Sign in to continue learning
Design, Develop, and Deploy Multi-Agent Systems with CrewAI
Beginner
3 hours 45 mins
Topics
Agents
Next up, you're going to hear about ABInBev. They own over 500 iconic brands across the globe. They are the world's largest beverage company, and they're using CrewAI for multi-agent systems across their entire business, from their commercial digital platform, all the way to automating back-office use cases. They're a massive company with a lot of potential for using AI agents, and a huge partner of us in this journey to reinvent the workforce by using AI agents, and you're going to hear from them next. So let's jump into that one. Hey there, Daniel. Thank you so much for coming in today. I'm very excited. I mean, people have been hearing me talking about agents and just learning all the things, the ins and outs, and how to view them, how to test them, and everything in between. And the reason I'm so excited to have you specifically on this is because you're on the other side. You're one of the companies that are building. When you folks decided to explore AI agents, what were some of the main challenges that kind of led to that? What made you think about, well, maybe we should do AI agents? Super happy to be here with you today because, you know, we are big fans of CrewAI, and CrewAI has been our big partners in this journey since the beginning. I think just to start, João, so we started using CrewAI first with the open source version, and I think there's an important caveat here. CrewAI actually helped us on building our strategy for agents because at the beginning, we were like more trying to understand and to learn how it's going to be now the processes, agentic oriented, I would say, right? And this concept of the crews and the flows was something that helped us a lot and actually create an internal framework that now is helping us on mapping the processes, which brings me to the answer of your question. I think the processes are the biggest challenge that we have on how can we actually get the process mapped on a way that we can translate into, first, what are the actions that must be taken? And then what are the agents that we need to create and what are the tools that those agents must use to take those actions? So far, I think this shift on translating our processes into this view of crews, flows, workflows has been something that has been very helpful for us so far. And I know that you folks already have some use cases live and have even bigger things planned for the future. What are some of the early impacts that you're seeing on deploying these agents? Yes, I think first, the productivity on developing the agents. At the beginning, to be honest, I think we struggled, but because of the lack of knowledge of our teams, right? So I think at the beginning, we were struggling on how to build the agents, what is the best way to build the architecture for the agents, and also integrating into our systems. Because one thing that was super important for us was to not create new systems, right? So we really wanted to embed the agents into the existing ones, because then we can actually transform and get the efficiencies into the processes. So I think from the technical perspective, we had a lot of productivity gains in terms of overtime, right? Because when we got proficiency on building the agents using CrewAI, then it's been fast, man. And after we got into the enterprise version and started to use the studio, even the business folks have been using CrewAI now. And man, to be honest, my life has been turned into a crew life here, right? Because as everyone is using CrewAI, not only the tech folks, but also the business folks, at the end of the day, this is very good for us because we believe in creating this network, right? So if we can create an agent and then this agent can be replicated, we use it across the functions, across the internal teams, and also between businesses and tech teams, then we can achieve productivity, right? So this is something that has been very, very powerful for us. So what you're saying is like, I want to embed these agents into existing systems so it's easy for me to distribute. So people already are using and they don't even notice. It just gets better. And by doing that, it makes it easier for them to adopt. And that's one part. But the other part that you mentioned is this idea of like, well, we need to have this repository, right? We want to make sure they're not doing the same things twice. We want to do once and then roll it out globally, make sure that we're using across all the different units. That is a very interesting take, and I'm glad to hear that we're helping with that. Now let's talk about the challenges. What do you think? You mentioned like getting the team educated was a challenge at first. Is there anything else that comes at point and how did you fix that? Yes. Getting the team up to speed was a big challenge for us, and I'd like to highlight here the way that we have been, because it's already ongoing, right? The way that we have been doing the workshops and the training sessions across the globe. And for us, this is something that is actually big, because we're talking more or less about 40 countries, right, that we are expanding. So for those who doesn't know, ABInBev, we have around six zones, which is our compositions of countries. Out of the six zones, I think now we have three to four already using CrewAI and expanding. We're going to reach all the six zones using CrewAI by the mid next year. But one of the biggest challenges were to scale, right? Because then when we got CrewAI up and running, then we need to train the teams. I used to call this internally as our pre-season. We need to run the pre-season with the team first, then we go to the tournament, right? So that's more or less the framework that we created, partnering with you guys. We created internal training sessions, training our teams in our internal platforms and tools right after that on CrewAI. Speaking about the technical challenges, I think scale is the biggest challenge, man, especially for us because we are so fragmented across the globe. But on the flip side, the challenge, the use cases are almost the same thing across the globe. Because if you talk about supply chain or even procurement or logistics, the challenges are more or less the same thing. Because at the end of the day, we have different systems across the globe. So this is also another big challenge for us. Because if we think on ERPs, internal systems, even payroll, right, which is something that has been building agents to automate a lot of our tasks, man, we have at least more than 15 different payroll systems across the globe. So this is a big challenge because we can replicate the template of the agent, I would say, right? Internally, we have been calling those agent templates. But we can replicate the template. But at the end of the day, we have something that are specific per country, right? So this is a big challenge because when you talk about, for example, MCP or connection with tools, so we can't just build one MCP server that's going to cover everything in the payroll domain, right? So we need to build some things that can be scaled across the globe. This is a huge and massive engineering work. But also if, as we are leveraging crew and having the same agentic framework being used by the whole company, then it's much easier for us to scale the agents across the globe. But this is something that has been very good for us, right? But it wasn't our first catch, right? So it came with the mistakes that we made at beginning. So far now, I think we found a path. Like a lot of these companies, they're trying to get efficiency gains. And the way that you get efficiency gains is like how you manage to train your team at scale to start using and deploying and building this. That is a very interesting take. And you mentioned about CPG as an industry. To talk more about that, it seems that across the entire different sectors and business units, there's a lot of commonalities going on. What do you think are some of like the critical considerations for these companies on CPG that are trying to understand how to apply AI agents into their day-to-day? One of our biggest learnings is to go deep, right? Try not to go so wide and go deep. I would say, for example, that we defined a few priorities in terms of processes that we may want to transform. Because when you talk about CPGs companies, usually you are talking about big operations, right? Heavy operations, logistics, distribution centers, in our case breweries, huge sales operations. So we have people spread across the globe. So it's very difficult when you think on how can we transform the processes so far. Of course, we have business services centers, which makes our lives much easier when we want, for example, to focus in a few streams. But at the end of the day, taking a decision on going deeper, deep in one process, it's been a big difference for us. So in our case, we try to tackle the ones that we have a certain level of knowledge and we can actually map the process end-to-end. After that, we evaluate, for example, the maturity level of the technology, right? So for example, if you don't have digitalized systems in this stream, people are using Excel, I would say, right? Which is pretty common across all the companies in the globe. And it's going to be slightly different from actually integrating the agents into existing systems. So different discussion, different tech approach, different way on mapping the process. So we consider a few factors on evaluating on how to scale. But at this moment, we prioritize more things that we can control internally. I would say, for example, functions like legal, finance, not so much sales, but things that we can actually control end-to-end, different from sales distribution. For example, that we depend on our partners in the field, which is going to be much more complex to implement things in there. But the functions that we have control end-to-end were our decision to get started. I would say that in terms of the challenge, the technical challenge, because now we are sharp on using CrewAI, I think the biggest challenge now is more on mapping the processes and actually designing how it's going to be the agentic flow for those processes. And on the flip side, to not actually try to create a lot of unnecessary complexity, right? Because sometimes this happens, you know, right? So the engineers got excited with the technology and they tried to create things that are not actually needed. You go into any company, as you said, they will have some similar pieces on the stack, right? There's always like a data management, and then you also have ERPs, CRMs, you need to integrate with all of those. How you're thinking about that and navigating that, because I know that's something that every company is thinking, especially when they're considering other options. Yeah, this is very challenging because in our case, we, for example, are parallel running big programs for digitalizing some processes, right? So everything is different when you look across the globe. So that's why I think the agents can play a big role, because I think if you compare to the past, where I think almost 15 years ago, when we got the API gateways, right? So all of the companies that have been trying to start their digital transformation journey, the API gateways played a key role, right? Because you have a big legacy here, but in a way you must deal with this legacy, but you need to actually to build the progress, you need to build the future, right? So I can compare now the agents that we are building. So this whole agentic strategy powered by, in our case, CrewAI plus our internal tools, but also with this new flow with the API gateways in the past, right? I think it's more or less the similar situation, but right now we have GenAI, right? So we have the LLMs powering this process, which man, makes us 10, 12, 50 times more productive than in the past. I love that. And yeah, I think you got something right. That's absolutely correct, because what I have heard from other companies, right, is they learn from their mistakes, right? They have seen what works and that worked in the past, and they got burned in certain movements that they did. They got vendor locks in certain places, so they're being very mindful about it going into this, knowing they want to move fast, they need to get results, they see the potential. And for the ones that are getting use cases like that, like yourself, you're seeing the results and you're doubling down, right? But they also want to make sure that they want to keep things as flexible as they can, because the market's moving so fast. A new model comes up, you want to be able to use that on your use cases. A new offering comes up, you need to be able to integrate it. There's a lot going on. I really appreciate you having the time to share and talking with us today. I know that you're super busy, like changing ABInBev from within, and that's exciting. I don't know if you have any parting thoughts or anything that you want to tell for the students that have been tagging along up to this point, but I do want to thank you for making yourself available. No, man. Thank you for inviting me. I think that's important to spend this time spreading the word, right? I think it's super important. As a final thought, João, I think one of the biggest learnings for us was exactly, and I think it connects to the course, right? So do a good pre-season, right? So it made a big difference for us. That's why I believe this course is going to be very helpful for everyone which is considering using CrewAI. And my last takeaway is right now, with so many things coming up in the AI space, in the market, I think it's good for us, people that are taking the technical decisions with the companies, to double down with some technologies and actually to make sure that we can build some things that can scale. And for that, I think the way that you guys designed the whole framework, man, it's very powerful, right? So first, let's do a good precision, and second, let's double down at least in one good technology that connects into our ecosystem. Because of that, we can make things scale. So those were my takeaways for this conversation. Great advice. You did a killer pre-season. I love the games that are going on. I'm excited for the Super Bowl. That's it. That's what I want to see. Thank you so much, Daniel. I really appreciate the time. This was amazing. Thank you so much. Thank you, João.
