Auto-generated transcript Do you ever wake up one morning and just think to yourself, hey, I want to automate my entire customer support department? Because I did. And I came up with a plan. Let me show you. Basically, we're going to automate our entire support department using AI agents. And if we do a good enough job, we can have agents that are as good as all of these, right? By the way, I know this is fake. I know this is a meme. Do not fucking call me out in the comments. But yeah, we're going to automate our entire support department. So let's begin. oh yeah before we do anything else let me just show you an example of this right so i have this url over here let me just open it up so this is an ai support agent built by salesforce i picked salesforce because i think most even non-technical people know about salesforce right so this video will be a lot more relatable to you guys but basically you have an agent like this you can ask it a bunch of questions like i don't know how do i um i don't know how salesforce works or what it does so i'm not really good at coming up with these questions but let's just say how do i reset my password right i could ask it a question it's going to come up with an answer and it's an ai agent so it actually understands my questions and it's not just gonna recommend me a bunch of articles it's going to give me step-by-step instructions on how to solve my problem right like look over here right it gives me a step-by-step plan on how to solve this problem how do i go about resetting my password right and it's referencing all this information from over here right from say this article this is what we're trying to build right instead of having to go through like support articles and just read through long articles on how to solve some problem we can have an agent that just lets our customers ask a bunch of questions in natural language and just get an answer right so that's what we want to build in this video okay now that i've gone through that let me show you my really awesome really smart chart over here this is what we're going to achieve, right? You've got a bunch of headcount in your company, or if you're just doing this yourself, you're spending way too much time on your support department. We're going to get rid of all of that, right? We're going to make it so you don't need all this employees and time and effort for your support department. And we're going to replace all of that with AI so that AI does all of it for us so that we don't need to do any of this stuff. And this right here is the step-by-step you know plan we're going to follow so let me go through this entire thing for you the very first thing you need to do is you need to have some sort of a knowledge base right a knowledge base is basically just a bunch of like articles and help information support information that you've got on your website that customers or users can read through and understand how to solve common problems in your product right so a very simple example is MailChimp over here so this So this is the MailChimp Help Center I could go over here and read any of the articles and solve my problems right So let say I a new user and I want to import contacts to MailChimp I go over here and then I would just read through this whole really long article and maybe watch these videos as well. So I understand how MailChimp works and how I can import contacts in it, right? This works. Every single company on earth is doing this. And to their credit, most of them are pretty useful, like the docs they write, but it's just so much work, right? As a user of MailChimp, I have to go read this entire thing. And it's just such a, like, it's just so much effort, right? I mean, compared to something like Salesforce, where I can just ask the AI agent a question, and it'll just give me specific step-by-step instructions. MailChimp needs me to read a really long article to solve the same problem, right? But that kind of knowledge base, those articles are a precursor to the AI agent that we're going to build. So you do need that, right? You do need some sort of a knowledge base you can use. Just a big chunk of articles that describe all the issues and how to solve them, right? Now, the good news is you most probably already have these, right? Most companies do, right? You've got some documentation or some sort of support center for your product where you've explained all these things in detail, right? So you probably have these already, but if you don't, go write them right now. You can use AI to help you write these just as long as the information in there is accurate, but you need a knowledge base. And once you have that, move on to step two, which is to develop a natural sounding AI, right? This isn't all that technical. This literally just comes down to the system prompt you're using, right? You would just tell the AI to sound like a human basically, and basically just make it sound like a support agent at your company, right? You don't want it to sound like some generic chatbot because users don't like that, right? Nobody likes to talk to some generic robotic chatbot. So you want to make it sound natural and you want to make it sound like someone at your company, right? Now, the specific instructions for this are going to be unique to you, to your company, to your use case. But some really common things that you want to avoid are emojis. AI loves emojis, especially chat gpt i haven't seen it used as much in cloud and gemini but chat gpt loves emojis so no emojis make sure you're avoiding them no m dashes every single ai llm is just spamming m dashes everywhere by default so it's a really common way to tell that you're using an ai and we don't want that so no m dashes finally no common llm phrases right no fucking you know you're absolutely right or it's not X, it's Y, right? It not you know like you wanna avoid all of these common LLM LLM phrases as much as possible right because they just give away the fact that you using an ai not only that but they just so low quality right because the internet has just been flooded with all of these llm output llm writing that everybody's seen stuff like this and we've become desensitized to it we don't really care about these things anymore so make sure you're avoiding all of these things i'll try to put a prompt in the description because I do have a really detailed AI prompt for this to just remove all of the LLM phrases and make your writing sound like a human. I've got a really detailed AI prompt for that. I'll put it in the description if I remember to. And you can use that for like your starting point over here. I'll try to put it in the description and you can go use that as your like kind of like starting point and just tweak that as per your use case, right? Now, step three is to build a rag pipelines let me explain what rag means rag basically means retrieval augmented generation it's not as difficult as it might sound you literally just like all it literally means is retrieval you're gonna retrieve documents and training materials for the ai that's what we do in step one in the knowledge base over here right augmented you're gonna augment the ai agent with those documents that you retrieved, right? So that the AI knows things about your specific product, the issues that may come up with it and how to solve them, right? And finally, generation. The LLM is going to generate some text output, some answers that the users are going to read and go fix their issues with it, right? We want to use RAG to train our AI agent on our knowledge base. And this does get a bit technical, but I'm sure you can figure it out, right? All you really need to do is have a knowledge base, detailed answers for every single problem in your product and you just feed all of that into AI using something like a vector database and embeddings and chunking. Google what those terms mean because I can't really go through how to set up a RAG pipeline in just this video. It would just become like a two hour long video. But yeah, you need to build some sort of a RAG system so that you can train your AI agent on your knowledge base. And once you've done that, you're pretty much done. At that point, all you need to do is just deploy the bloody thing on your website, right? And here's the thing. Here's the neat part. I'm sure I have the neat part somewhere over here. Yeah, there we go. Customer support is extremely cheap to automate, all right? This isn't some, like, this isn't a task that requires frontier models or frontier intelligence or really high reasoning, right? You could literally use cheaper models like Gamma 4 or Coin 3.6. If you don't know what these are, these are basically just really cheap open source models that you can deploy on your hardware they don't cost a lot to run you could literally have millions of tokens spent on these models and still the total cost would just be a few dollars right these are really small really cheap models and they perfect for a task like customer support because customer support is just so easy and cheap to automate right You don need some really fancy really frontier intelligence model. You could easily make do with these cheaper ones. So that's a bit of a good news. And some additional add-ons you could add to this whole thing, because like we already built the AI agent, right? But if you want to make this even better, here's some ideas, right? Number one, you could organize your customer support tickets using AI, right? You would have an agent that reads the entire support ticket, figures out what it's for, and then just categorizes it like, I don't know, account issues or payment issues or product issues, whatever, right? Just so it's easier to manually review later on. And also, you could analyze customer satisfaction from your support tickets. You could have an agent that reads the entire support ticket and just gives you a score out of like 0 to 10 or 0 to 100 on how satisfied the customer is with your support team, right? Or number three, you could find out your customer's most common pain points by simply reading the entire support ticket with AI, right? So if you notice that, I don't know, there's like 100 support tickets in your database and all of them point to the same problem with your product, that's probably a very common pain point that you should probably fix in your product right and an ai agent analyzing your support tickets can help you find something like that so that's the whole video that's how you create a support agent like salesforce and deploy it on your website this does get very technical and i can't go through all the technical bits like the rag and the deployment in this video because it would just become like a four hour long video So you're going to have to Google some information over here. But yeah, this is the overview. This is the big picture of how you can develop something like this, right? And if you want more guidance on this, I do have a resource over here, which is this GitHub gist. And if you go to that URL, you'll open up this system prompt that I've written for you. This is a really long, detailed prompt that's going to help you develop a sales, a support agent, just like Salesforce, but for your own custom product, right? Just copy this entire prompt, paste it into some LLM like Cloud or ChatGibby, and just let it help you develop this entire thing. I've put a lot of guidance into this prompt, and I've made it so that specifically non-technical people who aren't exactly programmers and coders can actually use this prompt to develop a support agent for their company, right? So even if you don't know how all of this technology works, the prompt is literally written for people like you. So just copy this, paste it into chat, jibble your plot, and just let it guide you on how to set all of these things up. I'll put the link for this in the description as well, along with all the other stuff. Go check that out.