for coding if it wrote the function slightly differently and it works, still you don't care. If it wrote the text slightly differently and it still conveys the idea, you don't care. Commerce is completely different world because if you think about commerce stack, basically everything is deterministic. Like your order has a specific price. You cannot just put whatever. Your taxes have to be specific. You cannot just be like, that's good enough. They have to be accurate.

[00:08:50] They get out of that. Yeah. IRS will not be happy if your taxes are just about right. It's close enough. Close enough. Yeah, same products. I mean, customer wants this specific product. If it was, it is a slightly different size of t-shirts, like one size larger, like they will not be happy one size smaller. And it's like the only things in around commerce, which kind of may not be deterministic is like product descriptions, but even then it's okay.

[00:09:17] So, but vast majority of the data is pretty structured and you want the agent to be very specific and very precise in what it, what it can do. And this is why kind of the, it boils down to, okay, this almost looks like you should treat an agent in e-commerce, like a platform choice, like an application where you want it to perform certain functions and very kind of precisely. And with great degree of accuracy.

[00:09:45] And going back to your second, second part of the question in terms of example. So for example, the customer calls a sales rep and they ask them about their order. And so the sales rep pulls up the window and asks, okay, the agent, this customer asks, ask me, where is my order? This is like the order number. And then the agent goes and pulls the data. And of course you want it to be accurate.

[00:10:09] You don't want it to make sure that if the agent says, well, the order will come tomorrow by 2 PM, that it actually comes tomorrow by 2 PM, not like two days from now, or it messes up the address or whatnot. And this is kind of the important part. And the big question is how do you make this happen? And like, what is the difference between making this happen and not making this happen?

[00:10:29] And it goes back to having the harness and for it to be able to talk to different systems because agents, they don't just can go and magically do stuff. They need to have, talk to a system via an MCP, model context protocol server, or via API. But in the end, it's the same basically integration question as you would have if it was not an agent, but another system. It would be the same kind of integration question, the same data question.

[00:10:58] You want to make sure your data is available, accessible by the agent. It knows how to read the data because for example, if you're, if you just tell agent, well, the status of the order is five. It doesn't know what it means. It can go to documentation, read about it, but it will burn tokens. It will be inefficient.

[00:11:15] And so the point is it's about, it's about the agent knowing about the systems and knowing what the data and having the, the proper instructions, how to interpret the data, which goes back to harness. So this case, if somebody would think, oh, we will just buy the commerce platform. It has this agent and it will automatically talk to order management. That's not going to happen. You will have to do the integration part, no matter how you approach it.

[00:11:43] And this is kind of, I would say this is the challenge and everybody needs to get comfortable with it so that, okay. You need to treat agent as a application that needs to be integrated properly with all the systems, have all the access to the data and being able to work with it properly. It sounds like some of this is almost like hard coded context, right? Cause we know that, you know, AI tools like Claude, they don't work super well if you don't give them the right context.

[00:12:12] So part of this is giving it context, but you said, you know, if you don't do it, this, if you don't do it properly can burn a massive amount of tokens. And I just saw, you know, I think we all saw the story about Uber burning through its entire AI budget in four months. And then having to put a hard limit on how much folks were using it and the ways they were using it is part of the agent selection, you know, consideration set.

[00:12:39] How much is this going to cost and how can I make this more kind of increase my ROI or make this, you know, give me the most bang for my AI buck? Yeah. And this, yeah, this is true. And this brings kind of to the next important part of the agent, the model switch and they bring your own model because the reality is every single company is moving towards the usage model.

[00:13:02] Maybe the consumer part will stay on subscriptions with like hard limits, but everyone is moving towards the usage based in enterprise and business B2B. We saw a copilot earlier this year from Microsoft originally, they had a subscription tiers move towards usage based model. And in general, that's the only way to go. And the prices will probably go up.

[00:13:24] And in this environment, because nobody's going to pay like, like $500,000 per month for you, if they're like a commerce vendor or a customer service vendor will not pay for tokens for you all this insane amount of money because they need to make money out of you. And the only way for you to have full control in this situation is to be able to switch models because you never know who will increase prices first. Will it be Antropic? Will it be OpenAI? Will it be someone else? What will be the terms?

[00:13:54] Like how it will happen? You will want to have the ability to switch towards open weight models, which are inherently cheaper. And so you want to have this optionality. And that's why like we believe at Kiva that bring your own goal is extremely important. And it's the only way that's the only reasonable way for anyone to proceed. Okay. So we're thinking about what is the harness or what are the instructions around the agent?

[00:14:20] We're thinking about, you know, models being virtually interchangeable. I know there are significant differences between them, but functionally they are interchangeable. We're thinking about making the best decision for costs and keeping it very efficient and effective. So if you're a commerce leader, whatever part of your digital supply chain you're thinking about introducing agents to, like what is the number one thing that you need to know to make a good choice that you probably don't know today?

[00:14:49] I would say like the three aspects that a commerce leader like need to keep in mind or how to approach the agent evaluation would be focus on like three things. First is composability or flexibility. And I will elaborate on each, the second one is simplicity. And the third one is ability to start small and scale. And the reason is because that's like, everything is going in AI is evolving rapidly.

[00:15:16] Like even what we're discussing now might not be true in six to 12 months. Might something might change. We never know, but it most, but it is true at current moment. And so in terms of composability flexibility, it's about being able to change, to choose your model. It's about being able to switch to a different agent theoretically. So the software needs to have the MCP exposed for another agent to use. So you want to have those kinds of options, which boil down to just to summarize. So switch your model.

[00:15:46] We have the ability to switch your model, have make sure that all the software has proper endpoints for agent to use like MCPIs and agent to agent protocols. Because well, you might want to use your own agent in the future. The simplicity goes back to the fact that everyone has a very different agent strategy. If you look at the vendors, some of the vendors, they have their position as having like 20 agents. And I'm not exaggerating.

[00:16:14] And I went to one of the vendors' web page and they literally had like 20 agents. And it is very hard for you to navigate this space because, well, there's one, like for example, one software has 20 agents. Another one has six agents. And yet another one like CMS has like another 10 agents. And then you have like suddenly 15 agents and you need to understand how they interact, how they kind of going to work together, what use cases are where. And it kind of can become overwhelming.

[00:16:41] And in terms of simplicity, that's kind of what like we at Kiba our vision is. You have one single agent experience, like basically one entity to deal with. And then there, of course, there are multiple agents in the background. But you want to make sure you simplify how your organization deals with agents. Because unless you are a company that is extremely large and you have already, I don't know, internal development teams, multiple development teams that work in an agent.

[00:17:10] You want to simplify because you want to make sure you can measure the ROI. You know the use cases. It's easy to use. And then this start small and be able to grow. It's kind of an extension of this because, okay, you want to start small. You want like ideally one agent experience for everyone to use. So there's one agent they can go ask a question and deal with it. But then in the future, you want to make sure all use cases are covered.

[00:17:37] And this is kind of how we think about it at Kiba. And I think it's a pretty good way to think about it. We don't think about agentic in terms of splitting it into specific functional agents. Like for example, supply chain agent or returns agent, because sometimes the borders might blend depending on what you want to do. An agent might interact with multiple systems. We think about it in terms of intent. So there are five functions we have.

[00:18:07] Engage, which is basically when the shopper interacts with an agent, it's engaged. That's it. Think of customer service agent. Then there's explain function or where if you ask it a question, what's going on with something, with an order, with a product, with a promotion, explain it to me no matter what. It gives you a descriptive answer. Then there is a configure agent where you ask it, I want to change this. Basically, you want to make a change to a system.

[00:18:34] Then there's analyze agent where you ask it for data slice and it kind of provides you with analytics information. And then there is finally optimize agent, the boss of all agents, which kind of runs in the loop to optimize your operations. This is kind of autonomous loop agent. But that's kind of the way you kind of how we're thinking about it.

[00:18:55] And we think it's a pretty beneficial way because as a commerce practitioner, it's very easy for you to understand, okay, I want an agent that will deal with my customers. Okay. That's an engaged agent. It's very simple, easy to understand, or I'm not sure I want my employees to do changes in the system with an agent. Okay. You don't need configure agent. You can focus on just start small, see how many queries employees do.

[00:19:20] You can do research in terms of what questions employees want to ask it, and you can kind of start small and expand later. And this is what we see as well, because when we have some customers deploying it in beta mode, and we already see that explain function is the most popular because people are still kind of scared about using it. Just, you know, this is agent, it will do whatever.

[00:19:44] But when they have a function that just like answers their questions, the same as Google search, they're much more comfortable, much more likely to use. So that's kind of to summarize my long winded answer is that make sure you have like start simple, start small. Make sure you have the, the vendor has the, all the capabilities. If you want to go larger, you want to increase scope and also make sure you have the optionality to go composable to switch agents, to switch models.

[00:20:13] Because you basically, the last thing you want to do is you don't want to end up in a situation where you go to a vendor, you sign a three years or five years, huge deal with a lot of kind of commitment with multiple agents. And then you're locked into a specific vendor with specific LLM and then you realize, well, actually, you know, how many employees don't use it? And it actually turns more tokens than we wanted. And we don't have kind of an ROI. That's the biggest risk you might end up with with agent. Okay.

[00:20:42] B-Y-O-A coming to a platform decision near you. All right. This is super illuminating. I love that, you know, as cool and futuristic and high tech as this all is, it really boils down to the same kinds of decisions we've all been making for a long time. And if we use those skills, I think we'll be in a good spot. And Atoli, thank you so much. If people want to learn more about you, what you do, find you online, do you want to tell us where to find you or would you prefer to stay anonymous?

[00:21:12] Yeah, I'm on LinkedIn. I'm on LinkedIn, although it will not be easy to find it because my LinkedIn tag is my name, last name, which is not going to be. Spelling is easier than saying it maybe. I'm not sure. We'll link it. We'll link it on the episode. All right. Thank you very much, Floyd, for your time. I'm so glad to be here. Thank you so much. Thanks for tuning in to this episode of Data vs Commerce. New episodes drop weekly.

[00:21:38] So if you're responsible for any part of how products get from a database to a doorstep, subscribe now on Apple, Spotify, or wherever you listen.