Stop Doing Boring Work Yourself…

- July 28, 2025 (7 months ago) • 49:22

Transcript

Start Time	Speaker	Text
00:00	Sam Parr	Alright, this is super cool. This type of stuff only happens, I think, on MFM. But here's the deal: I have a friend named Wade Foster. He started a company called Zapier. Zapier is a business that was bootstrapped to hundreds of millions in revenue, and it's worth about $5 or $10 billion. I had the CEO and founder, Wade, on MFM and I said, "Look, share your screen and show me how you're using AI to save 10, 20, 30 hours a week." > "Look, share your screen and show me how you're using AI to save 10, 20, 30 hours a week." It's pretty amazing — he did it. He broke down three or four different ways that he has automated parts of his life and his business. It's really, really cool because this guy has an incredible perspective: he runs such a large company and he helped create this whole automation industry. So if you want to save a bunch of time, this stuff is not complicated. I don't know AI that well, and I'm going to implement a ton of the stuff that I've just learned. By the way, if you're listening to this with audio only, you might want to go to Spotify (video) or YouTube to watch — it has a lot of visuals. You can still get a ton of value on audio, but it's just better, I think, on YouTube. So give it a watch, give it a listen, and let me know if you dug it. Talk soon. I'm thinking about how to introduce you. Basically, I met you in 2016 or '17, something like that, when you spoke at one of my events. Then I asked you to speak at one of my events because you founded and ran a company called Zapier. You guys took off before AI — way before it was ever a thing. If I remember correctly, you only raised, at the time, $1 million, but you got it to nine figures in revenue and then eventually raised a Series A at something like a $5 billion valuation. Is that about right?
01:45	Wade Foster	Close. So, the seed round was $1.2, and then the $5,000,000,000 number — that wasn't; that was all secondary. So none of that stuff went on to the balance sheet of the company. </FormattedResponse>
01:56	Sam Parr	How much did you guys raise? Was that public?
01:58	Wade Foster	It wasn't—we didn't raise anything.
02:01	Sam Parr	Well, you took money.
02:02	Wade Foster	Early investors and early employees got some liquidity from it, but nothing came to the company.
02:07	Sam Parr	How big are you guys now?
02:09	Wade Foster	What is the public number? I don't know. The public number — nine figures is public, though.
02:12	Sam Parr	Nine — you're at nine figures in revenue, and it's one of my favorite companies. I never thought in a million years I'd be fired up about such a dorky product: connecting APIs, but basically my entire company — and everyone I know — runs on Zapier.
02:29	Wade Foster	Well, it's wild. The first decade of existence—that "dorky company" thing—was 100% true: no one cared about workflow automation. We thought it was cool, and there were definitely people who did, but Silicon Valley wisdom was to raise a bunch of money, throw bodies at the problem, blitzscale, go nuts. So people would say they cared about automation, but their actions often betrayed them, especially in tech. We were kind of out here doing our thing; there weren't a lot of people in our space. Now, you get into the AI world and it's totally inverted: everyone cares about automation, everyone cares about AI. The market potential for us has just ballooned enormously, but of course there are a lot more players in the space now. </FormattedResponse>
03:24	Sam Parr	Yeah. At the time—when you guys were just getting started—it was basically you and maybe one other company doing this. Mm-hmm. Now there are so many more.</FormattedResponse>
03:32	Wade Foster	Yeah. And, you know, everybody has their angle—their niche: this vertical, that vertical. There are more dev-centric approaches and less dev-centric ones. Everyone's kind of trying to get a piece of this: what is going to be a $1 trillion opportunity?
03:47	Sam Parr	"In ten years, how big do you think your company will be in terms of revenue?" </FormattedResponse>
03:52	Wade Foster	Ten years? Well, shoot — I think we should be well past a billion in ARR if we do our jobs right. </FormattedResponse>
04:00	Sam Parr	Dang — and you think you will still not have raised the money?
04:06	Wade Foster	"I don't think we will."
04:08	Sam Parr	So, you're going to be one of the largest—one of the larger companies ever that only... I mean, I don't know if you still say "bootstrapped," but you only raise... you turn $1 million into something worth tens of billions of dollars. I mean, that's gotta be one of the more efficient stories ever.
04:23	Wade Foster	Yeah — this is how they used to do it. If you look at Microsoft's fundraising trajectory and Amazon's fundraising trajectory, those companies obviously went public much earlier in their life cycles, but they didn't raise huge gobs of money.
04:40	Sam Parr	"Yeah. How much did Google raise? I think they raised, like, low tens of millions of dollars."
04:45	Wade Foster	Yeah, I think it was something like that. Google had a big round — I think it was their Series A or Series B [Series A or Series B] — that was huge, and people were like, "Whoa, this is nuts." Nowadays, the AI companies, the foundation-model companies, will do a $1 billion Series A.
05:01	Sam Parr	Yeah, it's like...
05:02	Wade Foster	Woah — this is different. So, this is different.
05:04	Sam Parr	I want to ask you about all of that. But I've noticed that I have better conversations with people while we are doing stuff. That "doing stuff" is this thing where we basically ask people like you — people who run big companies — how many employees do you have?
05:25	Wade Foster	700 and change.</FormattedResponse>
05:26	Sam Parr	Okay, so you have 700 employees, and you're trying to get everyone to use AI at your company. You guys are an AI company — you were kind of an AI company a little bit before that was really popular. The thing that I want to do, and that I've loved doing lately, is I want you to share your screen and show me how you're using AI in really practical use cases. You said that sounds good, and so while you're showing the screen I might ask you questions about the background of the company and things like that. So I asked you to make a list. You have a list. One of them stuck out, right — you want to tell me what that was?
05:57	Wade Foster	Well, the first one is a pretty basic tool I use day to day: an instant dossier creator. You can use it for all sorts of things. It's particularly handy when you're out and about — for example, when you're going to a dinner or an event and you have a list of names, or there are name tags everywhere and you're wondering, "Who are these people?" In that use case I usually feed Claude [the AI assistant] some details about the person. Claude then returns a quick dossier that includes public details about the person, as well as contextual information: are they a customer, is there anything in our HubSpot account, or can ZoomInfo tell me anything useful? It gives you quick hitters to help you think, "Is there something I can talk to this person about?" It's kind of like, if you've watched Veep, when Selina has Gary in her ear telling her things — you get your own little version of that. </FormattedResponse>
06:57	Sam Parr	And so, do you do a lot of these dinners with customers and potential investors, things like that?
07:02	Wade Foster	The cool thing about this one is you can use it in all sorts of ways. If you're just sitting in your home office and you've got a string of meetings coming up, you can use it for all the people you're about to meet that day. If you've got leads coming in on your website and you want to enrich those and send them over to your sales reps, you can do the same. This process works and is applicable in many different scenarios. Mostly, what you're doing is amending how you utilize it based on the situation you're in. I can maybe do a screen share on this.
07:39	Sam Parr	Yeah.
07:40	Wade Foster	You know, Claude... the thing I love about Claude is Claude has tools connected. The Zapier MCP server is available here, and you can see—if you click into here—you can connect all these different things. With Zapier, you have access to 8,000 different tools, so I can just turn on all sorts of different tools to use within Claude. For example, you can see here we've got eight tools turned on. Basically, these tools are a series of HubSpot capabilities: finding contact information, finding company information, finding deal information, and ZoomInfo endpoints—like finding contact information and finding company information. The other thing you can do inside of Claude is you can make projects. I don't know if you've used projects before...
08:32	Sam Parr	I've used it with ChatGPT. It's like: "I have a health project, I have a therapist project, I have a business project."
08:37	Wade Foster	You can assign these tools to projects, or you can give those projects certain system prompts to help them perform specific tasks. Within one of my projects, I basically coach it on how to handle contact information for me. I'm like, "Hey, make these dossiers for myself so we can actually use them." I actually made a quick one for us, which is this My First Million demo. I have a buddy who said it would be cool if I use their name on this stuff. So, tell me more about Lars from Social Puppy.
09:18	Sam Parr	"And so, what was in that file that you uploaded?"
09:22	Wade Foster	So, I didn't upload any files. Basically, what this chat has access to is: - The Zapier HubSpot account. - Our ZoomInfo account. - My system instructions, which basically say: > "Step one: look up the contact and company inside of HubSpot. > Step two: look up information inside of ZoomInfo. > Step three: go find anything on the web to do this." You might have seen some errors popping up here. This is one of the tricky things about MCP and Claude right now — it's still a little buggy. MCP (and their tool stuff) has been out for maybe two or three months, so it mostly works, but there are occasional issues. </FormattedResponse>
10:13	Sam Parr	Hold up—just a really quick break. I know you're watching this and thinking, "This is a ton of information." Our producer, Ari, just listened to the entire episode and wrote out all the important stuff. If you want to implement a lot of these workflows—or whatever you want to call them—she wrote them all out. You can use the QR code on the screen [QR code on screen], or there's a link below in the description you can click. You should still watch and listen to the rest of the episode, but the thing she made is a PDF and it has everything written out. You can just click and link off to all the important, great stuff. Alright, back to the episode: What does MPC stand for, and what is MPC?
10:51	Wade Foster	Yeah, MCP is called Model-Context-Protocol. Effectively, this is a way for agents to talk to data. In the old world we would have APIs where you would say, "Hey, I want to talk to the Gmail API," or "I want to talk to the HubSpot API," or the Slack API. These were ways for SaaS companies to talk back and forth to each other. That protocol works really well for the SaaS tools. With agents, they don't exactly know how to utilize all those API endpoints. So MCP is basically this layer that sits in the middle, helps them find different tools they can use, and then take advantage of them.
11:42	Sam Parr	Got.
11:42	Wade Foster	Okay. Right here you can see this one is—oh, it worked. But the cool thing is you can see it's iterating over a bunch of different endpoints. It's saying, "Hey, this didn't work; let me try a more direct approach." "Okay, that didn't work." So this is Claude trying to figure out how to do this task. They're like, "Hey... I'm not exactly sure. I've got access to a bunch of different tools. Some of these things might work; some may not." And then, at the end, it'll spit out information. So it pulls up a LinkedIn URL. You can see that Lars is a Product Manager at Zapier. He's in Vancouver. You can see all the different things that we know about him. Then Social Puppy happens to be his side gig — it's a way for people who own dogs to do meetups together. Of course he's not paying for HubSpot, so they took a look at HubSpot and were like, "Yeah, there's no deal associated with this for Lars."
12:40	Sam Parr	"See what it says." "It said, 'Colleagues describe him—'" "Where—wait, what was that? 'Colleagues describe him as...'" "What does it say?" "Yeah: 'as incredible drive and enthusiasm for the goals.'"
12:47	Wade Foster	As having incredible drive and enthusiasm for the goals. Yep.
12:51	Sam Parr	Okay, and he's known for being fun and energetic. That's cool. Okay.
12:56	Wade Foster	And I would say that that's "act knowing Mars." I would say that's accurate—yeah.</FormattedResponse>
13:02	Sam Parr	So, when you're going to—let's say you have a work dinner. Let's say you're flying to a different city and you want to meet with about ten customers and host a cool hang to get to know them. Let's say they're big customers. Would you just upload all six of the names and then tell it, "Make this in note card format"? Or... how would you do it?
13:22	Wade Foster	Well, okay. I want to show off something separate. If I was doing that, I would actually—this is an internal tool that we built. Now, the thing I was just showing off is kind of for basic, on-demand stuff. You can see it did a good job, but if I'm trying to show up to a dinner and I know who's going to be there in advance, I want to come in well equipped. I don't want to do this sort of on-the-fly thing. I want to come in having done my research. So we built an internal tool — this is a Company Brief Generator. It now uses Zapier Interfaces, so this is a pretty easy thing to build. What it does is you put in a domain name. You could put in Netflix, Shopify, Slack, or whoever — you're meeting somebody from this company — and then it goes and retrieves information from three sources. One: it pulls from a web search. It goes and finds what we know about this company based on what's on the internet. Two: it does a Glean search. We use this tool called Glean internally, which is like an internal Google, you can think of it. It'll go search all your internal stuff.
14:41	Sam Parr	Glean is basically—I've been trying to use it. You just log in to everything, and then it makes... yeah, okay.
14:48	Wade Foster	And so, it'll search Slack, Google Drive, and all the tools you give it access to. It'll do a clean search. Then the third thing it will do is hit our actual customer database — well, a replica of it in Databricks — and help understand what usage is going on inside the company. On the other side, it spits out a report. I actually did Hampton before this call.
15:16	Sam Parr	"Oh my God — let me see."
15:17	Wade Foster	Alright, you wanna...
15:18	Sam Parr	See it. Who's going to be using this? And, at your company, what roles are using this?
15:22	Wade Foster	Everybody uses this internally. So, if you're customer-facing at all, you'll do this. It doesn't have a crazy amount of information on Hampty, but you can see here the company summary. This is pulling a bunch of public data on it. Now, I did some research on it, and it's sort of correct... I'm betting you're spotting things here that you're like...
15:49	Sam Parr	Yeah.
15:49	Wade Foster	Oh, that's not quite right.
15:50	Sam Parr	I think the goal of having this is—if your intention is to connect with me at a dinner or something like that, you have ammo there.
15:57	Wade Foster	Exactly — that's what I'm getting out of this. I'm not, like... I looked up [the company] before, and it doesn't look like Jordan is the CEO anymore, right? So that's something I'd probably avoid talking about. There is, however, a membership community in New York. Those are things I can riff with you on. It pulls up a bunch of information, and you can see our Glean summary, which lists more issues about what's going on with the account and things like that. So maybe there was a payment issue with you all at some point in time. </FormattedResponse>
16:29	Sam Parr	That's so awesome.
16:31	Wade Foster	Yeah. I can be prepared where I'm like, if you come up to me and you're like, "Wade, you really screwed me over in 2024," I can say, "Yeah—hey, I'm really sorry about that. Here's what we did to fix this: we did this, this, this, and this." All of a sudden you're like, "Okay, Wade's really on top of it; I feel better about Zapier as a vendor." Versus, if I didn't have this and you were just like, "Wade, I'm still pissed about what happened in December 2024," and I'm like, "What happened? I don't know what happened," now you're like, "Man, he doesn't even know what's going on in his own company." So having a tool like that matters. Lastly, it didn't actually pull much, so I don't know how much usage is going on here. But for companies that are really using Zapier, the visualizations area pulls up a whole bunch of metadata about how their account is being used. I can see things like: they use Slack, HubSpot, and Google; they're using AI or not using AI in the account; here are the users that are using it— all that sort of stuff.
17:31	Sam Parr	Well, who built this? I think...
17:32	Wade Foster	It was one of our data analysts who built this tool.
17:36	Sam Parr	On, like, Replit or something.</FormattedResponse>
17:39	Wade Foster	This is mostly just a zap — it's a Zapier interface. You put in a domain name, then the zap goes out and hits three areas. - Step one: it does a web search. - Step two: it does a Glean search. - Step three: it does an internal search on our Databricks instance. The output is an HTML file. </FormattedResponse>
18:03	Sam Parr	"What would I Google to find the landing page that allows me to make this?"
18:08	Wade Foster	So we have a bunch of templates that you can check out: zapper.com/template/use-cases. This has a bunch of places where you can go to figure out how to do versions of this for yourself. Anything under Lead Management will give you great examples for ways to build a version of this for yourself. The second place you can check out is a product called "Agents", which is in beta. It has its own templates, which are pretty slick as well. Company research is the number one thing here, so you can go build a version of this to help you with company research.
18:47	Sam Parr	Have you seen The Office? They show—like, I guess—Michael stole Dwight's notes on his customers, but he forgot how to... like, he didn't know how to read them. He's like, "Oh, I see you are Eric from PA, right?" And mhmm, "You have a gay son, right?" And he's like...
19:05	Wade Foster	Well, they're, like, weird. It's like green is color-coded—green is color-coded to mean this, to mean this, which means "stop."
19:11	Sam Parr	You're just... Oh man. Alright, what's the second one? Because you have a bunch here. </FormattedResponse>
19:16	Wade Foster	Yeah. So we could go to the "agents manning inbox" one, or there's another, even more basic one. Like — how basic do you want to get? </FormattedResponse>
19:22	Sam Parr	You drive, baby. I'm basic. I'm pretty... I'm a—I'm a—I'm a huge noob when it comes to this. </FormattedResponse>
19:29	Wade Foster	Alright, so here's one that I do all the time. What's this use case? If you run a company, you probably get a Google Doc—like a business review or a strategy memo—shared with you all the time, and you're trying to figure out what's going on in it. The use case I like to do is to just go back and forth with ChatGPT to try and better understand the document. I didn't want to share our own internal stuff, so I actually pulled one externally. Some of you might have seen this: MrBeast had this leaked memo about how they do their production stuff, so I figured we could just...
20:14	Sam Parr	Try it for the viewer: about a year ago, MrBeast, the famous YouTuber, has a production company. He has a 50- or 100-page document that breaks down how he wants his team to behave in different situations. For example, one famous thing he would say is: "Hire consultants. Consultants are amazing — they'll save you time. I want you to post this many videos." He has an outline for the company's processes and culture.
20:39	Wade Foster	So here's what I like to do: I give the document to ChatGPT and tell it what it is. I say it's a rough draft, and then I want it to succinctly describe back to me what is in the doc. [the document] I mostly don't want it to put its opinion on it. You run that, and it describes back to you the core philosophy—what makes a YouTube video go viral. It goes through the doc a bit, but it's kind of generic, so I'm like, "I don't know that I love this." Then I might say, "Be 100x more specific." I stole that from a woman who runs product at Whoop, Hillary. I saw a YouTube video where she did this, and I found that it produced a lot of nitty-gritty details where you can actually see what's going on. All the while, for internal Zapier stuff, I'm able to get a better sense of what the team is trying to tell me.
21:54	Sam Parr	So, when would you use this? You're using the MrBeast one as an example, but give me a realistic, Zapier-like scenario—so you're within your company.
22:02	Wade Foster	Say we're trying to decide. Maybe I'll give you an example: we're trying to decide—"Should we launch a voice agent or not?" Right?
22:12	Sam Parr	Okay.
22:12	Wade Foster	A hypothetical thing: some product team will say, "Hey, I want to go make a case for this," and they'll write up a strategy memo to explain, "Here's why we should or shouldn't do this." If they're really good at making their case, they'll include qualitative data and quantitative data to support their evidence. They'll also present multiple options for how to tackle it: "We could go Route A, we could go Route B, we could do a hacky version, or here's what the all-in version looks like." Then they'll provide a strong recommendation for how to proceed. That's what a great strategy memo looks like to me. But not everyone comes in and does that. They don't always tee up the information in the way I'm best at interpreting it. So I usually start by reading the raw document myself so I understand what the team is trying to say in their own words. Then I'll upload it to ChatGPT and start asking, "What does ChatGPT think about this?" to augment it and get a thought partner on what I'm going to try, what types of questions I should ask the team, and what types of things I'm trying to get at at the end of the day.
23:32	Sam Parr	Got it. Okay.
23:32	Wade Foster	So that's how I end up using this. The interesting question is: you kind of ask it a couple of questions like this, so ChatGPT starts to warm up, and then you ask the real questions.
23:46	Sam Parr	The "100 times more specific" one — that's a really good hack. Another hack I use is: I tell ChatGPT the problem I'm trying to solve and what I think the prompt should be. Then I just say, "mhmm, now write the prompt for me."
24:01	Wade Foster	Totally. Yeah — meta-prompting, right? You're like, "I need a prompt to do X." This is the way I see some people — the ones I think, "holy cow, you're really using ChatGPT," or using these tools exceptionally well. They don't just ask their question right away. They usually ask warm-up questions to gather a bunch of context and get specific information going. Once they have enough things in ChatGPT's memory, they start asking the real things they want to know, which is: "Where are my blind spots?"
24:40	Sam Parr	But the blind-spot question is: let's say that you're using the voice AI. So you're asking, "Where are Wade's blind spots in deciding if this is a good idea?" Is that what you're saying?
24:51	Wade Foster	Well, basically—because I'm presenting it as my own idea—and then when I ask, "Where are the blind spots?" Sorry, I'm scrolling really fast.
24:59	Sam Parr	I understand what you're saying. So where am I blind? You're really saying the author — where's... [trailing off]
25:03	Wade Foster	The team's, yeah.
25:04	Sam Parr	Where are the team's blind spots? I got it. Okay.</FormattedResponse>
25:06	Wade Foster	Yeah, and you know, as you read through this now — so, like, if I was, you know, MrBeast and I was going through the blind spots, you would say, "Oh, okay, interesting: role-specific guidance." There are different roles inside the team. So should I add, like, training that's specific to different roles or should I not? You know, there's no feedback system for creative feedback — do I care about that or not? Now when I ask this blind spot question, I'll find that there's usually one or two things that ChatGPT will catch where I'll go, "Oh yeah, that's a good one." Then there will be a bunch of things where I'll be like, "Nah, I'm good. That's not an actual blind spot. That's not real — we'll be fine with that." So I find this simple back-and-forth to be good at just catching stuff. It catches simple, obvious errors. It doesn't do the work for me or do the thinking for me, but it augments it. It's like having a thought partner there that can help you navigate any of the sort of tricky decisions that you're talking to. </FormattedResponse>
26:12	Sam Parr	What are some other good warm-up questions?
26:15	Wade Foster	You had, you had a good one, which was the meta-prompt. The second thing that you can do is just blab a bunch of context to it. A lot of folks use things like Super Whisper or Whisper Flow. You talk to it... you talk to it.
26:32	Sam Parr	Yeah, I love that.
26:33	Wade Foster	So, yeah—my co-founder goes on long walks with his dog, and he will literally just talk back and forth, back and forth, back and forth. Then he'll take the transcript of that, input it, and say, "Hey, based on all of this, now I want you to go write a strategy memo on X, Y, Z," versus saying, "Hey, just write the strategy memo on this."
26:57	Sam Parr	Can you actually walk me through that? You know, I'm a history buff, and I go for walks with AI—ChatGPT—and we're just conversing about World War II. It's so funny; it's great. I'm like, "Wait, so why did they do this, and how did these people react to that?" I have a historian I talk with. Obviously, using it this way is significantly more productive. Can you walk me through his conversation with it on ChatGPT? And then does he say, at the end, "Now transcribe all of this"?
27:32	Wade Foster	He — I see. I'm trying to think how he does it. So, if you talk to ChatGPT, it does give you the transcript. I suspect he'll just take the transcript out of it. Maybe he actually tells it, "Hey, give me the whole transcript of this conversation in JSON format," or something like that, because he's an engineer and he likes stuff in JSON format. He'll be able to do something more sophisticated with it and then upload it back into the next prompt.
28:00	Sam Parr	Got it. Okay — this is awesome. So, good ways to warm up questions. I like the meta prompt: "Be 100x more specific." Another one that I like to do — that I think people... yeah, I like.
28:15	Wade Foster	"Describe it back to me."
28:16	Sam Parr	Oh, that's a good one. So you'll say, "Describe it back to me." I'd like to say, before we get into it, if there are any questions you think you need answered so you have full context, please ask them now.
28:29	Wade Foster	Yep — I do that as well. That's 100% a great one. I have a really long-running "chatty bitty" project around my personal health and wellness. There's a whole bunch of stuff I don't know in the health and wellness space, so I'll take exports from my apps. I have a workout app and I'll upload the outputs from that. I also have a sleep app and I'll upload that. Then I'm like, "Here's a bunch of data on me — I want to know how to improve my overall health and wellness." Tell me if I can give you more information that gives you better suggestions for me.
29:12	Sam Parr	"Yeah, dude — this is awesome. So, you basically, this whole episode is turning into a thing: 'How can you save me 20 or 30 hours per week?' Yeah, you've done a good job so far saving time. What are a few other ways you're using this stuff?"
29:27	Wade Foster	So, Zapier Agents is another one that's interesting to try. Most people, when they talk about agents, are usually referring to chat agents — where you're going back and forth with ChatGPT or Claude. What's different about Zapier Agents is that these are fully automated agents. These are things where you can say, "Hey, I want you to just go do this job for me, always." For example, let's make an agent that replies to your email. Actually, I don't want it to reply to my email because I'm a little scared it might respond in a way that's not up to my standards. What I actually want is for it to create drafts for my emails. So I'm just going to do a really basic thing here: let me start from scratch. You can just create a new agent here, and we're going to make a custom agent. I'm just going to kind of blab into this box here.
30:25	Sam Parr	So, is this... this is Zapier?
30:28	Wade Foster	This is Zapier Agents, and it has access to all of Zapier's tools and its trigger infrastructure. These agents wake up based on events that might happen in your world. If you get a new email, a new lead, a new customer, or a new project—basically any event that can occur in the SaaS tools you use—you can use that to wake up your agent and have the agent do something. In this case, we're going to use Gmail: anytime you get a new email, wake the agent up and have it reply to the email. We'll use a really basic prompt (this is not a prompt you would actually use, but I want to show how it works). For example: "Hey, anytime I get a new email, write a draft reply for me." Building these agents takes a little bit of effort, but you get a copilot that starts to suggest and guide you through the workflow. The copilot asks for more details because a simple instruction like "write a draft reply" usually isn't enough to make a great agent. You need to provide these agents with lots of details. The copilot will prompt questions such as: - "What email service do you use?" — I use Gmail. - "Do you want to save it as a draft reply inside Gmail?" - "What should the draft reply include?" — I get many different types of emails, so that matters. - "Should this apply to all new emails, or only emails from [specific senders]?" These questions force you to think through the details so the agent can be effective.
32:48	Sam Parr	Yeah, so I see only internal — only my coworkers, or something.
32:54	Wade Foster	Yeah. Well, so I was thinking… actually, I don't want it to draft or reply to every email. What I really wanted to do is reply—reply to everyone who is asking about a job at Zapier, because I get a lot of people emailing me and saying, "I love to work at Zapier. This would be an amazing thing."
33:14	Sam Parr	It's basically just a super-smart autoresponder, like when I go on vacation.
33:21	Wade Foster	Yeah, you used to have these autoresponders, but they were all pretty dumb. </FormattedResponse>
33:25	Sam Parr	Dude — this is great because, for MFM, we get dozens a day of people wanting to come on the podcast, and it totally ruins my inbox.
33:36	Wade Foster	Yeah, alright. Okay, so now here we're going — it's starting to build the actual prompt for the agents. > "When a new email is received in Gmail, analyze the content to determine if the sender is asking about a job or a career opportunity. Look for keywords like 'job', 'career', 'position', 'hiring', 'application'..." So, like, we're making it — it's actually making a really good prompt for ourselves. Now I start to go, "Ah, this is interesting. I can come in and edit this directly." Oh, cool. "Thank them for their interest in Zapier." Yeah, I'd like that. "Direct them to the careers page." Yep, that sounds great. "Mention that they can find current openings and apply directly through the careers page." Yep, I like that. "Keep the tone welcoming and professional." I like that. Okay, but you know what I really wanted to do is...
34:25	Sam Parr	Rub them the wrong way.
34:26	Wade Foster	No, I'm, like, you know... let's see. I want them to be polite but really brief, because I find that these agents get really wordy, and that's not how I write an email. I do really short emails where I'm like, "Hey — thanks for checking us out. Did you check out the jobs page?" Maybe I actually should ask them, thank them for their interest in Zapier, and ask if they've already applied for a job. Something like that.
34:59	Sam Parr	Yeah.
35:00	Wade Foster	If I was being really fancy, what I could do is say, "I want you to actually go look inside our applicant tracking system and go..."
35:11	Sam Parr	"That's crazy."
35:11	Wade Foster	If they've already replied, you could do something like save this as a draft reply in Gmail so you can review and send it manually. If the email is not job-related, take no action. That's really important. I do not want you writing draft emails for, I don't know, a customer. So there you go.
35:33	Sam Parr	And I...
35:33	Wade Foster	It can be like, "Alright—so this feels pretty good here. I like this." So, go test the agent and turn it on. It will go in now and look at my own inbox and see, "Hey, do I have anybody that is asking for a job in my inbox right now?" I actually cleaned out my inbox before this, so it's probably not going to find anything — which is why there was a problem testing this agent here.
35:58	Sam Parr	How hard has it been to encourage— I guess, I mean, your company is very technical. But how hard has it been to encourage your staff to really embrace this stuff?
36:06	Wade Foster	So, for us, I think we have it better than the marginal company because we're an automation company. Our employees nerd out over this stuff. We have a company value: "Don't be a robot — build a robot." We're literally trying to teach people that automation is a core primitive. But even for us, there still is a learning curve. Yeah, we employ a bunch of engineers, but we also have accountants on staff and HR folks on staff — people who probably haven't, by default, been exposed to this stuff as much. So we really do make it our mission to help people and make this technology a lot more accessible. When we were building this agent, we were just assuming that somebody's going to come in and do a bad prompt. We just know that, because most people don't come in breaking the problem down step by step, step by step, step by step like this.
37:04	Sam Parr	Do you have, like, a full—like a... you know, at 700 people you almost need a five-person team. Is it all "I'm going to..." or do you just send them a bunch of YouTube channels? I don't know. How do you train, because the shit changes every three weeks?
37:18	Wade Foster	We had a wake-up call when ChatGPT launched. We were like, holy cow — our roadmap and how we operate the company needed to shift. There is so much more opportunity and, candidly, threats to our business if we are not paying attention to this stuff. We did a handful of things that got our usage of AI from effectively zero to just shy of 100%. The last time we ran the stats, about 90% of our employees were using this stuff daily. The three things we did were: First, we called a "code red" and ran a hackathon. I stopped the company for an entire week and said, I don't care what job you're in — whether HR, accounting, support, sales, or engineering — we're all going to press pause for the week and build stuff with AI. If you're an engineer, maybe you'll build a feature for our products. If you're in recruiting, maybe you'll see how you could use ChatGPT to write job descriptions or do basic research. This was in 2023, so it was pretty basic stuff back then. The hackathon was important for people to start getting familiar with the tooling. At the end of the hackathon we did a show-and-tell. Everyone had to show off what they built. That did two things: one, it promoted accountability — people took it seriously because they had to demonstrate it to their teammates; and two, it promoted knowledge sharing, because you got to see how other people were doing things. That has been the most impactful part of my own learning. I would ask, "What prompt did you use? Show me how you did that," because many people are constantly experimenting and nerding out on this stuff in ways I don't have time to try. I benefit a lot from watching how other people are using these tools. The second thing we did was commit to running these hackathons periodically — about every three to six months. Now we don't do a full week each time; usually it's one or two days. That cadence gets people to keep coming back to the well to see what's changed and what's new, and it helps them refresh their mental model of what these models and tools are capable of.
39:57	Sam Parr	"Capable of awesome. Man, this is so awesome. Do you want to do one more, or do you have one more, or no?"
40:02	Wade Foster	I don't — I can't demo this one because it's a lot more sophisticated to demo. I want to show what great can actually look like. I think mostly what I've shown today is, candidly, pretty basic starter stuff. But this is where, if you have a couple people embedded in all your functions inside a company, you can really start to use AI at a pretty impressive rate. For example, a couple of weeks back the internet went viral because there was this guy who had been hired by a bunch of different YC [Y Combinator] companies. He was — yeah — he had like five or six jobs at once.
40:37	Sam Parr	"Tell — tell that story. So, basically, I think I forget the guy's name. I was on a mixed panel and called him out, but he was like, 'Just so you know, I just caught this one employee working for us. Turns out he's had another job.' Then dozens of other companies were like, 'Dude, he worked for me too.' Turns out 100% — he got called out. Then he did a podcast where he said, "Yeah, I've been working three to four to five jobs at any given time. I've been doing this for two to three years." And everyone said the same thing, which was, "He passed the... he was amazing." Like, I thought he was just an amazing employee, and he nailed the job interview — everyone nailed it.
41:15	Wade Foster	The job interview.
41:15	Sam Parr	"And everyone was like, 'How—how—how?' That was the most impressive part: how on earth did this guy crush the interviews so well that he got these positions? So that's the story, yeah."
41:25	Wade Foster	Yeah, and it turns out there's a whole subreddit about this. There's a subreddit called "Overemployed."
41:32	Sam Parr	"It's crazy. It's crazy."
41:34	Wade Foster	It's... I mean, hats off to them. I think if these people worked that hard at starting a business, or even in their core job, they would be incredibly successful. These folks obviously have some unique skills; they just employ them in, like, you know...
41:50	Sam Parr	Nefarious ways, yeah.
41:51	Wade Foster	It's nefarious ways.</FormattedResponse>
41:52	Sam Parr	I've always thought—there was a book, Freakonomics, that looked at drug dealers and how much work they do. It turns out they only make about $14 an hour for all that work. The conclusion was: "You guys should get normal jobs—you can make more money, and you're clearly very hardworking."
42:10	Wade Foster	Totally. So what is this? This is a template that I think would have caught this guy: it's a "candidate risk detector." Effectively, what you do here is hook into things like Ashby, Slack, Verifone, the IP API, and run through applicants. It then tries to score their risk — is this applicant potentially fraudulent? You get an applicant that comes in. It takes the details provided by the applicant, runs checks on the IP address, phone numbers, and a whole bunch of other metadata, and then compares them to other applicants to try to spot mismatches, suspicious patterns, or things like that. Ten years ago, this would have required machine learning engineers building this type of system. It would have been really tough to set up. But this was built by Casey, who is on our talent team. She's just part of the talent team — that's what she's...
43:23	Sam Parr	Okay — is Casey like Frank Abagnale Jr.? Do you remember the movie "Catch Me If You Can," where he's a check fraudster and after 20 years he gets caught and the FBI is like, "Hey, do you want a job catching other fraudsters?" So did Casey have, like, 10 jobs and you caught her?
43:37	Wade Foster	No, she didn't. But that would make a better story if...
43:40	Sam Parr	Yeah.
43:40	Wade Foster	If it... if we did.
43:41	Sam Parr	"Casey should have done more illegal stuff."
43:44	Wade Foster	Yeah, so you can see here, pretty much, how the process—how the template actually works.
43:50	Sam Parr	Do... does the way your business work? If you scroll up, it said, "Can this..." Is this like an agent that someone made, or do you guys call it agent or templates?
43:58	Wade Foster	Yeah, this is... this is actually — I mean, you could call it an "agent." This is actually just a straight-up workflow in this case, but it has AI as part of it.
44:07	Sam Parr	"Are you guys going to become a platform where people can sell the agents they create?"
44:14	Wade Foster	"I like that idea. No comment."
44:17	Sam Parr	Yeah, understood. I mean, that's a no-brainer. So that would be awesome, and it would definitely blow you guys up — very similarly to Microsoft and Shopify. The platform model is an amazing model.
44:35	Wade Foster	Yeah, and these are the types of templates you could sell, right? The thing I was showing before — the "email-reply" one — that's super simple. But this one takes some effort. Casey — I'd have to ask her, but I would guess — this probably took her a couple of days of trying to think it through: breaking it down into all the different steps and finding all the tools she needed to use. But at the other end, it saves our HR team a huge amount of headaches, because now we're not spending time on candidates who are maybe trying to pull one over on us.
45:16	Sam Parr	What do you call your industry—automation? Like platform automations?
45:19	Wade Foster	So, we think about this as like AI orchestration. At the end of the day, it's AI orchestration, workflow automation—like AI automation. This is the kind of stuff that the most sophisticated teams are doing right now. They're building things like candidate risk detectors for hiring teams, and they're using that to solve problems that, candidly, they couldn't solve before or to do work they just couldn't do before.
45:48	Sam Parr	Who's the biggest in your space? Are you the big guys?</FormattedResponse>
45:51	Wade Foster	There. I mean, yeah — we are like... it's us. It's, you know, Microsoft has a product that does stuff like this. Workato, for example, has become much more enterprise-oriented these days. Zapier is very enterprise-oriented as well. And then there's a handful of small startups that are doing things like this, too.
46:15	Sam Parr	Dude, this is awesome. Do you remember? I used to have this trick, and you're the first person who was a victim of it—I will admit that this was my trick. For years, from about age 24 to 28, I would host an event called HustleCon, where I'd get people like you to come give a talk. I would lie to the speakers. I'd say, "You're talking at 3:00. You have to be there at 10:00 a.m. for the mic check and all that stuff." At conferences, there aren't mic checks—the mics work. It's the same mic; it works fine. The reality was I wanted you to come backstage to just hang out. I wanted to, not necessarily hang out with me, but to see Wade talk. I think it was like 16 or 17—I forget exactly when—and you were there. I didn't even have to lie to you. You were there from about 8 to 7, two days in a row, just sitting on this couch with me. I don't know if you were doing this on purpose, but I'm pretty sure you were doing the same thing I was doing. It was awesome because that was probably the biggest impact I've ever had on my business—or my life. I remember being with you and the founder of The Athletic, Miguel, Casey Neistat, and all these ballers. I thought, "Wade might be a little different, but most of these guys—they're not that much smarter than me. They're not a thousand times smarter than me, but they're a thousand times more successful than me. Why does that gap exist?" It's because they are fearful, but they do it anyway. That was very inspirational.
48:00	Wade Foster	Well, I think that's… there's this meme that goes around the internet of, like, "you can just do things."
48:05	Sam Parr	Yeah.
48:05	Wade Foster	And I think that that is—if I could go back to my former self, or just talk to the graduating class or whatever—it's: just do stuff. I think most people are so scared that they're going to have egg on their face. But usually what happens is that when you fail, nobody even notices; nobody even cares. So if you mess up, who cares? Nobody saw it. Try again. There is so much advantage to be had from just trying things.
48:37	Sam Parr	"Well, you did just do stuff—you did the damn thing. I appreciate you coming here and doing this." "No, thank you for being so gracious and making this happen."
48:47	Wade Foster	You bet. When I come back, we'll actually have to do the MFN thing and jam on business ideas and stuff like that.
48:52	Sam Parr	Right after we get done hitting off of this... we'll get you scheduled.
48:57	Wade Foster	"We'll do the second one, right?"
48:59	Sam Parr	I'm down. No — I'm... we're in love with it. Alright. God bless. Thank you. That's it. That's the pod.

Discover

Listen

Catalog

Guides

Connect

Stop Doing Boring Work Yourself…

Transcript