CrewAI Tutorial: Automate your Life with AI Agents

Key Takeaways

hi guys in this video I will introduce you to the amazing tool crew AI which will allow you to create an army of little AI assistants that can solve almost every task you can think of I will first introduce you to crew Ai and its capabilities then show you how crew AI works and lastly walk you through setting up your own crew and running it and at the end of the video we will look at the results together so you've got this big project and think maybe chat GPT can help me out but soon you find out that chat GPT struggles with the really tough stuff it's like Chad GPT is a bit behind the times not knowing what's happening in the world right now and multitasking forget about it you have to piece everything together yourself and the really smart tasks higher level thinking and creativity aren't exactly chat gpt's strong suits plus if you're looking at a big project chat GPT might not keep up especially if you're asking for something really long since chat GPT can't really handle things much longer than 500 Words let's be real chaty PT is cool and all but when it comes to more complex tasks you're kind of on your own since for the complex stuff fine-tuning your prompts for chat GPT to make it do what you want costs more time than just doing it yourself but what if there was something better something that took chat GPT and supercharged it to work on those tricky projects that's where crew AI comes in what's crew AI all about first of all shout out to Joel MOA who developed and keeps developing crew aai as an open source project on GitHub think of crew aai as that dream team you've always wanted this isn't just another chatbot it's like assembling your dream team of assistance each equipped with chat GPT and access to all the stuff on the web from today using their own unique expertise and chat GPT fine-tuning skills for tackling those big brain busting projects so let's look at some of chat gpts big limitations with crew AI you will basically get rid of most of these limitations while barely having to do any work yourself crew AI takes chat GPT to the next level giving it access to the latest info from the web and customizing it for your project whether it's putting together a travel itinerary coding a video game putting together investment advice or preparing your next Instagram posts crew aai has got you covered imagine having a team of experts each making chat GPT Smarter with up to the minute info from the web this isn't just about using chat GPT it's about making it better tailoring it with detailed expert knowledge to find solutions that really hit the mark But putting together a team like this by yourself that sounds like a lot of work and it will probably cost you more money than you are able to spend right now that's where crew aai steps up it brings together a virtual team of specialized AI agents each ready to boost chat GPT with their own expertise and the latest data from the web so we just got hyped about crew AI right it's like hearing you've got a ticket to the coolest show in town but then it hits you hold on do I need to be some kind of coding wizard to use this I get it that's what I first thought too you're probably thinking this sounds awesome but also like it might require a computer science degree or something and let's be honest not all of us want to spend hours upon hours learning to code from scratch right we've got ideas projects that we're passionate about and we don't want to get bogged down in the technical weeds you see crew aai is designed with us in mind you don't need to be a programming Guru to make the most out of crew aai if you've got a handle on the basics of python and can craft a clever agent description in plain English you're golden I would say if you've ever read and run a basic python script and know how to make chat GPT do what you want it to you will be fine the crew AI agents will be doing the heavy lifting crew AI is for those who have big ideas and think I wish I could afford to hire a team to help me execute on my ideas so let's delve into some practical examples here are some example Crews that have been set up already by Jon the developer of crew AI all of the examples can be found on Jan's GitHub which I will link in the description the stock analyst crew this particular crew specializes in aggregating and interpreting stock market data to provide actionable insights by harnessing diverse sources of financial reports and market trends it generates conscise recommendations on buying selling or holding stocks what distinguishes this crew is its ability to process vast amounts of data efficiently delivering comprehensive analyses in a fraction of the time it would take a human analyst the trip planner crew imagine planning a vacation tailored precisely to your preferences without the tedious research the trip planner crew automates this process offering personalized itineraries based on your interests and budget this crew evaluates various factors including destinations activities and logistical details to curate an experience that aligns with your expectations it exemplifies how crew AI can transform the planning process into a streamlined enjoyable task the game builder crew for those interested in game development the game builder crew presents an intriguing application of crew AI by simply describing your vision for a game this crew translates your ideas into functional code Beyond initial development it iteratively refines and enhances the game based on testing and feedback showcasing the Adaptive and creative capabilities of crew aai in a development context by examining these examples we gain a deeper appreciation for crew ai's role in facilitating complex problem solving across various domains it's important to recognize that these scenarios are within reach for users with foundational python skills and the ability to articulate their project goals clearly so let's actually look at how crew AI works and get an actual crew running together in the interest of time I will not get into depend Tendencies and environment set up here there are plenty of resources online both on Google and YouTube so I don't want to waste time on it if you are having trouble with it leave me a comment and I can make a video on that in the future the first thing we want to do is think of a way to assemble our crew as we would in real life for that you first need to decide on the type of crew you want to assemble or what crew AI calls the process currently crew AI supports two types of processes sequential task execution and hierarchical task execution but more complex processes like consensual and autonomous are being worked on an example of a sequential process would be a researcher and a writer where the researcher hands off the output of his research to the writer who then writes a blog post or newsletter with the research provided by the researcher a hierarchical process simply means building a crew with a chain of command so basically one of your agents becomes the manager of all the other agents and can delegate tasks to his subordinates and analyze their results and put it all together to get you started as quickly as possible I will just walk you through one of the easiest setups so you can already get experimenting with crew AI if you want other processes or crew setups covered in the future drop me a comment so I'm going to show you how to set up a crew with a simple example of a crew that is writing a job description for a client company imagine you're a recruiting company and you have a client that wants a job description written for a particular position and you can define a crew that does this for all of your clients so for each overarching goal that you define the first thing you have to do is Define the subtasks necessary to achieve that goal and Define the agents necessary to execute these tasks let's say the overarching goal is writing a job description the subtasks might be research the company culture do an industry analysis for that company research the role requirements draft the job posting and then review and edit the job posting you might have a crew of three people so you can imagine them as actual people that you can hire that have access to chat GPT and Google and so on on and can do this job for you you might Define a research agent that takes over the first three tasks so research company culture industry analysis and research requirements that agent has access to the web and Google search so they can just look at the company website for example and also search Google the research agent can then hand off the result of their research to the writer agent and the writer agent is tasked with drafting the job posting they again have access to the web Google and they can read files because we want to provide an example job posting for them to build that draft off of and then lastly the writer agent will hand off the draft to the review agent who will then review and edit the job posting with again the same tools this is an example of a sequential process so each agent will produce some output that will then be handed off to the next agent and the final agent will output the final result which is then the job posting that is the overarching goal crew AI also allows other types of processes like hierarchical processes where you have a manager that can give tasks to different agents working in parallel and then review all of their work and put the final answer together for you but I'm going to stick to the sequential process for this video if you want to see an example of the more complicated but more powerful hierarchical process I can do a video on that in the future drop me a comment and I can do that for a future video okay so as I just explained we are creating a job posting for one of your clients and basically what you have to do is Define the crew that is going to perform the task for you the subtasks of the main task all the agents in the crew and how they're going to work together so it's fairly simple you just need to understand this basic file structure and be able to read some python code but really no coding is really needed you just need to be able to give instructions to your agents in English so in the main.py file you always have to have a main.py file and a agents. p file and a tasks.py file which defines the agents the tasks and the general process here you import yourn file which has the your API keys for um open Ai and serper serper is our Google search API that we're going to use to do Google searches in the beginning we Define some context for the job which is basically every company that we want to write a job posting for the company domain the hiring needs and the specific benefits for the drop posting so you might want to reuse the screw for future drop postings for different clients and every time we run the crew we can just specify these four things on the command line and it will run the job with these four answers in mind and create a different job posting for different jobs then we uh call our three agents we Define them in the agents. py file we'll get there in a second and then we define the five tasks at hand for these agents we'll also get there in in a second and then we Define the crew and pass on the agents into the crew and the tasks to be done by default the agents will work on in a sequential process but if you want to run a hierarchical process you just have to specify it when you when you initi instantiate the crew here and last you just kick off the crew I'll show you how to run this after going through the other files but it's fairly easy so first we're going to look into the agents here we defined the agents and as I said you don't really need to do a lot of coding actually you just need to copy and paste these templates and change instructions in English in written English so up here you define the tools that are necessary for the agents to execute their tasks we're going to use just three tools the web search tool which is comes out of the box or all three come out of the box in C tools which you need to install there's the web search tool there's the serer def tool which is just a Google Search tool that's what you need the serer dev API key for and the file read tool the file read tool is to read the example job description that we pass on to the crew so this is the example job description and this is just to use as a template for your job description it will ignore all the actual text but just to give an overview of what the job posting should look like so let's get back to the agents and here you define the agents that are supposed to do this job for you so as I showed before you have three agents you have the research agent the writer agent and the review agent and for each of them all you need to do is in English or Define these three agents and explain to the llm so for example imagine you were talking to C gbt and describing to CH gbt what kind of agent they're supposed to do to execute your task at hand so the first role is the research analyst and then you describe the goal for that research analyst so the goal is to analyze the company website and provide it description to extract insights on culture values and specific needs and you can also train that agent to become a certain person by providing a backstory so for example that agent is an expert in analyzing company cultures and identifying key values and needs from various sources including websites and brief descriptions so again just think of giving really good prompts to CET GPT to describe to CET GPT what kind of assistant you want CH GPT to be right now to give you the answers you want we do the same thing for for the writer agent we Define the job description writer role give it a goal use insights from research analyst to create a detailed engaging enticing job posting and a backstory skilled and craf and compelling job descriptions that resonate with the company's values and so on and for each of the agents we also need to pass on tools that it's allowed to use so in our case the web search tool and the serde tool which again is just the Google Search tool for the writer agent we also needed to give it the file read tool because it needs to be able to write to read the job description example file to base its job posting on and then lastly we have the review agent and again just in English review and editing Specialists give it a goal review the job posting for clarity engagement grammatical accuracy and alignment with the company's values and so on pass the tools it's allowed to use to the review agent and lastly a backstory that defines the type of person that you want to execute this task verbos equals 2 just means give us very detailed output when we run the crew which you're going to see later so lastly we're going to go to the tasks and the tasks again it's very easy you just need to basically copy this template and change the task names the context that you pass on to the tasks and the description of the tasks in English that's all you need to do so you define a task that you give it you give it a description to so the first task is research company culture and basically as if you were talking to jbt and giving J gbt really detailed prompts in English you're just telling it what to do so here we pass the agent supposed to do the task into the task and some context that is based on the answers uh to the prompts on the command line so the company description and the compy and the company domain because we want the agent to go online and actually research the company so it goes to the company website and actually researches uh details about the company that's what we're telling it to do here analyze the provide company website and the hiring manager's company domain at company domain uh with this description focus on understanding the company's culture values and Mission identify unique selling points and specific projects or achievements highlighted on the site compile a report summarizing these insights specifically how they can be leveraged in the job posting to check the right candidates and then you give it an expected output so a comprehensive report detailing the company's culture values and Mission along with specific selling points and so on so the output of this will then be passed on to the next agent then we Define the other tasks in the same fashion so we have the second task research row requirements giving it to an agent to execute um with the context of the hiring needs and again we just in play in English we describe what the agent is supposed to do based on the hiring manager's needs pass in the hiring needs identify key skills experiences and so on and an expected output a list of recommended skills experiences and qualities for the ideal candidate the fourth one is drafted job posting so here we give it to another agent with with the company description hiring needs and specific benefits and again in plain English describe what this agent is to is supposed to do with an expected output and the same for reviewing and editing the job posting the context provided here and then just in English tell them what to do for this particular task and the expected output in this case we also have an output file which is the actual job posting which it goes into job posting. MD which will just appear in the same folder where you run the crew and lastly the industry analysis task same thing give it some context company domain and Company description um and then describing the task in plain English and giving an expected output and the agent supposed to solve this task so now let's go ahead and run the crew on the command line so let's now run the crew and have it generated a job posting for Tesla so this is my idea let's see what files we have here so those are all the files the main file the tasks file the job description example and the Agents file so let's go ahead and run the crew so all we have to do is type python Main do. py what is the company description Tesla is an EV car manufacturer the company domain is tes.com what are the hiring needs so let's get an actual job description I think there was python generalist for the Tesla bot I think that was an actual job posting what specific benefits do you offer 60 days pay time off and a free purple giraffe and now Crea is getting started and will give us output along the way to tell us what it's doing so the crew is now done and it actually did a lot more than I expected it to do because last time I ran this I ran this from my old employer and it was very quick because there's not a lot of information on the internet to research for my crew but now it actually went on a lot of websites because I ran it for Tesla and it took about a half hour and it I think it went through hundreds of websites to research Tesla company culture and uh job types and so on but this is basically what it came up with let me show you the resulting job description so this is the job description it came up with python generalist Tesla bot job posting at Tesla we're accelerating the world's I to sustainable energy we're revolutionizing the automobile and energy Industries with our electric vehicles and Energy Products our culture is driven by a commitment to Innovation collaboration and respect for individuals as a member of our team you contribute to groundbreaking initiatives and work with some of the industry's brightest minds and then it did all of this and I didn't give it any of this information it found out where Tesla sits it found found out what a what a python generalist does it actually went to the actual career side of Tesla and looked for similar job postings as well as other job postings on uh similar positions at other companies and you know it gave me feedback all along the way so here's some responsibilities develop an enhanced python training infrastructure for efficient robust training processes again I didn't tell it to do any of this it like figured it out on its own requirements practical experience in Python and R C+ programming which it found out online experience with training Frameworks pytorch all of this stuff I didn't tell it to do and it even came up with compensation and so on so this is actually very exciting um it was much as I said much much quicker for my old employer because it's it was a smaller company uh it took about a minute back then now it took around half an hour but it actually went to hundreds and hundreds of websites so you can scroll through everything it did and it actually did a lot of stuff and you know sometimes it got some errors when looking for websites and so on but all overall very very impressive so that's it guys running this new crew took us about 30 minutes or $5 of open credits to be fair the crew went a bit overboard with researching Tesla when I ran it for my former employer it took around one or two minutes and under $1 of open AI credits just because there was just less to research I guess so you can find- tune your instructions to the crew a bit here overall I'm very impressed the crew really went out there and researched Tesla its company culture researched similar positions at other companies the EV market and so on I think the same amount of work would have taken a crew of humans a few days assuming of course they had never heard of Tesla or python or E because the crew had to learn about all of that first so you would probably spend a couple hundred on this task I think the task was quite straight forward though and I think the difference will be even more extreme for a more complex task especially when we create a hierarchical crew with more complex tasks if you want to see me do that in a future video leave me a comment if you want to get started on using crei but don't know how to get an open AI key or how to clone a GitHub repository I've got you covered check out my other videos on this channel I also have python tutorials for beginners for those that need to refresh their coding skills thanks for watching and see you next time