# Build an Email Manager AI Agent | Build Along Workshop

## Метаданные

- **Канал:** Zapier
- **YouTube:** https://www.youtube.com/watch?v=SVu_9bNPLqE
- **Дата:** 04.05.2026
- **Длительность:** 1:00:51
- **Просмотры:** 268

## Описание

Get started with Zapier for free at: https://zpr.io/XUiAHTJw2csj

Subscribe to the Zapier YouTube channel so you don't miss other important tutorials and integrations:  
https://www.youtube.com/Zapier?sub_confirmation=1

Need customer support? https://zapier.com/app/get-help

## Содержание

### [0:00](https://www.youtube.com/watch?v=SVu_9bNPLqE) Segment 1 (00:00 - 05:00)

Hello everybody. Happy Thursday. It's Thursday for me. I'm in uh Madison, Wisconsin where it is firmly afternoon time. I'm so excited to see so many of you here right at the top of the hour. Thank you for the validation. It is Thursday. Uh, for those of you who are already with us, you probably know by now that I have a favorite icebreaker I love to force people to do in the chat in every meeting I've ever been in, which is tell me the best thing you consumed in the last 48 hours. So, could be something you ate or something you drank or something you listened to or read or watched. Think of it broadly. What is the best thing you consumed in the last 48 hours? Oh my gosh. I'm going to reveal a dark secret to you because I saw the answer in the chat of onion soup. Okay. Oh, now there's another soup. Two soups. Butternut soup. Here's my dark secret, friends. I cannot. My brain and my mouth do not understand soup. Feels like a beverage, but people tell me it's a food. So, so far my brain and my mouth, we can't agree with you on soup. But I want to like soup because in like all the cozy picture books, people are eating soup. Cereal is soup. Doug, that is controversial. And man, I'm seeing a lot of disagreement with me in the chat about the soup thing. I didn't mean to make such a controversy. I'm so sorry. I've lost friends and followers in this moment. I can't do soup. Uh but uh a lot of people sharing other things they consumed in the last 48 hours that were their favorite. Share them in the chat. I will tell you that I had uh Thai uh basil stir fry for lunch and it was fantastic. So that's definitely my answer. Keep it coming. We've got a few hundred people I'm losing trust with about my soup take. So, let's get started. Let's kick it off. We're a few minutes in. I'm so excited to hang out with you. If we've never met before, uh I'm going to give you an introduction. My name is Emily Maybe, and I work at Zapier. I'm a AI automation engineer for people teams uh for groups of people who often like have told themselves for a long time that they're not technical. And so I love the opportunity to do workshops like this where I might take someone from never using AI and automating with it before to having a fully functional AI automation in less than an hour. That's why we're here today. Today we're going to be tackling the question of email and the inbox that's always on fire. But before we do, some housekeeping. Here it is. First of all, our most popular question every single time is Emily, is this session being recorded? It is being recorded. And the other question we get more than any other question is, I'm stuck. Can I please have help? Now, the great news is we can get you unstuck. The bad news is there's only one of me and there's hundreds of you. So, if something is not super clear, if you're feeling stuck, if you're working on like something slightly different while we're working, there's something I'm going to show you called C-Pilot. It is your best friend for troubleshooting. You can chat with Copilot just like you would chat with me about what's wrong, but Copilot will actually be able to unstick you. And the last thing I want to say is you don't need any previous experience in order to do this workshop today. You just need to sign up for a free Zapier account. Uh Bryce is going to be able to share a link so that people can sign up. Uh my friend Bryce, by the way, is also going to be the one who's in the chat and in the Q& A tab. Uh so if you're interacting with someone in those spaces, it's probably Bryce. Bryce and I, we go together like mayonnaise and pretzels. Just great. Um, since you all were here first, over 300 of you were here right away. I'm going to deputize you. Okay? You have a special job. Your special job is when people join late, I want you to make them feel welcome because I want them to come back. But I also want your help answering these three questions, these three housekeeping questions. Usually when people join late they will ask is this being recorded or like how do I get help or do I have to be a Zappier user? Those kinds of questions you all know

### [5:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=300s) Segment 2 (05:00 - 10:00)

the answer so you can help us in the chat. That is super helpful. Okay you've been deputized. Now just a little more housekeeping and then we'll get our fingers on the keyboards. If you have questions during the session please do me a favor and put them in the questions tab. You can also go in there and upvote questions that you really like. That way, they'll rise to the top of the order. They might get voiced over by me during the Q& A. So, that's where questions should go. If you put them in the regular chat, they will get lost because, as you can see, hundreds of people are typing in there and stuff moves kind of quickly. So, questions, put them in the question tab. And then the other thing I want to point out is if you want me smaller like my face and you want the screen I'm showing bigger, you can slide the adjuster and toggle the questions and chat sidebar on and off by clicking those buttons to adjust your screen. Uh so please do that. Make it work for you so that you can see what you need to see. Speaking of seeing things you need to see, if you need to see some really great uh talent acquisition focused workshops, boy have I got a treat for you. My incredible and also very nice teammate Casey uh is a senior talent acquisition AI automation analyst at Zapier. Really fancy title. That means she makes magic. She teaches me new things all the time. and she's got a pair of workshops coming up just for AI agents for talent acquisition. So, if you're on like a people team or you respond to candidates or you review resumeumés, these sessions are for you. Be sure to register. Bryce already shared a link because he's a gem. And I want to see you there. So, if you're interested, sign up. Okay. What are we doing today? Today, we are going to start with a really quick intro. get to know each other. Then we'll transition into our problem and the tool that we're going to use, Zapier agents, and then we'll talk about what we're going to be building live together. You guys are actually going to help me figure out the design today. We're going to run it a little differently to help keep up with AI and then we'll do some Q& A. So, start us off, please. I would love to know. I gave you my intro so that you can build some trust with me. I want to know about you. What is your role and what do you want to do with AI? So, I have an example up here. Um, I'm a marketer and with AI, I want to better manage my content calendars. That's a great example. Tell me more. What is your role? What do you want to be doing with AI? Oh my gosh, Stacy. Okay, I saw Stacy's answer right away. We have um please keep typing everyone, but we have an incredible suite of templates from our executive assistants that showcase like how they use agents and automation for their uh to manage their execs and it's amazing. Hopefully I will remember to drop it uh in the chat later, but uh stay tuned for that because it's really great. Programs manager. Oh, you're a founder and you want to use it. Oh, okay. Great. Let's see. Oh, they're going so fast. Real estate workflows. Oh, automate front desk type workflows. What a great use case. Want to get to inbox zero. Yes, we are going to work on that today. Oh. Oh my gosh. I want to have more time to focus on strategy, less time focusing on emails at the end of my day. Oh, I'm a photographer. I want to automate tasks so I can spend more time actually in the field. What a beautiful answer. And also bonus points to the honesty in the chat uh about like I'm actually not even sure how AI and automation can help me yet. These are beautiful and uh I love by the way all the help people are giving each other in the chat already with questions. Thank you so much. It is indeed being recorded. I appreciate that. All right, let me read one more. Ideally, creating an inbound email flow as a catchall. These are fantastic. Let's get into the tool. Keep those like what I want to do top of mind and let's talk about the tool we're going to use today. All right. So, if this is your first time with us, welcome. This is my favorite metaphor to talk about what we're going to do and it's to answer the question like what even is an AI agent, which is what we're going to build with today. Almost every single one of us has talked to AI by now. We've chatted with chat GPT or Claude or Gemini. When you talk to them, it's just like

### [10:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=600s) Segment 3 (10:00 - 15:00)

using Google Maps. You open up Google Maps, you say, "I want to get to the movie theater. " It gives you an answer for how to get there, and then you got to go. You're doing all the work. That's what it's like chatting with an LLM. It's a manual and one-off interaction with a really, really smart assistant. But what we're going to do today is more like a Whimo or a self-driving car where you open up your app, you say, "I want to get to the movie theater. " And then it does all the work to get you there. Right? You get in the car and it takes you on the route to fulfill that goal. That's what it's like working with agents. They're automated teammates that work towards a predefined goal. And there's a lot of stuff that they're really, really good at. agents can gather context from the internet websites they can use. It's now over 9,000 of Zapier's connected apps with tons and tons of actions. And it's really great at tasks like extracting or summarizing or classifying or writing, which is exactly why it's such a great tool for managing our inbox. So with all of those things, you can go and take actions in the work apps that you're already using every day, but you can have AI take those actions and it can just do those things automatically if you'd like. All you have to do is tell it when to wake up and what to do once it's awake. Now, that's sort of an overview of AI agents. The agent that we're going to be building today is a Zapier agent. So that's the agent product that we'll use. When we build with Zapier agents, you can create as many agents as you want using plain natural language and you give each one a role and access to your apps and your business data. And then you give it instructions so it can do work for you. You've likely seen this example from me by now. I like to include real examples of agents that I use every single day. Here's one of mine. It's a Zapier agent. The whole idea was I'm really bad at remembering to schedule my PTO. So, this agent uses human language and a few tools to do that autonomously. And you can see you don't need to be able to read code to read it. It's written like a person. Uh, but how do I build it? And how will we be building together today? The first part of building an agent is asking yourself what should trigger it. I like to think of this as um what should wake your agent up to start getting work done. So you can see in this example my trigger was when it's four weeks out from an event starting. And when we fill in the blanks in a sentence that sounds like when blank happens then blank happens that first blank is always our trigger. So again here my trigger was when it's four weeks away from an event starting on this calendar. In this case it was the US public holidays calendar. And then we have to fill in the second blank. So when it's four weeks away from the start of something on the US holidays calendar then I want you to and we fill in that blank using plain natural language. No code. So, in this case, when we're four weeks away from that holiday, right, then I wanted my agent to check, did Emily schedule some PTO? If you don't find it on my calendar, I want you to go into our HIS and request it. I want you to send me a DM to confirm, maybe even DM my manager. That's how we write the then actions just with plain language. And finally, we have to add in the tools. So, whenever you need an agent to reach outside of itself and use an app or like browse the web or a specific site, you've got to let it know how to do that. Um, so for instance, here I needed it to go look on my Google calendar to see if I had scheduled PTO. I needed it to go into our HIS to request PTO if it didn't find any. I needed it to go into Slack and send a DM to me. So I had to drop those tools inside to get it to execute that work. But you might notice that I don't have any clawed or chat GPT steps in here because the agent itself is running on a large language model as its foundation. So I don't have to go add in chat GPT. The agent itself can already summarize, evaluate, categorize. I don't have to add claud or chat GPT into it. Okay. Last thing I'll share with you about how to think about an agent is I like to remind people that agents can be really simple or agents can be very complex. But when I say that I don't mean complex

### [15:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=900s) Segment 4 (15:00 - 20:00)

as in like hard to build. It just means that the agent could take multiple steps and actions and execute a variety of tasks. So you can see on my screen where I like scroll through it. It's still written in normal natural language. It doesn't mean like it was harder to build. I just gave it more instructions and more tools. So what does it look like to actually do clicking and building? We're about to dive in. And I want to remind you one more time, especially if you joined late. We are about to build together. So we're going to put our hands on our keyboards and our mouses and we are going to build an agent. But there are hundreds of you in here with me, which by the way makes me feel very warm and fuzzy inside. So, thanks for hanging out. But it also means I can't troubleshoot one-on-one with you today like I could if you had invited me to come to your company and do like a smaller gathering. So, if you get stuck, I'm going to ask you to ask Co-Pilot for help. I have a picture of that on the screen. Okay. You can use co-pilot for things like, hey, you know, this doesn't look like it's working right. What's going on? Or what should I do here? You can talk to it just like you would talk to me if we were one-on-one. It can do a bunch of other great stuff, too. Like you can chat with it to build stuff. It can give you new ideas. Uh, but today I want you to remember if you're troubleshooting, open C-pilot right in your agent. But I'll point that out as we build too. For now, let's get started. Here we go. If you've got two monitors, now is the time to set up one window where you can see me as like a video stream and in your other window you'll be building your Zap year agent. If you've only got one monitor, that's totally fine. I will be working from one monitor today. Uh but if you've got two, it's helpful to not have to click back and forth. Okay. Today, the problem we are going to solve with our agent is we get dozens, if not unfortunately, hundreds of emails. Hitting our inbox every single day, and staying on top of them can feel like a full-time job. I heard lots of great use cases from you in the chat about this already. We want to be responding faster. We want to also set healthy boundaries so our inbox isn't running our whole day, but we also want our communication to sound personal and actually like meet the needs of our users. So today we're going to build our solution together and you are going to help decide what it looks like. So get ready. This is going to be a little different than if you've built with me before. The first thing we have to do is open Zapier agents. Now, Bryce shared a link already, but if you didn't see it yet, thank you, Bryce. He popped it back up on the screen again. What a gem. You can go ahead and click that start building button. If you're trying to get there from the Zapier homepage, zapier. com, I'm just going to head over here, upper leftand corner. I'm going to go down to agents. Agents right there. Okay, this really cool box pops up. Oh my gosh, look at this. I could build one of these templates. I could tell it in one sentence what I want to build, but today I want your help building it from scratch. So instead, I'm going to click the purple button that says start from scratch. I'm going to click that and it's going to open our blank Zapier agent. Here we go. Right now, it's an empty field and we've got to sew our ideas inside of it. So, what I want to do is get some inspiration from you. Here's what I think we're going to use as our trigger. And remember, our trigger is the first half of the when then sentence. When blank then blank. So, I'm going to say when a new email comes in, I'm going to use Gmail for this. Okay. So, if you missed where I clicked, I clicked right here. Add a trigger. Okay. My trigger is Gmail because that's the email inbox I have access to. And I want to wake our agent up whenever a new email comes in. So, I'm going to choose new email. And you might notice that mine found my Gmail account right away. That's because I've authorized a Zapier agent to use my email before. If you don't see your name in the dropdown, there's a button that says connect new account, you can just click that button and it will take you about 20 seconds to connect your Gmail account. Okay? So, if you haven't authorized it before, stick with us. It'll just take you a second.

### [20:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=1200s) Segment 5 (20:00 - 25:00)

And if I click show all options, the purple button, you can see I can get even more granular, they call it even more specific. I could say only tell me when new emails come into this inbox or that inbox. I'm going to leave that alone. I just want it to wake up whenever I get any email. Okay. So, I'm a big fan of dropping screenshots while I work so that if you are ever like two clicks uh behind where I am, you don't feel lost. Here's how I set up this trigger. I'm going to click add trigger. And now I have half of my agent already built, if you can believe it, because I have half my win then sentence done. When a new email comes into my inbox, then blank. So, we're halfway there, folks. Already, now here's the exciting part for me at least. You are going to decide on the second half. So, here's how it starts. It starts out when a new email comes into my inbox. Now, I don't have to type this out because it's already in the trigger. I'm just doing that so you all can see what we need to fill in next. Then what should happen? And I'm going to challenge you. I want you to tell me ideas in the chat. What would you like in a in an ideal world? You have the power of AI and agents at your fingertips and they are going to just handle your inbox. What do you want it to do? Oh my gosh, the ideas are already beautiful. I want it to draft a reply. sort out the spam. Don't even let me see it. Oh, scan emails for cold sales and archive them. Label them. Okay, so help you categorize. Make a to-do if it's like a task. Send a Google chat to a specific person. Wow, these are really varied, everybody. These are great. Amazing. Okay, keep them coming. Apply labels and then send me a digest so I don't miss anything I care about. Big plus one on that one. I don't live in my email inbox as much as I live in like texts or Slack, so I love like send me something else. Create. Okay, awesome. Yes, go to the grocery store and get my groceries. I know you're joking, but I do have a colleague who has a a whole setup where um based on her communications, it does auto place the order for a grocery pickup. So, it's a great idea. Um, okay. If it's from my boss, check the email sentiment and then send me a DM. Uh, Richard, that one speaks to my heart. Uh, because I am someone who gets email anxiety. If I see an email that might be heated or sad or something, like I will avoid opening it, I'd love a little like, "It's okay to open this, Emily. The sentiment's fine. " Or like, "Go read this right now. The sentiment's really heated. You got to resolve it. " I love it. Send me an email with a meeting prep checklist. Sort the emails. Oo, delegate them, Jenny. That is next level stuff. These are fantastic. Let's take a few and just plug them in here. By the way, this is gonna sound wild, but if you were able to type that in the chat, you can build the agent. So, if I don't take your exact use case, if you're like, Emily, I loved my idea about um making a to-doist task and you didn't build that one. I promise that you can build that one. Okay? So, I'm going to try to take some that I'm hearing over and over again. I heard Oh, boy. I heard a whole lot of help filter stuff out. So, let's add that as our first uh step. So, I'm going to say when a new email comes into my inbox, then I'm going to say step one, if it's junk, and I saw in the chat a few examples of this. Um, okay. Spam, you said notification emails. Okay. Very good. Okay. I heard sales emails one, two, three, four times. So, I'm gonna I'm going to make that one a separate one because I noticed you talked about them as sort of their own category. Uh, so I'll just say etc. I'm just going to say mark them as red or even we could just say delete them. I get nervous about having it delete stuff. Uh, and something I love about agents is you can control exactly which actions you let it take. So, if you're like, "Don't delete anything. Just mark it as red so it gets out of my inbox or label it and move it to a folder," you could do that, too. So, I'm going to say, "Okay, if it's junk, just mark them as red so I don't even see them in my

### [25:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=1500s) Segment 6 (25:00 - 30:00)

unread inbox. Let's use another one of your ideas. " Okay. Yep. Multiple mentions of cold calls from sales. I love it. Okay. So, let's go with um if it's a cold call from sales, I love the idea of labeling it. I don't know if I have a label for that in my inbox, but I could make one. I'm going to go with uh if it's cold call from sales, mark it as red. Um and draft me a polite no thank you reply. That's uh I will say friendly because that's what I'm known for. But a very firm no. Okay. Massage that however you like. Friends, make it sound like you. And then I want to grab one more of those. This one. Oh, Erica asks a great question. Eric, I love that question. If you put it in the questions tab, I bet we can voice it over. Uh okay. I see lots of ideas about like pulling something out of email. um you know, okay, let me know about it in Slack, put it in a spreadsheet, task tracker. Lots of flavors of that. Um or like if it's a specific person, take a certain set of instructions. These are great ideas. Let me see which one of these will allow me to show you something really unique. I'm gonna go with um in all other cases when a reply is needed. Okay, because sometimes people just send you like a funny meme. I don't need to know about that. When a reply is needed, example, um I have to do something or tell someone something. draft a reply and let me know in Slack that I've got an email to review. And you know what I love? Somebody in the chat shared like you could just get them as a digest. If it's not super urgent, just tell me like once an hour, right? Or once a day. And if it's really urgent, tell me right away. Okay, keep those stored up here because we can come back and add them back in. For now, I'm going to grab this whole chunk of text and put it in the chat in case you type at a different speed than I do. I'm going to grab a screenshot of what you came up with. These are just some of your ideas. Man, your ideas are so good. And there are so many. And I'm gonna tell you, we are almost done. If you can believe it. We're almost ready to turn this on. But we're not quite yet. If I turn it on right now, what would happen if I were to turn this on right now? What would happen? Oh my gosh, gorgeous work. Gorgeous work, Emily. It's not going to know how to do this stuff. I love it. Oh, we got multiple answers that are fantastic. Emily, you didn't add tools. It's not going to be able to do these things because you told it to go into labels. draft a response. You told it to go to Slack, but you didn't give it access. And then I also saw Emily, nothing would happen because you didn't hit publish. You are correct. Also big plus one to the idea in the chat that like so far a lot of what we're talking about could be automatic email labeling. What makes it valuable is layering in customization and logic. Absolutely. We'll get back to that in just a minute. So what we need to do, you're spot on. I shared it in the slides. You've seen it before maybe in your own work too. We have to add the tools. It can see the new email, but it certainly cannot mark it as red because we didn't give it that tool. So, let's add the tool. Right now, I'm going to use the slash key. When I do that, a box pops up that says add tools. If you are not using a keyboard, if you need to use your mouse, you can do it from this button down here that says add tool. Either way will give you the same result. I'm going to click add tools. And in this case, I mentioned I need it to mark junk as red. This is where I'm going to go into Gmail as my tool. And you can see lots of options in here.

### [30:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=1800s) Segment 7 (30:00 - 35:00)

I'm just going to scroll because look at all these things you could have it do. remove a label. You could have it add a label, archive an email, delete an email. I love your ideas, by the way. So many of you had so many different approaches. Some of you felt confident deleting stuff from an agent. Some of you said, "Let's just archive it. " In this case, I'm telling my agent, I just want you to remove the unread label. That's all I want you to do. Just remove unread so that it goes away. The way I have my inbox set up, I won't see it. If it goes into unreads, it looks like I've taken care of it. But if you were thinking, Emily, I wanted to archive mine. That's how you would do it, too. Okay, very similarly. So, we insert the tool. If it's junk, mark them as red. So, I don't even see them in my unread inbox. If I click the little gear icon next to the tool, I see these options. Okay, it asks me, "What label do you want your agent to remove? " Now, I think my instructions to my agent were pretty clear. I could let it choose for itself, but I want to show you that you also have the option in Zapier agents to set specific values. So in this case, I want it to remove the unread label. So all of my incoming emails before I get a chance to read them have the unread label. It's why they're at the top of my inbox. I want this agent to just take that label away and then I'm going to hit save. Okay, so step one, if it's junk, mark it as red. So I don't even see it in my unread inbox. And another shout out in the chat, I'm hearing folks posit like why they would rather archive instead. Something that I say a lot in my workshops is the most important part of AI automation is customizing it for you. So if you would not like it marked unread, if it would bother you and you're like, "No, Emily, I don't want it opened at all. I'm afraid of what that would do as a signal to like email tracking marketing campaigns. I just want it archived. Please pick a different tool. This is part of the magic, right? We each have different ways of working. It's really important that you build automations that work the way you work. So, really great stuff in the chat. If you're like, "Oh, I have a strong difference of opinion. " That's wonderful. Do that. Now, I'm going to go to step two. Step two, you all I saw this several times that we're struggling with sales messages, people, and I'm right there with you as a people pleaser. hopefully recovering people pleaser. I love that AI lets me get some help on cold call outreaches so that I can say no in a friendly way. Keep a positive relationship but make it firm so that they understand it's a no. So I am going to have my AI helper here check if it's a cold call from sales. If it is, I want it marked as red and I want it to draft me that polite no thank you. Now, we need to check for a second. And I'm actually going to scroll down here so you can see. Here's a list of all of the tools our agent has access to right now. So, if we look at step two, I've given it a few things to do. I said, if it's a cold call, mark it as red and draft me a polite no thank you reply. What tool do I need to add for step two? Exactly. So, this is kind of tricky, but you guys are nailing it. So, because I've already given the agent the power to remove the unread label, it can already do this first part. It can already mark emails as red. But you're exactly right. I have not given it the power to draft a no thank you. And I'm hammering this for a reason. If you're like, Emily, man, you're really beating this up. Here's the reason I build with people every day, all day. And something I notice is they see like Gmail in the trigger. They see it here in one of the actions. And in their head, they'll think the agent has access to Gmail. It can do anything I want it to do in Gmail. But it can't. It can only take the actions that we allow it to. And that's really important. That helps us keep like safe building practices. is we don't have to worry about AI like going rogue. But it's also important to remember that like even though we gave it access to one Gmail action, it doesn't have access to all of them. So let's go add in another one. So in this

### [35:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=2100s) Segment 8 (35:00 - 40:00)

case, drafting a reply. You were exactly right. I am going to use that slash key to add another tool. It's Gmail. And I'm just going to search for the word reply. And we've got two options here. I want to point it out. One is creating a draft reply from an email it got. The other one is sending the reply. So, not drafting it, just sending it off. You might get to a point where you've customized your agent, you feel really good about the emails it's writing, and you're okay with it just sending messages. But since we're just starting out, we're going to use draft reply. All right. So, in this case, it's going to create a draft for us in our inbox, not send one right away. And when I open the little tool icon, so you can see all of the fields here. There's so much that we can customize. We can customize like we could add a label anytime our agent drafts an email. We could have a special label so we could go back through and like audit those, right? Uh we could change up our signature like assist AI assisted email. We could do all sorts of stuff. In this case, I'm actually not going to do any of these. I'm going to leave them exactly the way they are so that our agent can make all those choices. But just know there's a lot of customization we can do. So, I am actually not going to change any of those fields. At this point, we are almost done. I I appreciate, by the way, so much the cheering for human in the loop. uh because I think that's one of the most important considerations is when we're building where do we want human control and to hand off to AI because humans are ultimately responsible either way. Now we're on our last bullet point here then we'll check it out see that it's good and we'll make it great. So bullet point three in all other cases when a reply is needed draft a reply. Now we just covered this. I don't have to add that back in because it has the tool create draft reply, but I do have to add let me know in Slack that I've got an email to review. Now, for most of us, we're probably going to want a direct message in Slack. I'm just going to warn you that I'm going to send it to a channel, mostly because I don't want to show you all my Slack DMs. That's private. Uh, but we're going to add it exactly the same way. And there have been cases I'm going to click add tools. There have been cases where when I'm building with teams, they actually do want a channel message because they might get emails in their inbox that are for like the entire team. Like maybe they work on a recruiting team and someone's emailing just them or the recruiting email and they want it shared to a channel. So, I'm going to use send channel message, but you might want to use send direct message if you want to just get a heads up for you right in your Slack DMs. Again, just a heads up. I'm going to use send channel message for this today. And that's just because who knows if I can trust you people. I'm not showing you my inbox. Uh, that being said, I need to hook up Slack. I'm gonna click the little gear. The little gear opens like this. Just like with your email, if you have hooked Slack up to Zapier before, you'll see it in the dropdown as an option. I have used Zapier with my Slack before. So, I see myself right here. If you have not, you just need to click connect new account. And it'll take you just a second. I'm going to screenshot with an arrow here. Connect new account just in case you're at that phase. And super important, there is a channel requirement. So, I'm going to find the channel I want and I'm going to use a channel that's just called test webinar because that's one you guys are allowed to see. Test webinar. But you can see once I hook up my Slack account, I can see all of the options I have. So if I search, it just makes it a lot easier. And I'm going to leave all the rest of this blank. I'm going to leave it all blank. I'm just going to choose my account and my channel. One more screenshot for you all. Okay, I'm going to hit [snorts] save and then we're going to test this out. All right, and you're going to see that it works and it's good, but your responsibility, get ready, is you're going to help me make it better.

### [40:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=2400s) Segment 9 (40:00 - 45:00)

So I am going to click at the very bottom there's a button called agent preview. And if I click that it opens up a box that says do you want to save your instructions and test which I do. I want to save them. I want to test it out right now. And what a great question to username. Very good question Aaron. If you are trying to uh set up a DM to yourself, the uh the two would be your uh Slack ID or your Slack handle, the person it's going to. All right. So, if you're testing with me, something to note is that um it will look for like a recent email. Okay? if it doesn't find one, if you haven't gotten an email in the last like hour or so, it might show you an error. And in that case, the way that I usually uh have solved this is just sending myself an email so that it can like find one. And if you want to like really test it, you could try sending yourself a cold call. See how it handles that since you gave it instructions. Uh, okay. So, in this case, I've got this email from a colleague. It's saying, "Hey, can you reschedu this calendar uh invitation? It's at a time that doesn't work for me. " And now it's classifying it. Look at that. It's doing exactly what I asked. It's saying, "Okay, this sounds like a legitimate Whoops. work email and it needs your action. Emily has to delete a calendar event. This is case three. It needs a draft reply and a notification in Slack. " So then it said, "Hey, I'm going to draft this reply. " Look at that. Here's the draft that it's going to create. Hey, Christina, thanks for a heads up. I'll go ahead and delete that. Looking forward to it. Here's the preview. How does it look now? You might be thinking, "Oh no, this is very manual. I'm having to approve this. That feels like more work. " Just a quick reset. That only happens in the testing window. Once everything looks good and we turn it on, this will happen in the background without you worrying about pushing approve, um, it's only asking for your approval while we're like setting the agent up. So, in this case, my draft looks good. That looks great. Doesn't need to be over complicated. Uh, I'm going to hit approve, which means it's actually going to create the draft reply right in my inbox. I'll be able to check it out there in a second, but it has more work to do, right? I said draft a reply and then let me know in Slack. So, it's got to plan its next steps and it's saying, "All right, here's what I'm going to send in Slack. A new email needs your attention. Here's who it's from. Here's the subject. And here's the action needed. " Great. I didn't even go into detail. I just said, "Let me know you've got an email to review. " And it even Oh my gosh, it even added a link to the email, which is my favorite. I'm going to approve that so I can check it out in Slack. Let's see it come through. Okay, I'm gonna go check Slack. Let me get out of there. Okay, look at that. This is what it looks like when it comes through in Slack. Hey, new email needs your attention. Here's the detail. And then it even linked to the draft. It said, here's the reply. We can go check it out. Let's go see [singing] it. Awesome. Okay. And by the way, I saw a really helpful Look at that. My draft looks beautiful. Okay. I saw a really helpful uh question in the chat. Here's what I'm going to do. I'm going to here I'm going to zoom in so you all can see uh my draft reply. Uh I'm going to answer that. But before I answer the great question in the chat, I want you all to think for a second. This is good. The agent's good, right? It works fine. I can go turn it on. I'm gonna hit publish. And now it's working. But it's not like great yet. How could we make it great instead of just good? What would you do to make it great? And while you're answering that, I'm going to answer the question in the chat. Some of you noticed that uh when you were hooking this up, you got a warning. It said the tools your agent is using could create an infinite loop. I'm going to explain that really quickly. Okay, you can see it on mine, too. It's saying when you get a new email, then

### [45:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=2700s) Segment 10 (45:00 - 50:00)

draft a new email. So, it's saying warning, right? You've picked tools that could cause a loop. So, here's why it's not going to cause a loop. The new emails we're listening for are in the inbox. Our drafts are in the drafts folder. So that's not going to create a loop, but it's just warning you because then if you like sent yourself that draft every time, you would create a loop. Your agent would keep running running running cycllically, but it's just giving us a helpful heads up and we don't have to worry about it for this build. Okay, what are we getting in the chat for ideas? Have it. Oh, Peggy took the words right out of my mouth. Peggy's like, "If it noticed that the email says you got to delete a calendar event, what if it just deleted the calendar event? " We could give it the power to take actions that it reads about in the email. Love it. Let's see more ideas. Have it use a preferred email signature. I love it. Uh, give it a knowledge source for our email writing style. Beautiful. Absolutely. These are great. Okay. These are great ideas. So, a few that I'm hearing, make it sound like you, right? That draft right now. Like, this could be what I sound like. I'll give you a heads up. It's not because I have a favorite emoji that I put in almost every single communication, which is this orange heart. So, it's kind of a dead giveaway that that's not me. Um, I also use a lot of semicolons and parentheses. Those are things I know about myself. Uh, so we could make it sound more like us. We could have it take the actions it hears I'm supposed to take in the emails. Great ideas. I'm going to do one of those with you and the other one I'm just going to talk about. So, first I'm going to talk about having the agent go off and do something, right? we could give it the tools like let's say I am um an executive assistant and a ton of my job is moving calendar events around deleting them creating them responding to requests I could absolutely give it the power to like if you've been asked to do anything with Google Calendar uh use these tools and do it and then I could just give it Google Calendar tools. Two thoughts there. Number one, you might want to have a step in the middle where it asks for your approval or if you know that the thing you're getting emails about all the time being asked to do uh is like very low risk and you're totally comfortable giving it those tools, then you should absolutely test that out. Uh, I say test that out because you might find that there are things you're more comfortable letting your agent do without looping you in and things you're less comfortable with. So, just putting that out there. Um, one very low stakes answer is like if you get emails a lot where people are like, can you put an agenda together for that? You can absolutely have your agent go make a Google doc and fill it with pertinent agenda items. That's pretty low risk. So, that might be a really good one. Uh, but I like this idea of personalizing it, making it sound like you. One option for that Alex shared in the chat is knowledge sources, which if I scroll down, you can see here, knowledge sources are like a document that your agent uses as context. Like if I have a whole document of like emails I've written that I think are really great and they sound just like me, I could put them all in a document. Every time my agent sends an email, it could read that knowledge source and make them sound like me. So let's do it. Let's do that right now. I'm going to say uh before you actually let's put it down at the bottom instead of a step. This is more like a rule. So, I'm gonna say rule. Whenever you draft emails, make sure they actually sound like me. And then here are two options. Okay. If you think you can go grab a few emails you've sent right now and put them in a Google doc, awesome. Then I want you to use the knowledge sources. Okay? That's what I'm going to do right now. If you're like, "No, I'm not. I can't do that. Here's what I would do instead. I would say, look up emails I sent in the past six months to get a good feel for emojis I use, turns of phrase, etc. And you can just have the agent itself go find examples and tailor the way it sounds to you. The benefit of that is over time if you start changing the way you communicate, like you get a new favorite emoji or

### [50:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=3000s) Segment 11 (50:00 - 55:00)

there's a new term that you're using a lot, your job changes, it'll keep learning from you because your agent isn't always going to be the one sending emails. Sometimes you're going to go in and write your own emails. So, just putting that out there. I'm going to try Alex's idea. I'm going to make a quick Google doc. Let's go to drive. google. com. Okay. I'm gonna make a Google doc. New Google doc. And I'm gonna call it Emily's favorite emails. Yeah. And I'm going to grab some of my scent mail. And I'm just going to drop it right into that. I'm gonna drop it right in there. All right, here's one from me. And let's say I put 10 in there. That's fine. I'm just going to start with this because I think that's plenty. I'm going to go back to my agent and down down down down down down. I'm going to write check out the knowledge source that has my old emails in it so you can match my voice, my favorite emojis, use of punctuation and slang etc. Okay. So, there's my rule. Now, I've just got to go down here and you know how earlier we added tools. We used the little slash command. We added tools. By the way, here is my rule. I'm just going to drop it in there. Okay. Uh, now I'm going to scroll down and I'm going to add a knowledge source. And you can see there are a bunch of different types we can use. So many different types. Mine's a Google Doc, but if you had a hard file, like a download, you could upload it. I really like using Google Docs because they stay in sync. Anytime I go add to that with more emails from me, uh it will be something that the agent can access. I don't have to like go recreate it or re-upload it. So, I really like using uh non like hard copy files when possible. Sometimes a hard copy is all you got, though. Okay. So, now we've got a knowledge source. It knows what I sound like. It's either going and looking up emails from me or it's going to read those messages that I have just given it access to in my Google doc, in which case I gave it a very funny single message to reference. Uh, but either way, if you haven't turned your agent on yet, or if you did and then you changed it, it's time to push the publish button. Pushing the publish button means that it is live. The next time an email comes in for you, your agent will wake up because our trigger is new email. Now, I'm going to head back to my deck for just a second because we've got a functioning agent. I'm gonna head back here. I want to know, you just built a Zapier agent from start to finish, but you had about a billion more ideas that we did not get to. You guys had so many ideas. I want to know what you want to build next. What is your next AI automation that your mind is just on fire for? Here's an example from a previous week. Someone shared, "I'm going to build a multi- aent system to get me past the blank page problem when I'm designing a new campaign. " Multi- aent system just means like one agent hands off work to another agent or another agent. An agent is orchestrating more agents. Tell me your ideas. Oo, automate lead verifications. Build Oh, build out a tool for your CEO's inbox. Build out a trading bot with a knowledge source of setups you've already used. I love when people can repurpose stuff they've already built, like documents they've already put together. Repurpose them for their use. That is great. Clean up your old email. Awesome. Build an onboarding system. Oh, we've got some great resources for that, too, on the website. These are fantastic. Okay, great stuff. As you're filling the chat with those ideas, quick reminder for you. Um, we have a really great workshop. I

### [55:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=3300s) Segment 12 (55:00 - 60:00)

mentioned this at the beginning, but for anyone who didn't catch it, we have a special talent acquisition focused workshop series coming up with my incredible colleague Casey. If you're on a people team or your job is about responding to candidates or reviewing resumes, these workshops will help you. Be sure to register. We'll see you there. Thank you, Bryce, for popping up the link. And we have five minutes. I would absolutely love to go through some Q& A. Bryce, I'm going to ask you to join me and give me some questions to answer. — Yes. Hello, Emily. And hello, everyone. Some great questions that came through. Thanks for leading us through that, Emily. Always informative and a great session. We have a couple good questions I think we could probably spend the last four minutes on. Okay. So, the first one is from Ezra and it's a bit more broad. So, I'm interested to hear what you're saying, what she'll say to it. So, the question is, what is one of your favorite success stories from using AI like AI integrations or building AI agents that you have personally helped with? — Oh my gosh. that I have personally helped with. I would say, okay, I've got one. Uh, I want you to imagine someone who is excited to start their own business. They have a lifetime of experience writing and performing and they want to start out like cut a path for themselves and they're like, "AI is not for me. I don't know where to start. " being able to build a system that a looks through first of all agents can access the internet which is amazing. so looks for mentions of them, curates a an RSS feed of like all of the times that articles they've written have been shared and starts to build for them a story about like how people use their work and what their work means to people and then helps them curate their voice and direction to like start their business and really get out there in the world. That's sort of a vague answer, but I don't want to share like personal details without asking. But seeing someone go from like an idea of entrepreneurship to building out an entire business using agents and not losing their own voice and perspective in the process, I think is a very powerful example for me. — Nice. It's awesome. Love it. Okay, this other question is from Britney. She says, "I appreciate all these webinars, workshops. I'm getting confident creating agents for my own email calendar apps. But how do we activate these agents for other members of our team? — H this is such a good question. Okay, forgive me because I'm going to shoehorn us into looking at something uh which is related to the conversation about executive assistance or executive ops because I believe two things to be true. I believe you need a mandate meaning like from the top down you need interest and excitement around AI and I also believe something that is incredibly powerful is looking at exec ops EAS as partners in this space because they already know about helping get technology into the hands of other people like their execs and helping to manage other people's expectations and tool use. So, I'm actually gonna head to our executive assistance page. If you've never seen this, I'm putting it in the chat. Here you go. Uh, the executive assistance page here showcases how our executive assistants use Zapier and you can really get a good feel for how they've helped proliferate it to teams uh like through their exec. There's also really beautiful um success stories of like getting agents adopted at a teamwide level. So, I know this sounds like a pass off, but I really encourage you to check out this page because I think it has really beautiful solutions and tips and you can learn from personal stories like here's a story of one EA getting their whole team to use Zapier as well as templates that you can borrow if you are interested in like using exactly the templates that we use here. So just sharing that. — So again, yes, the AA team here is amazing. There's also that does this give access to that doc that slide deck, Emily, or should I can paste it in the chat. If not, — what slide deck? Oh, oh, I know what you're talking about. Uh, if you can paste it in the chat, that's great. My text is not working.

### [1:00:00](https://www.youtube.com/watch?v=SVu_9bNPLqE&t=3600s) Segment 13 (60:00 - 60:00)

— There we go. So, also check out this deck. A lot of like practical Yeah. use cases from our Exec Ops team that you can look at, too. So, — yeah. — Cool, Emily. Okay. Well, we're at time. — I — So, do you want to say any final words? — I do. I want to just shout out Amanda's comment in the chat as we part the idea of what AI and automation unlock for neurodeivergent folks uh or for folks in any historically marginalized community I think is a beautiful thing to explore and I appreciate you voicing it. Thank you so much everyone for spending the time with us today. Bryce and I never take it for granted and you have been amazing participants way to be full of ideas. Have a great rest of your day. — All right. Thanks Emily. Thanks everyone. See you soon. Bye.

---
*Источник: https://ekstraktznaniy.ru/video/50743*