Automate anything with Ai Agents: 2025 guide

In this video, I wanted to go over how you guys can automate pretty much any process you like with AI agents in 2025. A lot of you guys probably know AI agents are going to be the big thing in 2025, and it's really, really important that you learn how to leverage them and how to actually use them effectively. So in this video, I'm going to be giving you 3 different agents you can make to automate some common tasks. It's going to be one agent that can analyze all of your PDFs and then answer questions based on the PDFs, one agent that can generate unique social media content, and then one agent that can manage your whole entire email inbox, categorize the emails you receive, and then draft responses for those emails. As always, all the agents that I make in this video will be available for completely free in the description. There'll be a little section that says free templates. You can click on the link, get all the templates for free. By the way, if you don't know who I am, my name is Rory Ridgers and I'm an AI enthusiast, and I help business owners to integrate AI systems into their business. So if that's something you might be interested in, then I've left a link to my website in the description and you can go ahead and book in a free consultation with me and we can talk about some ways to integrate AI your business. Let's not waste any more time, let's get into the video. Okay, so in this video we're going to be using Make.com and Relevance AI. You can see on Relevance AI here, these are the agents we're going to be making: the PDF query agent, the blog post generator, and the email sorter agent. So for these agents, we only have 4 tools, which is quite nice. We have the email sorting and sending tool, the PDF converter tool, and then the blog writer tool, as well as the email blog post tool. So obviously the email sorting and sending tool is going to be for the email agent. PDF converter tool is going to be for the PDF agent. And then the email blog post and blog writer tool is going to be for the blog writing agent. So as there's only 4, I can show you inside all of them. We're going to start off with probably the most simple, which is the PDF converter tool. Let's hit edit. We can look at the inputs here, and the only things we have are query about the file and then the uploaded file. So obviously this is where the uploaded PDF would go. And this is where the query about the uploaded PDF would go. So we have those two inputs there, and then the first step is literally just converting a file to text. So we just put the PDF URL from here into this step, and this will just convert that PDF into plain text so the LLMs can read it. And then we just have an LLM step, and we've just given it a really, really simple prompt that says, I want you to look at this data from an uploaded file and then try to find what the user is looking for from that file's content. I want you to display your findings in an appealing and easy-to-read way. And then we've given it the file content here, which is just the output from this step, and then the user's query, which is just this input here. So really, really simple. What I'm going to do is I'm going to test it out really, really quickly. So what I did is I found this pretty long PDF on LLMs, so everything about LLMs. I'm pretty sure it's about 50 pages long. So we've just uploaded that, and then we're going to give it a simple query like What is an LLM? And then you'll actually be able to see this tool working. You'll be able to see the file being broken down into text and then the LLM looking through that text and trying to find, in this case, what is an LLM. So depending on how big the file is depends on how long this step takes. Obviously, the bigger PDFs are going to take a little bit longer. Smaller ones are going to be pretty quick. So we're just going to leave it for a second and let it convert it into text. Okay, so it's worked, and then the LLM step worked pretty much straight away. So this is what it's come out with. What is a large language model? Large language model is a type of artificial intelligence model designed to understand, generate, and manipulate human language. And then key features, so architecture, training, capabilities, applications, challenges, conclusion. So perfect. It's obviously gone through that PDF, looked at what our question was, which is what is an LLM, and then given us the appropriate answer. So that's a really, really simple tool. You can hit save on that. And then come out of it. Now I can show you the PDF agent because that is the only tool that goes into the PDF agent. So PDF query agent, let's have a look inside it. Let's hit edit. See the agent name, simply PDF query agent. Agent description, this agent can take PDFs, answer questions on the content. So then the tools, we've given it the PDF converter tool to do that. You just click add tool and then select the tool that you like. Make sure to turn this from— by default will be approval required. You've got to set it to auto run so the agent automatically use the tool. The core instructions, really simple in here, you can see it's quite short. So we just said, role: your job is to extract data from PDFs from a user and answer questions based on the content. The task: your job is to take an uploaded PDF from a user, use a PDF converter tool to extract the data, and answer their questions on the content. Here is a breakdown of the tool. And then I've just given it a really, really simple breakdown of how the tool works. And the reason why this prompt is so short is because pretty much all this agent is doing is just taking the bits of data that the user inputs to it and then just inputting that straight into a tool. And then when the tool outputs something, it just outputs that back to the user. So the agent is kind of just being a middleman here. It's not really doing too much as there's only one tool that it's using. But anyway, giving it a breakdown of the tool, I've said this tool allows you to upload a PDF and then ask questions based on its content. So I said what the tool does, given that the inputs, the user inputs are query and file URL. I want you to put the user's uploaded PDF into the file URL and then their query on the file into the query input. So really, really simple. It's a very simple agent. All it's doing is just taking the user's input, inputting that into the tool, and then taking the output from the tool and outputting that back to the user. So really, really simple. Let's hit save changes and then we can test this out really quick. So you have to make sure to ask the question first and then upload the file for some reason. I'll show you why when I upload the file, it'll make sense. But we're going to ask it, we're going to say, how does LLM training work? And then we upload the file. And as you can see, it adds the link here. So you're better off just asking the question first and then uploading the file, because if you do it the other way around, you can mess up the link, which means the agent won't be able to use it at all. So just do it that way around, ask the question, and then put the file in. You can see here we have the link. PDF query agent's working, one step performed in the background. So it's using the PDF converter tool. Which is good. You can see these are the inputs. So the query, how does LLM training work, and the uploaded file is the link to the PDF. So that's absolutely perfect. Just give this a second to run. Okay, so it's used the PDF converter tool and it's just outputted this to us. So it's used the tool. As you can see, this is the output. We can have a look at it here. So large language models are trained using the combination of advanced techniques and methodologies that allow them to understand and generate human-like text. And then we've got the pre-training phase, training method, architecture, fine-tuning phase, training techniques, evaluation and benchmarking, challenges in training. So it's obviously gone through that pretty long PDF that I uploaded to it, and it's given us all the information that we want from it, which is absolutely perfect. And you can do this with any PDF you like, ask any questions. So a really, really simple but useful agent here. So now we're going to look at the blog post generating agent. This one has two tools. It has the blog writer tool and then the email blog post tool. And then we also use a Make.com scenario to trigger this agent. So I'll show you that as well. So first of all, I'm going to show you inside the blog writer tool. So basically what this agent does is it takes a transcript from a new YouTube video and then converts that into a blog and then emails it over to you so that then you can post it. You can obviously do this with LinkedIn posts or X posts or any text-based posts like that. But I just thought blogs are longer than all of them, so it showcases it a little bit better on what it can do. So the two inputs are going to be thumbnail image and YouTube transcript. I'll show you how we get that. That's from the Make.com scenario that triggers the agent, so don't worry about that. But thumbnail image, YouTube transcript, and then what we do is we have an LLM here and we've just prompted it saying, you're a skilled content writer. Your task is to transform a YouTube transcript into a detailed SEO-rich blog post. Follow these guidelines. And then as you can see here, I've just given it some guidelines. Tone and style, structure, SEO optimization. You can change all of this to your liking. I've also mentioned about image placement. So place the image within the content at the top underneath the title, ensuring it aligns with the topic being discussed. Then I've just given it the transcript here and then the image link as well. And then notes, always include the image at the start of the text. Make the image the same width of the text. And then for me, I've said this is the link to the website in case you need to include it when it says something like visit my website or booking a call. And then I'll put a link to my website because I've based this off of my YouTube channel. So it's going to be generating blogs when I post a new YouTube video. So for me, in a lot of my videos, I'll say like visit my website or booking a call. So I've just put in the link to my website where people can do that in case they need to include that in the blog. But obviously you change this to whatever your website is. And then what we do is we use another LLM step because we want to generate a nice AI image in the middle of this blog. So what we do is to say, I want you to take this blog post that was crafted based off YouTube video transcript and find an area where you feel like an image would be helpful. I then want you to output a prompt for an AI image generator for this image. And then I've just said to try to avoid infographics as the image generator has a hard time creating text and graphics. So I want actual images rather than infographics because the image generator struggles with text and stuff a lot of the time. So I've given it the blog post there, which is just the output from this step. And then what we do is we use a text-to-image generation. I think this one, yeah, is Flux. And then we literally just put the output from this step, which should be a prompt for an image that can go in the middle of our blog post. So we just put the output of this step into here, which should be a nice prompt. And then we use another LLM step to take the blog post and add in the image where you see it as appropriate. I want you to rewrite the blog post keeping the text exactly the same but just adding the image in where you feel it makes the most sense. So basically what's happening here is we're just giving it the full blog post with no AI image, then we're giving it the AI image, and then we're also giving it the prompt for the AI image so it knows what the image is. And then we're basically just asking it to rewrite the entire blog including the AI image wherever it feels like it makes the most sense. So it might seem a little complicated with a lot of steps, but it's actually quite simple when we run it through. I'm quickly going to show you the other tool before I show you the Make.com scenario, and that is the email blog post tool. This is really, really simple. All it is, it's just the blog post, and then we set up a Gmail step where I'm emailing it to myself in this case. So really, really simple. The reason why I didn't add this to the end of the blog post generating tool is because I wanted the blog post to go through the AI agent first, just so the AI agent can make sure that it's all correct and it's actually there and stuff. And then it can use the email blog post tool so it can email it to me when it ensures that the blog post is actually looking all right. So that's why I've done it that way. But yeah, it's a really, really simple tool. They usually just input the entire blog post and then it gets emailed over to me. So this is the make.com scenario that is triggering our AI agent to make these blog posts. So the first step is going to be YouTube watch videos in a channel, and this is just hooked up to my YouTube channel. I put my channel ID here and then publish before, publish after. You can obviously set this to whatever you want. And then the limit I've just put to 1, so it's only going to return 1 YouTube video. And then what we've done here is we're setting the variable for the video link because for some reason this step before, it only gives you the video ID rather than the video link. And we need the video link for the next step, which I'll show you. But it's all right because every YouTube video link just looks like this: https:// www.youtube.com/watch?v= and then the video ID. So we just type in this manually and then put the video ID here, and then this will set the variable as this with this variable added on. So it'll set that as a new variable that is the video link. That'll make a little bit more sense when we actually run this tool through, don't worry. And then we use this Dumpling AI Get YouTube Shown script. All we do is literally just put in the output from the variable before. So as you can see, I've obviously run this tool before, and as you can see, the YouTube video link with the video ID at the end. So this is actually the valid YouTube video link. So we put that in there. Include timestamps, it's up to you. And then timestamps to combine, whatever you like. Preferred language, for me, obviously it's English. And then we use a Relevance AI start a new conversation node on make.com. Set the role to user, and then we're sending this over to the AI agent. So we say, The thumbnail image link, you take that from these variables in the watch videos in a channel. So you can see the URL here and the high resolution. And then the video transcript, it's just going to be the output of this Dumpling AI step. And then we put in our agent ID. And if you don't know how to get that, I'll show you how to do that. It's just going to make sure it's actually talking to the right agent on Relevance AI. So that's the Make.com scenario that triggers it. Now let me show you inside the actual agent. So it is the blog post generator agent. Let's hit edit. And then to get your agent ID to be able to connect it to this node on make.com, all you do is you just go to API, go down to request body, and then you can see your agent ID here. And you can just copy and paste that in. Really, really simple. And it just means that make.com will be able to talk directly to your agents on Relevance AI. So we have the tools hooked up on this agent. We have the blog writer tool and then the email blog post tool. Again, make sure to set these to auto run. And the core instructions, quite short again. I've said the role: you're a helpful assistant that can generate blog posts based off YouTube transcript and then email it to me. Task: you will be triggered by an API and you will receive a thumbnail image URL and a YouTube video transcript. You only have two tools, which is the blog writer tool and the email blog post tool. I will explain to you how to use them now. And then I've just given it a breakdown of how to use both of these tools, so it's really quite simple. The Blog Writer tool, it literally just inputs the thumbnail image into the thumbnail image input, and then the YouTube transcript into the YouTube transcript input. And then the Email Blog Post tool, really, really simple. I've made sure that it uses it after it's used the Blog Writer tool, and it just puts the output of this into this input here, and then that will email the blog post over to me. You can see for the notes I've just said always use a Blog Writer tool first and then use the Email Blog Post tool. So really, really simple. You can pause the video and read through it if you like. But I'm just gonna hit Save Changes and then test this agent out. Okay, so we're going to run this Make.com scenario. Obviously this would be set to automatic, but for you guys I'm just going to manually run it. You can see it's got my video, it's got the build an AI agent that can scrape anything, so one of my more recent videos. And then it's set the variable of that video to this, which is obviously the link to that video. So that's perfect. Now we've got the valid video link, and then we get the YouTube transcript. You can see here This is the entire transcript. Send that over to our Relevance AI agent, which should have been triggered here. So yeah, you can see this is what it got sent through, a massive transcript. Just to make sure, we can click on the thumbnail image link, which is perfect. That is the one that we're looking for. So then we can scroll down, and now this does take a second because it's having to go through so much text here and having to generate quite a long piece of text as well. So we are just going to give this a minute and we'll come back when it's done. Okay, so as you can see, it has now used the Blog Writer tool. We're just waiting for it to email it over to me. You see, this is the answer, and it might look weird to you, but it's generated it all in HTML. And the reason it's done that is because when it emails it over to me, you can actually email stuff in HTML and it will render it, which is really cool. So this way we can add in, you know, the H3s, the H1s, things like that, and then also the images So when it gets emailed over to me, it will look normal. It doesn't look normal here. Okay, so you can see it's used the Blog Writer tool and the Email Blog Post tool, and it says the blog post has been successfully created and emailed to you. And then you can see, if we look at our emails, you can see this photo is quite low resolution. That's just while we're previewing it through the email, so don't worry about that. You can see this is the blog post that's generated, which is really, really cool. Create an AI agent for market research. What is an AI agent? The tools you'll need. Website scraper, Instagram scraper, LinkedIn scraper, Trustpilot scraper, lead researcher tool. And then this is the image it generated and did a decent job, but it could do with a little bit of better prompting. You can see it's tried to do the Instagram scraper, the LinkedIn scraper, the Trustpilot scraper, scraper tool. It seems to have messed up a little bit. I did tell it not to try and make text or an infographic, but seems like it's done it anyway. So we can always refine that prompt, but it's got an image in there which is really, really cool. And then it's given us a breakdown of each of these tools that I obviously touched on in the video: Lead Researcher tool, Website Scraper, LinkedIn Scraper, Instagram Scraper, Trustpilot Scraper, and then a bit about integrating the tools into your agent, testing the agent, and then a conclusion. And then you can see here, booking a call with me. If we click on that link, it'll take you to my website. So that's really, really cool. So again, that's a relatively simple agent with two relatively simple tools. Obviously you do have to connect it to a Make.com scenario, which might be a little bit tricky for some of you guys if you've never used Make.com. Obviously the templates will be available for completely free, so you can download it and then mess around with it and try and get your head around it. Now we can look at the last tool and the last agent, which is the email sorting and sending tool. So for this, the inputs are recipient name, email body, recipient email address, message ID, and subject line. Again, this agent is going to be triggered by a Make.com scenario, so I'll show you that in a second. It's a really, really simple scenario, but we get all of these different data points here from the Make.com scenario. And then we simply have an LLM step here. And what we want to do first is look at the email that's come in and then sort that into the right label category. So all my emails, I've made some categories here. I've done business, socials, and urgent. Obviously this could be whatever you want it to be. You just have to change the prompting slightly of this LLM step here. And what I've done is told it to look at the email body from the email that we received and then try and output the label ID for the label that it thinks it should go into. So I've said The label categories are urgent— this would be anything that requires immediate attention, like a failure in a system or an angry customer. Socials— this would be anything that's related to my social life, like a friend asking to meet up sometime or go grab a coffee, etc. And business— this would be anything business-related, like a customer inquiry or a message from a member of the team about a project. So output only the label ID and nothing else, no summary or explanation. And then I've given it the label IDs here. So social, this is the label ID. Urgent, this is the label ID. Business, this is the label ID. And I've given it the email body. So should output the label ID of the label it thinks this email should go into. And then we have an API step here, and this connects to another Make.com scenario, which I'll show you now. So this is the Make.com scenario. We have a really simple webhook set up here. So if you've never used a webhook before, all you do is you place it down, give it a name, you'll get a URL here, and then you're going to put that into the URL section here. Set the method to POST because you want to be sending data over, and then you'll have to hit Redetermine Data Structure, then hit Run, and then they should pair, and then you should be all good to go. And you can send data from Relevance AI using this API step over to Make.com using a webhook. So for the headers, we're going to go Content-Type application/json, and then the body is going to be JSON object. And the reason we do this is because we're sending over two separate variables here, and that is the message ID and the label and we get that from the inputs here. So the message ID and then the label answer from this LLM step, and we send it over as a JSON object because this way we get this as two separate variables that we can use in the Make.com scenario. So all we have here after this is the webhook and then just a Gmail modify email label. So what we do is, as you can see from the webhook, we get two separate variables, which is the label and the message ID. Message ID goes into here. So this is going to be the message ID of the email we receive. And we want to change the label of that message that we received, and that is going to be the answer from the LLM step, which obviously is wrong here, but it should be the label ID. And if you don't know how to get the label IDs, because I didn't before making this video, all you do is you select the label here and then click Map, and it will show the label ID. So that's what I did for all of the parts here in the prompt. I just did this method where you click map on the label. But for this, we are going to map it and we're going to put the label, which should be the label ID from that LLM step. And then labels to remove, you can do this if you want, but for me, they aren't already in a label, so I'm not going to be removing any labels. And then that's the Make.com scenario. It's really, really simple. All it is is just a webhook and then a Gmail step where we modify the label of the email that we received. So we can come back over to the Relevance AI tool here. We have another LLM step, and what this is going to do is it's actually going to draft a response to the email that we receive. So I've said to it, I want you to draft an email response to this email. Act as friendly— act as a friendly personal assistant for me. Make sure to include the recipient's name in the response and sign it off with thanks. And I've also said, if you can't find the recipient's name, then don't mention it and just put hey instead of a placeholder. Because sometimes if the LLM step can't find the user's name when it comes to Gmail, it will put like placeholder name in brackets, which obviously looks rubbish. We don't want that. And one of the things we need to make sure to do up here is set the recipient name to not required, set it to optional, because sometimes on Gmail people don't have their names on there, they just have their email address. So we need to make sure to set that to optional and make sure to tell the LLM that if it doesn't get a name through, just say, hey, don't put the placeholder. So then we give it the recipient's name variable. The email body variable. And I've said don't output our subject line, only the email body. And then we use a Send Gmail Email step. The to email is going to be the email that we received at the beginning, so that should be the email address of the person that sent us the email. And then the subject, I've just put re, and then the subject line that they sent us. And then the email body is going to be the answer from this LLM step here. And an important part here, go to settings if you want to make sure this is only drafted and not actually sent. Tick this box which says draft, and then it won't actually send the email, it'll just put it in your drafts. Okay, so I'm going to show you how this tool actually works when we use it inside the agent, because we need to get a legitimate email through, because we need the actual message ID to be able to modify the labels and stuff. So we're just going to actually send an email to my email address, and then that will trigger the agent, which will trigger this tool, and then we can see how that bit works. So let's hit save, come out of it, and then look at the agent, which is the email sorter agent. So if we hit edit, you can see the agent name is Email Sorter. Agent description says you can sort emails into labels and draft email responses. The tools is obviously going to be that one we're just looking at. And the core instructions, again, this is really, really short. The role: you're a helpful email assistant that can categorize emails and draft responses. You will be triggered when you receive a new email and you will receive multiple data points. I want you to put these data points into the correct inputs in the email sorting and sending tool. And then I've just given it a breakdown of the tool. This tool can categorize and draft email responses. The inputs are recipient name, email body, email, message ID, subject line. And I've just said really, really simple, put the respective data points into the correct input. So the sender's email address goes into the email input, the email content goes into the email body input, et cetera, et cetera. I'm also gonna show you the Make.com scenario, which triggers this agent, which is again, really, really simple. All we have is a Gmail, watch email SERP, and this is just hooked up to my Gmail account and I've just gone all mail. The criteria is all emails. You can do only read or only unread. And then mark emails as read when they're fetched, you can do if you like. Maximum number of results, I've only put 1. Obviously you can tailor this to however you want it. And then again, we have a Relevance AI node here and we're connecting this to our Relevance AI agent. The role's user, the content is going to be the date, subject, the content, sender name, sender email, and the message ID. And we obviously get all these variables from the Watch Emails node before. And then again, we just put our agent ID for this new agent in there. Okay, so we're gonna test out this Make.com scenario. I've just drafted sort of a basic email to myself, something about a project deadline. I've said project deadline for the subject line. Hey, wanted to reach out to say this project is due soon. Are you up for a call to talk about it? So maybe an email from a member of a team or a client or something like that. This should go into the business label. We're gonna hit send. So I've sent that to myself and then we're gonna run this Make.com scenario, which should trigger and pick up, hey, once, yeah, once to reach out and say this project is due soon. Are you up for a call to talk about it? Perfect. Send that off to our AI agent here. One step performed in the background. Use the email sorting and sending tool. We look at the inputs and outputs for that. You can see Rory@autonomously-ai.com. See email address. Email body is what we sent. Message ID is this. Recipient name, Rory Ridgers. Subject line, project deadline. Looks like the agent has inputted the variables pretty well into the tool, so that's good. Okay, then as you can see, here's the email from Rory Ridgers has successfully been categorized and a response has been drafted. This is a response, but we can have a look just on my emails. So the draft RE project deadline, hey Rory, thanks for reaching out. I'd be happy to set up a call to discuss the project. Please let me know your availability and I'll do my best to accommodate. And then if we look into the business label, just had to refresh it. There you can see project deadline. Hey, want to reach out and say the project's due soon. Are you up for a call to talk about it? So that's obviously the email I sent. It's categorized that into the business label, which is really, really cool. So now those 3 agents work perfectly. None of them are too complicated, which is nice, but they all solve really, really common issues and are really, really helpful for business owners or just individuals. So I hope you enjoyed this video and should now all be set up to create your AI agents in 2025 that can automate pretty much anything you want and solve all of your problems. As I said at the start of the video, all the templates I made for these agents will be available for completely free in the description. You should see the link that says free templates. You can go ahead and grab them. And also, if you were looking to work with me, as I said, I've left a link to my website. You can book in a free call and we can talk about some ways integrate AI systems like this or something more complex into your business. So that's it for this video, but I'll have loads more great videos coming out on AI agents in 2025, so make sure to subscribe and follow along the journey.