# 148 Learn Zapier in 30 minutes: AI Agent Automation Tutorial For Beginners (2026)

# 148 Learn Zapier in 30 minutes: AI Agent Automation Tutorial For Beginners (2026) Speaker 2: The easiest way to get started with AI agents is using Zapier's AI agent tool. Now, Zapier lets you build AI agents from a single prompt. You don't need to know how to code. You don't need to know anything about automation. And in this video, we're going to walk you through how to build your first agent from scratch. From beginning to end, how to test it and how to put it live into production. I sit down with Tom Nassr, who's been doing this for over five years, and he's going to show us a masterclass on how to build AI agents from scratch using Zapier. So enjoy the video and let's dive in. Tom, what are we building today? Speaker 1: We are going to start building with Zapier agents. Now, I know you might be thinking agents, AI agents, it's a scary new frontier, but we're going to make it really simple for you today. We're going to show off Zapier's agent builder, and I'm also going to show you how to build an automation that does some similar things, but also uses AI in the middle. So that's what we're going to focus on. Speaker 2: Awesome. And Zapier is such a powerful tool. And I think the agent feature really helps take it to the next level. So right before you jump into your screen, Tom, why don't you just give us your background and kind of tell the audience why it is that folks should listen to you? Speaker 1: Sure. My name is Tom Nassr. I'm co-founder, CEO at X-Ray. We're a workflow consultancy. We have helped hundreds of different companies from dog groomers to VCs build better operations and design the way that people work. We're a people-oriented organization. We design work for humans, and that happens to use AI and automation on the back end, but by orienting around the person and the work that the person wants to do, it really helps figure out what the right tool is for the job. So we started about five years ago. We have a handful of products of our own, and really we're one of the thought leaders in the space. Speaker 2: And you also happen to have a really awesome YouTube channel too, which of course we'll link in the description. But for folks that want to go deeper on this stuff, I'd say check out Tom's YouTube. So with that said, Tom, why don't you go ahead and share screen and we'll dive right in and look at some Zapier agents. Speaker 1: Let's do it. So here I'm inside of the agents product by Zapier. On their home screen, just click on agents, right? Assuming you logged in, assuming you have an account, chances are you're already paying for Zapier. If you're not, you could start a trial, but Zapier's free plan really doesn't give you much to play with. So you're probably going to end up putting up some dollars for Zapier. You can just go ahead and click on Create Agent, and there's a bunch of different agent templates that they have for you. Now, they're pretty good, but today we're going to focus on a new type of agent that we're going to build from scratch. You can follow along and just explain in natural language what it is you actually want to do. In our case, we're going to say, look for emails from or about Orchestrated connecting in my Gmail. And Orchestrated Connecting is a peer group that I'm a part of. Speaker 2: Okay. Speaker 1: So we're going to say, look for these emails. Then at the end of every week, send me a Slack message with a general summary of these emails and link me to them. So I can respond. Or read it. So very general prompt. Great, let's start building. So when you click, you know, the send the prompt, what Zapier agent builder is going to do is actually parse the message that you send to create a set of different triggers. So triggers and actions. So in this case, it's looking at a weekly trigger for a preferred time, and then it's going to do a Gmail search, just like you would go to Gmail and search, you know, subject or body contains, in this case, orchestrated connecting. It would do that search on your behalf inside of your Gmail and then send me a Slack message. So what I'll end up doing, just zoom out a little bit here so the UI looks a little bit better, is we're going to add a trigger and it's going to be scheduled by Zapier every week. And we're going to choose the day, Monday, how about 645. And that's it. We just added that, right? Now, I did that manually. We could also just say to the copilot, like, looks good, build it. Now, the Copilot has gotten a lot better in the last year or so. It used to be a glorified tutorial. Speaker 2: I said that on purpose, but I can't do it. Speaker 1: What did you say? Speaker 2: I was just saying, yeah, it used to be that there wasn't much that it could do. It was kind of more of just like an interactive knowledge base. Speaker 1: It would point you to the help docs all the time. It was really a pain. So now it's a lot better. So in this case, right, you see it, it's filling out instructions, it's adding actions to the instructions. One nice thing about Zapier agent is like the action that you want to perform inside of any given app is like inserted into the directions. So you really know when it's calling certain things. So I need to connect my Gmail and I need to connect my Slack account. So let's fix this setup error. I'm going to connect my Gmail. Great. Continue. This is just regular OAuth login through Google like you would normally do. If you haven't done it on your account, you'll need to do that. That's just how Zapier works. The search string, we could let the agent determine the value for this field or we could be specific about it. In this case, I'm just going to let the agent decide for now and just save. So that find email step, the agent is going to actually draft, you know, the string inside of the search that is going to search for emails about orchestrated connecting. And then I have a Slack step that's broken, so I'll connect that. I am not sure if this is going to work the way I want it to work. Speaker 2: It's basically saying that, like, hey, this Zap isn't going to work until you authenticate your Slack account, obviously. Right. Speaker 1: Exactly. Yeah. So let me just open up my one password here. To get these fancy codes. Normally you would have already done, did this. I actually logged out of everything to give people a authentic, authentic look. Obviously that's biting me in the butt right now. Speaker 2: I mean, no, I think it helps because again, it's like when you're kind of building it from scratch and walking through it step by step. I mean, the idea with this, this podcast and this channel is to, Help people, you know, give people something that they can follow step by step. So a lot of people don't know. It's like, oh, well, yeah, naturally I have to connect my Gmail and I've got to connect my Slack and I've got to give it access to all the tools that I want it to operate within. Speaker 1: That's exactly right. Uh, maybe. Of course. Oh, actually, wait, could I sign in with a password instead? Unknown Speaker: There we go. Speaker 2: There we go. Speaker 1: Just needed to trigger the right way to do it. So I'm just going to log into my X-Ray Slack account. And now I'm in. I should be in. Should have already connected. Speaker 2: Sometimes I have to do it twice too. That's happened to me before. Speaker 1: Really? Speaker 2: Weird. Speaker 1: Okay. And now I have three accounts. There we go. Hey, you know what? Everybody's dealing with a bug. All right, so here we go. So we'll save this and great. So now that is all done and we have default tools like visit, site, and web search there. I don't need a knowledge source, but you can think of a knowledge source as like the database that could be consulted if the agent needs to find additional information. Here's basic instructions. You saw me do pretty much no setup, right? Like this was, I mean, we're, what, five minutes in? Speaker 2: Right. Speaker 1: Maybe. So I'm just going to click Preview Agent here down at the bottom to just see what would happen. Speaker 2: Oh, I love that it lets you, I didn't realize that lets you preview, which I think is like kind of a, people would think that that's an insignificant feature, but it is actually very important because there's nothing worse than going through a workflow and doing all this work to get something set up. And then you go to try to test it with your real data and it breaks and you've got to go back to the drawing board and figure out what didn't work. But Zapier lets you preview it and test it before it goes live, it looks like. Speaker 1: Yeah, so this is actually, it's triggering right now. So the time is obviously sort of voided, right? But the action is actually doing a find email step. And this is going to find one email. And in this case, it's from David, right? This event that I attended yesterday. So it found this one email and You know, I was hoping it might find multiple emails, but this is a find email step. This is a singular step. And maybe I actually want to add a different action to this. We'll get to this in a second. But so this is going to also send a Slack message. And actually, it found multiple things. Fantastic. And look at all the direct email links. Now, where is this actually sending it? That's the question. Speaker 2: Yeah, like which channel within Slack. Speaker 1: Yeah, see, so the channel is dealer's choice. The agent could send this anywhere. Probably don't want to do that. You probably don't want to do that. So let's instead set a specific field here and we'll do this tutorials channel because this is a tutorial. And I'll save this. So now this is going to specify the tutorials channel down here at the bottom. And I'll approve this because I'm ready to get this message. And I'll pop open my tutorials channel in Slack here. And boom, here we go. I'm just going to switch my screen to share my Slack with you instead. And just like that, weekly orchestrated connecting email summary. I have 11 emails over the past week with some key highlights, positive feedback, a bunch of direct emails if I want to go directly, some thank yous, and consider joining. It actually gave me action items. I didn't ask for that. Speaker 2: I love that. Yeah. Right? Speaker 1: But the agent did that for me. Which is pretty cool. So I got 10 direct emails from these different folks and now I can action them if I want to action them. Speaker 2: I think that's a great use case too because like I'm a part of multiple communities and I don't necessarily have the time to stay up to date with every single thing that's going on with every single one of them every week. But if I just had kind of like a weekly digest of like, hey, here's all the communities you're a part of and here are suggested action items, because some weeks it's like there's not much going on and I don't really have to do anything. But some weeks it's like, oh, there's this event that I really want to remember to show up to. And if I don't have an agent like this, maybe proactively letting me know, I'll completely forget. Speaker 1: Well, that's really one of the use cases that AI actually shines in, which is like summarize some body of information. Where it's really not going to shine in is like post it to the right channel. It has no idea what channel to post it in. Speaker 2: Yeah. Speaker 1: Right. That's a topic that we actually just released a video about on our X-Ray automation channel on YouTube about the difference between deterministic programmatic automations and AI automations, where AI deals with this like fluffy summarization, categorization, ambiguity. Exactly, ambiguity, non-deterministic things. You don't know what that message every week is going to look like. It's different every time. It's supposed to be, right? But the deterministic side is search for this particular phrase every time, right? Don't get cheeky. Search for this phrase, right? And then post it directly to this channel. Don't guess what channel. I don't want that same message or that unique message every week in a different channel. That wouldn't work. Right? So the balance between non-deterministic things and deterministic things are really where you get to design a workflow that makes sense and it's faster and cheaper and like the right tool for the job. You don't need a sledgehammer to hang a picture on the wall, right? That type of thing. Speaker 2: Yeah, I think that's a great example because traditional Zapier automations, like traditional Zaps are, it's like it's got to be for the most part predictable input, predictable output. We're just automating the steps in between. The agent function, like you said, can deal with those non-deterministic inputs or outputs, right? It can take unstructured data or more ambiguous decision-making logic and still take action on that because of the AI ability built in, right? Speaker 1: So there's one extremely important step that I didn't show you yet, which I'm going to show you and then we can talk about if we need a different demo or not. Speaker 2: Yeah. Speaker 1: But that is None other than the publish button. We just did testing, right? This is a nice test. That was cute. But we actually need to publish a V1. So then when you publish it, you get this nice little green, you know, published here. You can actually see your activity. So, you know, today we actually ran it and it would cost you credits. So you don't actually pay for test runs, which is unique in Zapier. Some other automation providers do charge you to run tests. In this case, you're not charged to run tests. And I have 400 activities left in my fresh account. To you know, I could do 400. Summaries, basically. Each of these steps counts as one. So in the AI agent that we just built, that would cost three steps per run. Because we're triggering it, you're finding it in Gmail, and you're actually posting it to Slack. So there would be three activities consumed per run, I believe, unless Zapier has changed that. Yeah, monthly that each agent can perform. So you can actually dig into the data usage. Data usage is actually measured in this side of Zapier. In the automation side, data usage is not measured, which is a big deal if you're moving big files and that's a part of the equation. After that, you're pretty much done. Then you can get into Zapier and build a Zap, an agent, or something else that you need. That was our agent demo today. Speaker 2: That's sweet. That's awesome. That's a great use case. Do you have one more we could run through or is that what you had today? Speaker 1: Yeah, well what I was gonna do is show how to build it in a Zap form. Speaker 2: Yeah, let's do that. Speaker 1: Right, so if we... I'm actually gonna cheat. I'm gonna come back here, go to my agent, and I'm gonna copy... I'm gonna copy the whole thing that I just gave it. I'm gonna copy this, right? Come back, go to Zapier, Go to my Zap and you get this copilot again, and I'm just going to paste the same thing. Let's go to work Zapier, like build these steps for me in this regimented way. And one of the major differences between the automation building in Zap versus the AI agent is the ability to control and manipulate the model itself. So in an AI agent, you have no control. Zapier's AI agent is Zapier's AI agent. That is the way it is. You can't change any weights. You can't use open AI instead. You can't do anything. It is like Zapier's thing. Inside of the automation platform, and I'm really curious to see if this actually works and actually gets built. Inside of the automation platform, the AI by Zapier step. Actually has things that you can configure. Speaker 2: Right, because you can choose like which model you want to use, right? Whether it's Claude or OpenAI or Gemini or whatever, right? Speaker 1: You can choose which model. You can choose response format. Like if you want the subject line and the body to be, you know, drafted as an output from the AI step, it will do all of those things distinctly. So if you want to actually control the format, and you see in the search string here, from orchestrated connecting or subject orchestrated connecting, newer than seven days, that search string, you could actually go to your Gmail, search that, and that's how Gmail's search works. That's the syntax for Gmail. Speaker 2: So it sounds to me like, tell me if I'm understanding this correctly. So like the use case that you demoed here, right, is like going to Gmail and finding Emails that fall under certain criteria and then every week sending you a Slack summary of those emails. So it's like we could accomplish that objective with a Zapier agent or we could accomplish the exact same objective with just a normal Zapier workflow. But you're saying that what differentiates them is with like a regular Zapier workflow, sounds like you can get a bit more granular with like With really all steps of the process, right? With an agent, you're leaving a lot of that interpretation up to the agent of like, hey, based on Tom's prompt, this is probably how he's going to want the data. But with a true zap that we're building right here, it's like we can really get in the weeds of like, oh, I want it formatted exactly this way, and I want to make sure it includes exactly this information. Am I understanding that correctly? Speaker 1: Exactly right. That is exactly right. AI agents by Zapier are gonna be more expensive, they're gonna be slower, they're gonna be more error-prone, and you're gonna get a wider variety of... Speaker 2: Of outputs. Speaker 1: Of outputs, right, because you just can't fine-tune it as well. Speaker 2: Right. Speaker 1: Right? Speaker 2: But is the pro, then, that they're just easier to set up? Like, you spun that up in, what, seven, six, seven minutes? Yeah. And it got the job done, but it's like, like you said, maybe we want to get in there and just like, Tweak it a little bit and set it up to where it outputs things a very specific way every time, then it might make sense to just set it up as a Zap instead. Speaker 1: Exactly. And here I can control that search string. I can control if the automation fails if I don't find anything. I can create an email if it doesn't exist yet, which obviously I don't want to do. But inside of AI by Zapier, for whatever reason, the copilot gave me an old version of that, so I'll just update that. I don't know why. But I can continue here and let me just pop this out because I think this is a really helpful view. You can choose the input fields. So I will say like emails found and then I could pluck. All of the emails that are found inside of the Gmail step, right? And then I can specify output fields. So I can say, you know, summary here, and I can make this a specific field type if I wanted to, like text or whatever. So this type of granular control is going to be a lot better for when you need field validation or you need to be inside of a specific format or if I wanted to bold certain components of the message inside of Slack or if I wanted a particular formatting, you could do that inside of configuring the Slack step. So this got me here. This got me closer. I'm not going to spend the next 15, 20 minutes configuring that because it would probably take another 15 or 20 minutes to get this into a demo-able state, and I just don't think it's going to add value right now to your viewers. But you get the point. Speaker 2: The concept is what matters, right? That, to me, is coming into this, my goal is I want to know when I should use an agent What is Zapier agent versus just a regular zap? And you kind of answered that question. It's like, and again, correct me if I'm wrong, but it's like, you want to use an agent when it's like, we just need to get the job done. We don't care as much about the output being like exactly the same every time, as long as it gets the job done. And it, you know, I guess resolves the need that we have set out for it to accomplish and it's good to go. But if we want like, Super granular control over like it's got to be structured this way every time it can't deviate and we need these specific fields pulled in like that's when you'd want to create a regular zap, right? Speaker 1: Yeah, when you're dealing with soft verbose information, right, it's going to be best to use an agent. Because it's going to be able to adapt both, you know, flexibly. But when you're dealing with stuff like a database or tasks and projects or CRMs, like those types of things that have a rigid data structure that you really want to specify a number of fields and a number of data values, like it's worth it to spend the extra 15 minutes in the configuration because then you don't spend the extra 50 minutes in sweeping up with the agent Built For You because you didn't specify. That's the trade-off in time. It's going to be fast to set up and get something working, but you might need to do a bunch of cleanup if you were sloppy with how you configured it in the first place. It's going to be more expensive. It's a trade-off. You've got to weigh your options. Know that you might switch between the two and you might set up a prototype and be like, wow, this is really nice, but I want to make it good, right? Speaker 2: I validate it as useful. That's actually smart. I didn't think about it that way, but it's like you could, depending on the workflow, it's like, hey, I'm going to kind of spin up an agent and that agent is going to be kind of my MVP. It's like, is this even possible? And if the agent can do a decent job or get the job done, then it's like, all right, we proved this works. We've proven that this can be done inside of Zapier's platform. So now let's expand it and really dial it in and turn it into a more custom A granular zap. Is that kind of a way to think about it? Speaker 1: Yeah, absolutely. And that's the secret sauce from the 800-pound gorilla of Zapier. It's like everything that AI agents can functionally do is identical to everything that you can functionally do inside of the automation, like zaps portion. Those are the same API routes. It has the same integrations. It has all of the same information. So one last thing that I will share with all of you is this free resource that we ended up building here at X-Ray for figuring out where and what can I build with the tools that I'm using. So this is our website, xray.tech. And, you know, we have a whole bunch of different tools here for your, you know, simplicity. Search is really what I want to focus on. Search.xray.tech is the URL. And, you know, we'll just use Gmail as an example. Everything that we built today, Gmail and Slack. Those are the two tools that we talked about. So we'll just open up Slack's View Automation Abilities. And here you can see all of the triggers, actions, and searches that are available for Slack on Zapier, Make, N8n, Workato, Bardeen, and Power Automate. All of these things are indexed in one place. So you can say, oh, is there something that I can do with a channel? Oh, I can. I can watch direct messages in a channel on Make, or I can see a new public message posted anywhere, a public channel, right? Speaker 2: Zapier, right. Speaker 1: Exactly. So you can like really compare and understand, should I even use Zapier? Or is something like N8n good enough? Or like, you know, what can I actually do with the tools that I'm using today? Gmail has some really simple triggers, new label, new labeled conversation, new email. All these things are just triggers and actions. These are like the starting points of workflows or actions that would be in between steps inside of your workflow. Agents and automations can do all of the things on these pages. Speaker 2: And so I'll give you props because I feel, I'm sure this absolutely crushes from an SEO perspective, like that kind of directory style pages on your website. And it's funny, that's something we're looking into for our website as well. It's just like a programmatic SEO strategy. But I love how you took, because like Zapier kind of made that playbook, right? I'm sure folks listening to this and yourself and I certainly have, it's like, oh, well, what can I automate with Zapier? And Zapier has a really good, system for it's like, we'll just type in the tools you use. And it's like, Oh, you want to automate from Gmail to Slack? These are the triggers. And these are the actions that you can take. And you've taken that exact same model, but you've applied it to all the automation platforms. Cause it's like, well, you know, maybe Zapier can't do XYZ, but maybe make.com can, right? Or maybe neither of them can, but N8N can. And like, so I just want to give you like that. As soon as you were sharing that, I'm like, damn, we got to copy that. Cause I'm so good at it. Speaker 1: Well, it's a lot harder than it looks. It might be a little bit easier now than it was, you know, a couple of years ago, but yeah, it's, it's, you'd be surprised though, generally like the top 100 SaaS tools all have nearly identical support. Any tool that you realistically use day to day, it's going to be the same. It's the niche products, things like in the legal sphere, Clio is a big case management tool, or if you're in real estate or if you're in construction, there's some really specific tools that are there. That's the extremely long tail of Zapier. Like, Zapier has those integrations and Make and Power Automate don't. But, like, Power Automate would have integrations, you know, better support for, like, SAP. Speaker 2: Right. Speaker 1: So it's just, like, It's just different per platform. We have that data. We spent a lot of time and energy trying to figure out the right way to index it and present it because this was a tool that we were using for the last three, four years. With our re-release of our brand with our five-year anniversary this past month, we made all this stuff public for everybody. Speaker 2: That's smart, because again, I think from an SEO perspective, I'm sure you guys are getting a good amount of traffic from what you just showed us. Speaker 1: It's helpful, you know, and builders like you and hopefully your viewers, it's a free resource. We're never going to charge for it. If you're trying to answer the question, what can I automate with my tools? Just do the search. Speaker 2: What's the URL real quick, just so that maybe if people are listening on a podcast platform? Speaker 1: It's search.xray.tech, T-E-C-H. Speaker 2: Perfect. Yeah. And we'll include that in the show notes, of course. Speaker 1: Yeah. And you can access it from our website, just x-ray.tech. It's one of the products that we highlight. And if you have any feedback or other things that you would love to see, I'd be very happy to add it to the platform. We're really starting to get in a good cadence of shipping stuff right now. We just released all five of our internal products. For different things from error monitoring and reporting to search to onboarding, there's a whole bunch of things that are really helpful if you're trying to run an automation agency or an agency generally. And we're very competitively priced. We'll put it that way. Speaker 2: Yeah. Speaker 1: So happy to talk to you and figure it out if we can help. Speaker 2: Awesome. Well, Tom, this has been fantastic, man. I've certainly learned a ton. I know the power of Zapier and I've been trying to spread the word for years for other people to get in there and take advantage of the platform. And I think you did a really good job of breaking down really when you would use an agent versus when you would just use a regular Zap. Great work. Where can people follow you or find out more about you? Like, where do you want to send people? Speaker 1: Yeah, I would say follow us on YouTube or LinkedIn. We are X-Ray Automation on both. We post all of our updates there. You can find me, Tom Nassr, on YouTube or LinkedIn. But those are our two main platforms. So I would say follow us there if you want more updates. Speaker 2: Awesome. Well, Tom, thank you so much for the time, man. And of course, we'll put all the links he mentioned in the description and in the show notes. Be sure to go and follow Tom, and we will be back to build something new next week. Speaker 1: Thanks. See ya.