#2289 Building companies using Zapier’s AI automations

0:00Zapier's AI automations are so good people are creating whole software companies on top of them And it's so easy to do. Well, the founder is going to show you how easy. Coming up. Wade Foster is the founder of Zapier, the AI automation company.

0:18Why don't we start with this? I feel like everyone's going after HVAC and roofers these days in tech, you know? You have people who use Zapier to create their whole businesses. Give me an example of somebody. Well, it's funny you say roofers. We have one of our solutions partners, experts. He started as a roofer. So he was running small-time business, local roofing business. Hale would come in. People would call him up and be like, I need a new roof. And he'd go out and get him a new roof. And so that's what he did for a living. And as per roofers go, it turns out he's a little more techie, a little more nerdy than your average roofer.

1:01and he had set up a bunch of like technology to help him manage his business so he had uh you know he was running ads on facebook and google to like target local stuff he you know had a crm to collect his leads he would send out um you know invoices so people could pay via credit card versus like writing a check uh you know or handing cash over like you know stuff like that that you know is is pretty modern for what you might expect a roofer to do and you know one of these days he was like, you know what? I don't know that I want to be out on a roof all day, every day for the rest of my career, but maybe I could help my fellow roofers out with all this back office stuff.

1:38And so he set up a roofing automation agency where he works with roofers and basically gives them the playbook for like, here's how you can manage your back office way more efficiently so you can take on more clients and you can spend more time out on the roof, fixing more roofs, getting more billable hours. How does Zapier fit in? What are some of the zaps that he creates for them. And essentially what he's doing is he's selling zaps, right? Yeah, more or less. So, you know, if you think about how a local roofing, you know, business might work is first they got to get leads. Like, you know, when you, when you need your roof to get fixed, like, well, what do you do?

2:15You probably go to Google and be like, who fixes roofs in, you know, where you're in Austin, Texas. Yeah. Yeah. Who fixes roofs in Austin, Texas. And then, you know, list comes up and you're like trying to find these things. So a big part of a roofing business is like, well, I got to show up where people are looking for me. And so they might run ads on like Facebook or Google that are like geo-targeted and then drive them to a page or use like their lead form ads. And then there, that's where Zapier comes in. So we'll take the lead as it comes through Google or as it comes through Facebook, or if you have your own landing pages and have forms on your landing pages, and then they'll start to send them through an automated system.

2:53And so you can start to send out follow-up emails, follow-up texts with scheduling links, with payment forms, all this stuff that if you weren't doing automation, what that roofer would have to do is at the end of the day, they're hot, they're sweaty, they're like ready to go home and eat and have a shower. Instead of doing that, they got to sit down at the computer and go, okay, like, you know, how do I, do I need to call these leads? What do I got to do with these? Like a very manual, very bespoke process. That's not actually where they make their money. They make their money fixing roofs, not doing this stuff.

3:25But if you put automation in, you can create a much better customer experience, a much more responsive experience. And Zapier helps all these roofers do that effectively. I've seen whole agencies built on nothing but Zaps. They take on a client, they create a Zap, and the Zap now is getting smarter and smarter. In fact, to call it a single Zap is just oversimplifying it. Like what are the smart things that it's doing now because of AI that weren't possible maybe a year ago? The biggest new capabilities you get with AI are two things. One, you can add AI into workflows. And what that gives you is the ability to work with unstructured data.

4:04Now, unstructured data allows you to do some very specific stuff. One, you know, in the old school world, you might have an email template that was using a mail merge. Well, mail merges are like sort of personalized, but not really personalized. I sort of pretend like I know you. But with AI, you can actually go do deep research on somebody. So I can say, hey, tell me about Andrew and Mixergy and everything he's trying to do. And based on everything you go learn, now I want you to go write a specific email to Andrew using these key points and translate it to him in a way that he cares about. So instead of using those like old school vanilla mail merge, I'm able to actually generate a very custom email that is very specific to everything I know about you.

4:46That's something you couldn't have done before. So that's a great example of it. Another great example is you might have a business where you get a lot of customer feedback. Like maybe you've got a bunch of support tickets coming in or maybe you've got a customer feedback form. Well, in the past, you might want to go look through all that feedback and try and understand, you know, what are the common complaints I have? How do I go generate an FAQ? How do you all do all this stuff? Well, that required a human to sit down and like read all of them, categorize them in a particular way. And it was hard to do in a workflow.

5:18Like you just couldn't do it because the data is very unstructured. It's very messy. Again, this is where AI thrives. So you can send a big old block of text through a workflow to an AI and say, hey, what I want you to do is put it into a theme. Here's like five themes that could come in. I want you to categorize it. And then if it's categorized a particular way, I actually want you to go route it to this specific person on the team who specializes in these issues. And so they're going to be the rep that follows up with it versus having to do some sort of round robin system or something like that.

5:50So these types of systems allow you, AI allows you to put brains in your business in places that was really hard to put brains in the past. It was very manual, very expensive. And mostly, especially for small business, you're just like, I'm just not going to do it. Like, I just can't do it. It's just not possible. And so that's where AI is like so exciting. Yeah, I hear you. Even when you think about the mail merge, the fact that you can have a text message or an email fired off that says, Andrew, in your part of Austin, the freeze is so bad that you're going to have trees falling down on your roof.

6:22And that little bit of understanding is like, ouch, I do remember that. That literally happened two years ago. I need something to protect us. Okay, I get it. Now let's talk about what else is out there and then we'll go into what we could do with Zapier. Because I'm on LinkedIn lately. I don't love it, but I'm learning to like it. Every freaking guy on LinkedIn ends up getting a bunch of likes or hearts or whatever it is that they measure success with on that platform by saying, ChatGPT or OpenAI just crush Zapier. Let's take a look at what they announced with their agents. And then let's talk about how it differs.

7:00And then I'd like to actually see if we can build something together with you on Zapier. Cool, let's do it. But let's start by looking at what's on OpenAI. OpenAI launched their agent builder like two weeks ago. It's a pretty cool product. I am actually pretty excited about what's possible here. And, you know, I think it is pretty different than Zapier. While it looks like a workflow builder, under the hood, what it really allows you to do is build better chat agents. So you build a chat agent. So what is a chat agent? And how is that different from like an automation or something on Zapper? Well, a chat agent is triggered by a message to ChatGPT.

7:39So you literally are sitting at your keyboard and saying, hey, I want a blah, blah, blah, blah, blah, blah, blah. And then instead of going to the general purpose ChatGPT, you know, LLM that's like answering and responding, it'll send it to this agent that has a specific set of instructions. And that specific set of instructions will help give a better answer to the person sitting there at the chat. So that's a pretty different experience. Whereas on Zapier, what you're trying to do is you can wake up your agent in any different way. You can have a type form come in. You can have a new email come in.

8:15You can have a calendar event. You can have a new customer on Stripe. You can have all these other things that would wake your agent up. And then you can have logic that encompasses any of the 8 ,000 different tools we work with. Whereas this is really about giving very specific prompts to ChatGPT to make it answer you better. And so if you do that, you can see like start. Well, what is start? Start means talk to ChatGPT. It means literally type a message. There's no other way to make this start. and then inside your agent you can start to give instructions for it so this would be like an ai step inside of zapier where you'd say hey you know i want you to um you know let's see uh let's call it like sentiment analyzer right uh you know tell me if the message is positive or negative right so now you've got just like a very simple like uh agent here that's job it is is to assess the message coming in they're just like trying to figure out if you're a happy person or a sad person something like if we were if we were to create like a brief generator on my upcoming meetings it would be me starting a chat that would be the start in the chat i would put a screenshot of my upcoming meeting or i would ask it what's helped me prepare for my upcoming meeting Yeah, you would give the chat a bunch of context.

9:36Yeah, go ahead. And then it would be able to pull in my calendar, because I think it can. And then it would do some research. And then it would come back and say, Andrew, based on what you like for your prep for a meeting, here is what I found about this person that you just chatted me that you want information on. Totally. So you could say, let's call this the brief analyzer agent, right? So you would say, you know, write a concise brief based on the info you get from the chat.

10:11And, you know, use bullet points less than five, you know, blah, blah, blah, blah, blah, blah, blah. Right. So on and so forth. Exactly the way you'd want it. But here's the kicker. This is where actually Agent Builder and Zapier get really interesting. So there's a thing called MCP. Now, MCP might sound like this kind of techie concept. It stands for Model Context Protocol. But what's really powerful with Agent Builder is you can now call tools. Now, again, tools might sound kind of techie, but really all this is is connecting to apps. Think of all the apps you use. and if you wanted this agent to have access to your information you probably want to go use mcp to do it and here is where you can see open ai has a handful of apps that you can connect so your gmail your calendar your outlook your outlook calendar your google drive etc kind of some of the standard stuff but then they've got third-party servers and there's not that many but right here you've got zapier yeah and so zapier you can go hook in and now with agent builder the thing that you're kind of that you really want is I want access to all of my apps not just you know the the you know eight here that chat GPT offers out of the box I want everything so now you can use our MCP server to give your agent more ways to respond so ultimately that's how you know agent builder and Zapier are kind of like peanut butter and jelly in fact I would wager if you're not using agent builder with Zap, you're, you're, you're probably using like a much, um, it's, it's basically just a vanilla chat chatbot builder is what this is.

11:53Got it. Right. So if, for example, I wanted to look up a customer who's using, who is in my CRM, which might be close. I can't do that out of the box, but here I could say, create an agent in chat GPT's agent tool that also prepares me for a meeting, but looks up anyone else in this company that I've talked with and look it up in my close account, not on the internet. Got it. And that's where Zapier comes in and is helpful. And a hundred percent, but I don't even need, um, I don't even need their agent builder to do this. I can do this just directly in Zapier. And in Zapier, the nice thing is I don't have to go to the chat to trigger it.

12:30I can do things like press a button on my Chrome browser that triggers it. I could send a text to trigger it. I could do anything. It could just be time of day. It could be when this, when this meeting's about to come up, that kind of a Okay, so basically, to all the people who've said that they've killed you, congratulations on getting all the likes and the hearts and the attention. We're clearly not there. And if anything, it's actually sending people over to Zapier. But it's a nice way to introduce this new tool and tell your followers about it. Now, let's see the comparable experience. Like if you and I were going to build this type of brief in Zapier, what would that look like?

13:06So here we've got the Zapier Agent Builder. Okay. Now, this is a little different. So instead of having the nodes in drag and drop, what you can do with Zapier's workflow builder, with the Zapier agent, you're just describing what do you want the agent to do in plain language. And the nice thing is, I know a lot of folks struggle with thinking through all the detailed instructions that agents really benefit from. One of the things you can do with the agent is just describe it like you would be talking to a friend. And you wouldn't give a friend all the nitty-gritty details. And so you might just say, make me a meeting brief agent that, you know, gives me some details before my meetings.

13:56So this is a pretty bad prompt. Like, this is not what agents generally prefer to get. They like detailed instructions. This is not detailed instructions at all. But we're going to start with anyway and see what happens. And for people who are listening, you just press start. and now it's doing it. And so now you've got this co-pilot that's over here thinking and trying to figure out, okay, what exactly are we trying to do here? And it's like, hey, I'm happy to help you with this. To build this effectively, I'm going to need a little more details. So what calendar system do you use? So I use Google Calendar.

14:29Let's use this. And then can you send the meeting brief, let's do four hours before the meeting. So that way I know I've got the brief ahead of time. What do I want included? So let's see. Include details about attendees. I actually only want it to do only give me a brief for external attendees. I don't care to get a brief for people at Zapier. So only give me a brief for external attendees. and I'll just say ignore internal attendees.

15:13And then let's see, previous meeting notes or context. You know what might be helpful? Can you, let's see, use any prior email correspondence I've had with the attendee? So let's see if it can do that for me. And this, by the way, is the copilot's on the left and what it's doing is it's saying, look, before I create this on the right, I have some questions for you that I want you to think through. And when you're referring to all these questions, that's where it's coming in. You're right. You've got it. So send me the meeting brief and email. So that's what I want it to do. But you could send it in Slack.

15:50You could send it, you know, whatever other places. Where should the agent pull information from? CRM, previous emails, company database, LinkedIn. Oh, interesting. I didn't answer that. So let's see. Use, you know, internet research about the person. Make sure to include a LinkedIn profile. We'll just do that for now. Let's see what happens. Okay, so we're going to give it some more instructions because it said like, hey, I need more details to make this agent a little bit better. And you can see I'm not like, I'm kind of sloppy about this. I'm not even, there's a lot of typos. There's, you know, I'm not exactly being super professional about how I go about this.

16:35And you can see it's over here. It's starting to build it. And you can start to see your agent start to kick in over here on the right. And when you read through what it's starting to do, you're like, oh, that's better. You know, when a Google Calendar event starts in four hours, generate a comprehensive meeting brief for external attendees. So extract meeting details, including time, title, location, all attendees. Identify external attendees by filtering out internal company email addresses for each external attendee. You know, search through prior email correspondence. So, like, it's fleshing this out in a way that an agent is going to be like, ooh, that's good instructions.

17:08Like, this is the kind of stuff that you would put in an actual standing, like, you know, operating procedure. Like if you were trying to hire an employee, you would much rather give them instructions like this versus what I gave the copilot. I gave the copilot bad instructions. It's like this is not all that good. But over here, we're starting to get something that you're like, okay, like someone could follow this. So this is the cool thing about using – this is the other place where AI is really exciting is it's not just helpful as a tool inside your workflows. it's also really helpful for building workflows because building workflows is kind of hard like not everyone thinks in systems thinks in structure and so what we've tried to do with copilot is make it so that you know when you're you're just like have a vague idea of what you want i just kind of want a meeting brief thing like you can just tell it that and then it will coach you through uh how to go build all that stuff i and one of the things i told you before we got started that i really like about the software is it's not just giving instructions that make sense.

18:12It's also written in a way that's highly scannable. Like you look through it and this feels like a blog post almost, but with just the facts. And so it starts off with step number one, extract meeting details, including time, title, location, and all attendees. Step number two, identify external attendees by filtering out internal company email addresses. Step number three, for each external attendee, then there's a bullet point, bullet point, bullet point. Then it goes on to step number four and so on. And this is so easy to read that now I can go back in and edit it if I want to change it. So I might say for me, external means all email addresses that end in one of these three domains.

18:50Sorry, internal is all email addresses that end in one of these domains. Okay, I got it. I see it. Then we hit agent. We can go in and edit it in the instructions. Then we hit preview. Yeah, you got it. And so you can see up here, Google Calendar, it's a new event. So it's set up the trigger for us. Good call. I didn't realize that. This is like it's saying, this is what I know is going to kick this off every time I see a new event. Now you can go in and change it and say minutes before the new event, I want you to do it. Yeah. And actually, you know, I can see it here. I do think it's made a mistake.

19:23So this thing is not perfect. You're going to have to pay attention to this. You know, in this case, I actually think I want it to be Google Calendar. We want it to be when an event starts. And we actually want it to be, let's do my own calendar. And I actually want it to be four hours. So I made a mistake, right? That is what you asked it for, right, right. But, you know, making the change was like actually not that complicated. Like I just clicked a couple buttons there. And then as I read through it, most of the rest of this stuff looks pretty good. Let's go see what it did on the Gmail tool here.

20:00So it's basically going to go look inside of Gmail and then it's going to guess. I actually want it to guess. Like just go figure out, you know, if I've talked to Andrew before, basically, and tell me, you know, if there's anything in those emails that I need to know before this meeting, just include it. So I'm happy with this. You know, what else is interesting? You know, this is probably the one place where I might make some changes, you know, meeting overview, external attendees, prior correspondent, LinkedIn profile. You know, I might say like, well, here's like the style of a brief that I like.

20:32And so I might give it an example of like, you know, can you make it look kind of like this and structured this way, et cetera? Maybe somebody on my team wrote like a really nice brief once and I'd be like, follow this format for me. But by and large, like I'm pretty happy with it. And so you can go ahead and click, you know, the agent preview and it will go do, you know, a test for us here. And so it's going to go take a look at it in my calendar and we'll see if I don't accidentally leak PII here. so oh it actually pulled in the Mixer G1 so that's great so yeah what it's going to say I'm going to see what's going on here with Andrew everyone knows your email address now Andrew I'm good with it they all know yours too so it takes a minute to kind of test all this stuff oh this is the part that's going to make me nervous because it's going to pull up all our past email interactions yeah so we'll see you might have to blur out what pops up on the screen here I can't imagine there's anything that's private it.

21:29Yeah. It's probably just like, Hey, how are you? I think a bunch of it is just me showing off about how I have one of the, actually the first account on Zapier. And as a result, I still have all these features in my free account that I refuse to let go of. Um, because it's the classic account, even though we pay for a real account for our company, I still hold onto that. Yeah. Yeah. That is a little known fact is you were the first external user of Zapier. so for 12 years no 13 years ago now let's also say customer I paid on your way to the bank customer you did pay it's a wild story alright it's going did it complete it ok it found it so it looks like it found a bunch of emails etc so it's still thinking though so you can see up here we probably need to make this more obvious that it's thinking still that is the one challenge with agents versus workflows workflows are fast they're fast they're reliable they're consistent agents the llm's thinking you know it's like trying to work its way through this it's trying to follow the instructions um the more tools you give it the harder it thinks the more mistakes it's going to make so these are things that like as you start to get used to building agents you start to learn a little bit more about how to use them um yeah i find with agents it's easier to get started than with workflows but you run into more like, I call them like potholes.

22:55It's like not major problems, but they're things that like, it's not so good. Whereas with workflows, they're a little harder to get started, but they work. They're like, they're rock solid, very consistent, very reliable. Because workflows are just, you tell it what the variables are. You tell it what to do. You step-by-step guide them. All right, while it's doing that - It's not having to think, it's just going and doing the thing. While it's doing that, let me ask you a couple of questions about the business. How many customers do you have now? uh we're like 350 000 customers or so wow and revenue wow we would nice try okay hundreds of millions of dollars based on that we can say sure yeah we'll go with that i you didn't take on much funding right we did a seed round that was uh 1.2 million in 2012 Did you touch any or you hardly touched any of it?

23:47We treated it like the last money we were going to spend. We did burn money for about 12 months. We were burning money. The most money we lost in a month was about$20 ,000, which at the time was terrifying. But now I hear the stories of some of these companies that are literally burning a billion dollars a month. And I'm just like, oh, my gosh, that's insane. No, this is nothing. I mean, even smaller companies now burn a lot. Though, actually, I was just talking to Gary Tan, Y Combinator. He says the revenues are growing phenomenally for their companies. Yeah, I mean, there's a lot that are just doing really, really well.

24:27So, yeah, it's impressive how fast these companies are growing these days. How much are you going to continue to depend on outside LLMs for your thinking, for the AI internally that you're using? Yeah, I think, you know, we will probably always depend on, you know, external LLMs for large parts of this. I think these large language models, these frontier models, they're pushing the envelope. They're at the cutting edge. So it's really useful for getting, you know, a rock solid model that can do most things well. We'll see if we start to like sprinkle in, you know, our own like custom smaller models that are like really hyper targeted at specific tasks that maybe the large language models are just not quite as specialized at.

25:17So that's that's an area we're considering. But you're not doing that now. I kind of assumed that you were that you were taking on the smaller stuff for yourselves. Okay. All right. Here we are on our agent. So you can see it's gone through and created a brief based on this upcoming meeting I have with Zapier on Mixergy. And this is the email that it wants to send. And so you can see it generated this email. Of course, it's in HTML, so it's a little tough to read. But you can see it's generated one. And we're going to just go ahead and approve it. And so this will send me an email. And if you give me a second, I can change screens.

25:56And we can actually see it. we can actually go read what gets sent. While you're doing that, if I were doing it myself, I would probably say, I don't want any HTML. I just want you to do nothing but plain text, right? Yeah, you could. Let's go see what it does, though. Fair, fair. So this is the email it sent me. Meeting brief. Zapier on Mixer G. Here's the details of it. And then here's stuff about Andrew. So backgrounds and achievements, notable interviews, education, current focus, expertise, recent activity. Whoa! correspondence. Wait, that's interesting. It actually knows that I'm doing the series of AI software founder interviews.

26:33Okay. There you go. Yep. Key context from recent communications. You can see the outreach, the scheduling process, interview focus that you want to spend time on, topics that Andrew may cover, Andrew's interview style, action items, et cetera. Wait, I like that. Andrew's interview style, known for thorough preparation and fact-checking, focuses on actionable insights for entrepreneurs, experience with 2000 founder interviews, likely to ask about specific metrics, strategies and lessons, which is exactly what I did here. It nailed you, right? It really nailed me, yeah. It's done it perfectly. So you can see like we built this in what, five minutes?

27:10Yeah. And this is what you get with five minutes of effort. Now I can look at this and also go like, well, you know what? I don't like this part or I wish it did more. Maybe I wish it had more context on, I don't know. Like maybe you're education. I really care about you, what you did in New York University or something like that. And so I could go back and edit the prompt and be like, you know, pay a special attention to these things and less about this or make it shorter or make it longer. So you could give it more guidance. But, you know, even just out of the box with like a super basic thing.

27:42Yeah. You know, this is this is to me, I think, what is so exciting about AI is you get pretty good stuff pretty dang fast. What's interesting is it didn't get any of the past email interactions that we had because I used a different email address to book it. And it would have been interesting to see what came up over the years, but I'm glad of this. All right, we'll go on to the next thing, which is the agent that qualifies leads. Do you want to look at one of those? Yeah, let's do that. Do you want to team me up or do you want me to ask it or just dive in? You know what? I guess what I want to know before we start diving into what this is is who cares?