Welcome to this course on multi-agent systems with CrewAI, where you learn to design, develop, and deploy these powerful systems. This is built in collaboration with Joe Moura at CrewAI. One of the most important skills in AI today is how to build agentic AI workflows, and multi-agent systems, where you take a complex task and break it down into subtasks for different specialized agents to execute, lets you very efficiently build fairly complex workflows. In this course, you learn about the core fundamental thinking frameworks that helps you approach complex tasks and how to break them down into multi-agent systems to look at exciting examples, including code review, and building a deep researcher, and building a meeting preparation assistant, and much more. Delighted to have Joe back with us here again to teach. Thank you so much for having me. I'm so excited about this. And honestly, there's a lot that we're going to go over this course. We're going to talk about how you can, as an AI builder, use the CrewAI framework so that you're building something that is ready for production eventually, once that you're ready to bring it there. But we're going to talk about how you build them with agents and flows, how you track performance, how you implement memory, how you control versions, monitor traces, and continuously improve throughout the agent loop. We actually introduced a lot of features coming this course to enable low-level control, so you can focus on the fun parts of building agents. So this course is not only about identifying the use cases, but also building these agents and get to value as fast as you can. And that's one of the reasons I'm so excited about this. One of the reasons I always enjoy talking to Joe about agentic workflows is Joe is an early visionary that started building CrewAI, in which I'm privileged to be a minor investor, started building CrewAI long before multi-agent systems was popular. And what that means is I think Joe, more than the vast majority of humans on this planet, has seen so many implementations, practical, valuable commercial implementations, at scale, very valuable businesses. And so I know he's looking forward to sharing all those lessons with you. So I think of the basic process of looking around your life, your work, to try to identify what are the things that are amenable to rapid prototyping experimentation with a multi-agent system. And then also how to use CrewAI, say, to quickly build a prototype and what are the tools, Whether it's MCP servers, web search, all the way through to when you have something that works, what are the lessons to take that prototype, which may run great on your laptop, to scale it to thousands or tens of thousands or more users, hopefully without too much pain and too many pitfalls along the way. A thousand percent. And I think you got the nail in the hand there, where literally as you're building these use cases, what makes them interesting is the fact that they're non-deterministic, is the fact that these agents can tap in so many different sources. So they're not only using cognition, but they also have this ability to self-heal and to react at anything that you throw at them at runtime. But that cannot come at a cost of not giving you reliable, repeatable outputs. So that's something that you should watch for. So in this course, you're going to learn how the building blocks for these multi-agent systems, including not only agents, tasks, and crews, but you'll learn how to customize and optimize your agents with memory, tools, including MCPs, execution hooks, and guardrails. You also learn how to add a layer of low-level control with CrewAI flows and states that allows you to not only test your application using metrics, but also inform you how to even train your agents using human feedback and improve your application over time. And in the final module, you'll also hear from several companies that have been using multi-agent systems and CrewAI in production. So even if this is your first time working with agents or CrewAI, you'll get a lot of this course. And if you took our earlier courses in partnership with DeepLearning.AI, I think of this like a phase three, where you're building these things, not only for scale, but quality and reliability. And I'm very excited about that. So one of the challenges that I know you spend a lot of time thinking about is reliability. Very often, we can build a quick and dirty prototype for agent workflow. It works in demo state for that one input we try on our laptop. Maybe even works well for 10 inputs. But then as you go from 10 different examples to 100 in production, to 1000, then does this mean that, well, it's 80% reliable, it doesn't really work. But it turns out that for teams that know how to drive an agent development process, this agents only work in demo, only works 80% of the time, there are lots of tools to break through that. A thousand percent. And what we have seen is that there's a spectrum of things that you can do in order to improve the agents, in order to improve the reliability that you're getting. From using CrewAI training, to use guardrails, to use CrewAI testing, all the way to actually fine tuning a model if you really want to go there. So all that goes into agents memory, and we're going to talk all about this throughout the course. So it's very exciting. The genie is not getting back in the bottle. If anything, people are going to use more agents every day. But in order for that to happen, there needs to be at least three main things. One is, it needs to be extremely easy to build, and we're going to talk about how you can easily get this agents build out. It needs to be trustworthy, that goes back into this idea of reliable and repeatable outcomes, and also needs to be manageable, so you can get to scale eventually. And the scale is not only about running a million agents in an hour, it's about making sure that you're building this building blocks that can be reused. So you're having your custom tools to be reused across many use cases, or even your agents themselves. And that's very exciting about this course. I find that having that discipline process, building your multi-agent system, and then thinking through what are the observability mechanisms, so that you can figure out when it's working, when it's not working. And then also to drive that disciplined process to repeatedly, you know, quickly identify the errors, and then fix it with some of the tools, be it prompting, fine-tuning, or something else. That is probably a more reliable recipe than most people realize, which is why with this set of knowledge, it's actually not easy, but probably easier than is widely appreciated to build pretty reliable multi-agent systems. Now there's a lot of buzz, a lot of hype about agentic AI, and about multi-agent systems. And with all this buzz and noise, what we wanted to do was get together to create a definitive course on how to use CrewAI to build multi-agent systems, so that when you'll finish this course, you understand what are the core building blocks, what are the mental frameworks, how to put them together, and then also how to go all the way from prototyping to taking your multi-agent system to scale. And this will be a very powerful skill that I'm confident will let you build a lot more stuff than was possible before. And having known Joe for a while, I can promise that this course will also be a lot of fun, and I'm excited to see what you build with this. So there's a lot to cover, let's go on to the next video and get started.