The Top Scaling Lessons from Zapier's AI Transformation with Wade Foster, Co-Founder & CEO of Zapier

(Intro Narrator at 00:00:00) Today, we're talking to Wade Foster, co-founder and CEO of Zapier, the sponsor of today's episode. This one is all about how top companies are actually making AI scale within their organizations. You're listening to Joel Beasley, Modern CTO.

(Joel Beasley at 00:00:22) I'm particularly excited about this conversation because I've been a customer of Zapier north of ten years. So when your team reached out, I was like, yeah, this is gonna be fun. Can you tell me the origin story of how you started this company?

(Wade Foster at 00:00:38) Sure. So, gosh, it's 2011, almost fifteen years ago now. And, you know, my co-founder and I, we were working at a mortgage company. And on the side, we're playing in a blues and jazz quartet, and we're doing a bunch of hacking side projects, like building WordPress sites and plugins and that kind of thing.

(Wade Foster at 00:01:04) Anything just to make a buck. And, you know, we kind of just did what a couple of nerds would do, which is we would text each other cool startup ideas all the time. Be like, what if we made a product like this? What if we built a company around this? All this sort of stuff.

(Wade Foster at 00:01:20) And, look, most of them were just kind of just that. It was just like, hey, this is an interesting idea. And then one day, Brian had texted me, and he said, hey, I noticed that if you go to the forums of most of the early SaaS products—so think like Dropbox and Evernote and Zendesk and Salesforce and QuickBooks Online—and they all had these public help forums.

(Wade Foster at 00:01:46) And if you looked at them, there was a very common type of question, which was, when will you integrate with X? And you would go read those threads, and they all, again, follow the same pattern. It would be, me too, plus one, I can't believe you don't have this yet. And they go on and on.

(Wade Foster at 00:02:03) And eventually, a product manager would chime in and say, hey, thanks for the feedback, everyone. We'll take a look at this. And if you've ever worked in one of these companies, you know that's code for not gonna happen. And so Brian texted me this in not so many words and was like, hey, I think we could make a no-code version of this that abstracts away APIs and allows these folks to set up their own integrations between these tools. And so at the mortgage company, my day job is in email marketing. At the time, I'm in charge of setting up Marketo. And they've got this old school SOAP WSDL API. The documentation is in a PDF.

(Wade Foster at 00:02:49) I'm having a bad time. I'm not a particularly good developer. And so I hear this idea, and I'm like, oh, if this existed, I would just be using that product and not whatever the heck it is I'm trying to do over here. And so that was the nugget of the idea. And we're in Columbia, Missouri. There's a local Startup Weekend coming up. So we go to the Startup Weekend. We team up with our third co-founder, Mike. We build the first prototype. We have a lot of fun building it.

(Wade Foster at 00:03:15) And when the weekend's done, we're like, well, that was fun. Should we keep working on this thing? And we decided, yeah. Yeah. Let's go give it a shot.

(Wade Foster at 00:03:23) And, you know, we're in Central Missouri, so you don't go raise a bunch of money. You just build stuff. And so we keep our day jobs. Mike's still in school. And nights and weekends, we just kind of set out trying to build the beginnings of what is now Zapier.

(Joel Beasley at 00:03:42) Oh, it's amazing. Did the name happen at the time of the competition or later?

(Wade Foster at 00:03:46) So it happened in like five minutes. You know, we knew that the company was gonna be all about integrating APIs. And so I thought, well, it'll be pretty cool if we could fit API in the name of the company. And then E-R is a common suffix in English. So we're like, okay, what could go with A-P-I-E-R? And almost instantly, we thought, we have it. Snappier. And so we go buy the domain. We think we've got it sorted.

(Wade Foster at 00:04:20) And then after the weekend, you know, we're like, okay, we gotta go do things properly. And so we go out and start to search the space and see if, hey, is there anyone kind of in our world that's doing stuff like this? Turns out there's a quasi-competitor, sort of maybe, that had some snappy stuff going on. And we were like, let's just steer clear of that. We need to change the name. And the next obvious choice was Zapier, which is one letter shorter, so even better.

(Joel Beasley at 00:04:49) Yeah. Just like flip that S around.

(Wade Foster at 00:04:51) You know? There you go.

(Joel Beasley at 00:04:53) Now was your first integration the Marketo stuff that you were working on? Or was—

(Wade Foster at 00:04:56) Surprisingly not. But it was email. So we—well, let's see. The first thing we did was we had PayPal. We had Highrise, which was a CRM from 37signals.

(Joel Beasley at 00:05:08) You know? So that's—

(Wade Foster at 00:05:09) That's a throwback. There was Twilio. Gotta text people. And what was the other one? Oh, Twitter. Twitter was all the rage at that time. So those are the first four that we sort of added in.

(Joel Beasley at 00:05:26) That is brilliant. And then you made it so that they could integrate it themselves into their own stuff.

(Wade Foster at 00:05:30) Yeah. About a year in, we built out our developer platform that allowed people to start building their own integrations. But, yeah, for the first year, it was just us cranking away. I think we built the first 50 ourselves in, I don't know, six months, something like that.

(Joel Beasley at 00:05:46) And were you remote or in person? How did that go?

(Wade Foster at 00:05:49) Well, so it's a side project at first, so there definitely is no office. We're mostly just, you know, talking on Gchat at the time. And then our old boss at the mortgage company would occasionally let us come into the office at nights and use it. So we would meet up there. Then we moved to the Bay Area about six months in when we got accepted to YC, and then the three of us lived in a two-bedroom apartment.

(Wade Foster at 00:06:21) You do the math on that. Not exactly glamorous. And then after, Mike moved back to Missouri to be with his then girlfriend, now wife, as she was wrapping up law school. And that was kind of the first inkling of, okay, how are we gonna go build this company?

(Wade Foster at 00:06:40) Mike's in Missouri. We're in California. We gotta go—we had to hire somebody to help us with customer support. It was like, are we gonna try and hire someone in the Bay Area, or do we care? And the advice we got at the time, we had no experience hiring folks, was, well, you should just hire people you've worked with in the past that you know and trust, and that'll de-risk some things for you, given that you're woefully inexperienced at doing this on your own.

(Wade Foster at 00:07:06) And turns out most of the people we know are not in the Bay Area. So we hire a couple people. One's in St. Louis, one's in Columbia, one's in Chicago, and that kind of sets us down this distributed remote path. And, gosh, as best we could tell, it seemed to be working. Customer count was up. You know? Team was healthy. Revenue was up. Products were shipping. You know, there was nothing we looked at and said, oh, this model won't work. At least for us, it felt like it was working.

(Wade Foster at 00:07:36) So we were like, let's keep doing this because nobody else is doing it. And if we can do it well, that'll give us an advantage.

(Joel Beasley at 00:07:43) Now what point did you raise any money?

(Wade Foster at 00:07:46) We raised a small amount at Demo Day after YC about—well, I say a small amount. It seemed like it seemed like a large amount at the time to someone who has not seen that amount of money, but it was $1,300,000. Nowadays, that feels quaint. Yeah. But that was the last money we ever raised.

(Joel Beasley at 00:08:09) So you just raised once at Demo Day?

(Wade Foster at 00:08:11) That's it.

(Joel Beasley at 00:08:13) And then from that, you were able to push this thing into a profitable scenario and scale it up to a $5 billion valuation?

(Wade Foster at 00:08:21) That's what we did.

(Joel Beasley at 00:08:22) That's awesome. Are you still working there today?

(Wade Foster at 00:08:25) I am.

(Joel Beasley at 00:08:26) You still CEO position? Is that what it is?

(Wade Foster at 00:08:29) That's what it is.

(Joel Beasley at 00:08:29) Are you loving it? Are you still loving it? Are you—it's hard to stay—look, it's hard to stay excited. You can make a bunch of money. I've met some people who make a ton of money doing their dream, and they're miserable. Are you able to stay excited?

(Wade Foster at 00:08:40) I'll tell you what has been really reinvigorating. I felt like at the tail end of the last decade, it was starting to get a little, I don't know. It was like everybody's product roadmap—ours included—was stuff like Command K and dark mode, which, okay, cool, I guess. But now this age of AI, it sort of feels like this just unbounded opportunity has been unlocked, and that has just reinvigorated me and the company in just so many unique ways because it feels like a lot of the stuff we had wanted to do, but just couldn't quite get it working, now feels like, oh my gosh, we can do this and, you know, a thousand other ideas.

(Joel Beasley at 00:09:24) That is true. That's actually a really good way to phrase that. Everyone was working on dark mode, and then AI came out and we're like, oh, stop playing around, guys.

(Wade Foster at 00:09:33) Really is. It's like the—I mean, you know, nowadays, you just, you know, Claude. Give me dark mode. You're done.

(Joel Beasley at 00:09:39) Give me dark mode, and it will. It will. Give me dark mode in one style sheet.

(Wade Foster at 00:09:44) Yeah. It's like, it's not much to it.

(Joel Beasley at 00:09:47) When did Agents of Scale, your podcast that you host, when did that start up?

(Wade Foster at 00:09:51) Let's see. We did the first episode, I guess it's probably been six months ago now, give or take.

(Joel Beasley at 00:09:59) Nice. And then what are you talking about there? What's the whole angle of that podcast?

(Wade Foster at 00:10:03) So the angle is we tend to have leaders, executives come on and share what is working for them at changing how their company operates in the age of AI. It's a mix of folks, some technical, some not technical. We talk about actual practical stuff, but we talk a lot about culture too. I think a lot of the battle that companies are facing is cultural. It's like, hey, we have to unlearn all these bad habits—or it's not even bad habits. It's just these were the habits that worked in the last decade. And now in the age of AI, they have become bad habits without people actually realizing, oh, that actually is the old way of doing things. And so it's fun to just hear how different people are tackling that problem and how they're getting people to sort of shift and what tips and tricks are working, what tips and tricks are not working.

(Joel Beasley at 00:10:57) And you've been running it for six months. What's the one conversation that you remember the most?

(Wade Foster at 00:11:03) Gosh. I've gotten to talk to a ton of interesting people. You know, folks from, I don't know, HubSpot, Klaviyo, Gamma, NewFront, and many others. I think the one that I was maybe the most inspired by was the founder of Gamma. That product is just on a tear. For those who don't know, it's AI-generated slide decks. But even that, I feel like I'm underselling it. They would do so much better. But just hearing how they went from, you know, basically kind of puttering along in sort of the pre-AI era to just going all in on this AI-generated way of building and all the lessons they've learned, scaling up to, I mean, tens of millions of users. It was just pretty fun to see sort of a company that has kind of figured out a lot of things from first principles.

(Joel Beasley at 00:12:03) So they generate slide decks.

(Wade Foster at 00:12:07) Yeah. I mean, that's sort of the simplistic version. But I think what's more exciting about it is these slide decks can be dynamic. So, you know, if you think about the old way of doing this, you might have generated a slide template in Google Slides. And, you know, if you had a meeting with a sales prospect coming up, you would come in and you'd paste in their logo, paste in something relevant to them, and, you know, that would be what customization would look like.

(Wade Foster at 00:12:40) What Gamma can do is you can take that dynamic version, but you can, you know, dial it up to 11, where instead of just swapping out the logo, you swap out entire value props, entire positioning, entire chunks. And so you can start to prompt and say, hey, you know, based on what you learn about the prospect, change the visuals, change the reference logos, change the case study that you slot in, find the one that's closest based on our bank. And all that type of stuff gets to be really powerful. And I've seen folks actually hook it up with Zapier where they'll say, hey, you know, take a lead as it comes through, you know, maybe get a lead from Facebook or Google or something like that. Funnel it through Zapier. Hit these custom prompts, and then land it in this Gamma template that just goes and generates this entirely custom deck for that customer. And so, you know, that's where I'm like, man, just calling it an AI slide deck, it sort of feels like it sells it short.

(Joel Beasley at 00:13:40) Well, I've gotta play with it. This is me learning about it for the first time right now. So I've gotta check it out because it sounds pretty interesting. And, yeah, that's what we would do. When we were pitching our leadership software platform, we would just go and flip out the logo and make it as personalized as we could, but it was—

(Wade Foster at 00:13:56) Pretty much. Right? You're just like—

(Joel Beasley at 00:13:58) It wasn't a whole lot. Yeah. Interesting. Well, I'll take a look at that. What other conversations have you had that are particularly interesting to you?

(Wade Foster at 00:14:08) Well, I think the most common learning that is shared across these companies that I see from the ones that are actually driving AI change in their organizations tends to share a couple traits. One, usually, the founder CEO is preaching this stuff, but also walking the walk. So they're out there saying, hey, we need to go work this way, but they're also got their hands on the keyboards. They're doing the stuff themselves.

(Wade Foster at 00:14:39) The second thing is they generally have their procurement process reasonably dialed in. One of the biggest friction points these companies face as they get to scale is that folks wanna use it, but they haven't actually bought tools for people. And so they can't even get started because there's nothing for them to go use or play with. And so they're all stuck using, like, GitHub, you know, Copilot, instead of the most cutting-edge stuff. And then the third thing, and this is the one that I think is the absolute most common, is they're all running hackathons.

(Wade Foster at 00:15:17) And they're running hackathons for not just engineering, but sales and marketing and finance and everybody in the org. And I think what folks are finding when they do that—we've certainly seen this internally at Zapier—is that when people put their hands on the keyboard, the fear starts to fade into the background and the excitement starts to be more visceral where they're like, whoa, I can do this now. And they start to see, I actually can go solve problems that I couldn't solve before, but they also know what it's good for. They also know its limitations. They know that it's not perfect. And I don't know. It just becomes a much more energetic and pragmatic cultural shift that happens. Whereas, you know, when folks are not doing any of those, you sort of hear it—it's almost like the mainstream starts to drown out the actual boots-on-the-ground experience for folks.

(Joel Beasley at 00:16:11) Yeah. In my experience, there were two types of people. There were the people—or this is to your point of the CEOs being on board with it. There were two types of executives I'd run into. I'd run into the ones that put their hands on the keyboard for five minutes and were like, this is all hype.

(Joel Beasley at 00:16:26) They couldn't get it to do whatever they wanted it to do, and they were just like, "This is way—it'll take years." And then there's the people who pushed through that moment and either learned from a friend how to do it, saw someone do it, monkey see monkey do, or just pushed through and were persistent enough to ask it how to best use it. And then they got hooked. And so it's really just people are dismissive or addicts. Rarely do I meet anyone in the middle.

(Wade Foster at 00:16:51) It is kind of wild. Like, it does sort of feel like once you get it, it—I mean, addicts is a good word. Like, I think it is addictive. But honestly, I think it's kind of like the best kind of addictive. Like, in my mind, it's replaced my doom scrolling habit with prompting, and I'm like, yeah, that actually sounds like a win to me. Like, I'm great with this kind of addiction.

(Joel Beasley at 00:17:14) It is a win. I use it in my personal relationship. When I want to communicate things to my wife, I'm like, tell me the best way to communicate it to her. And it gives it to me, and I'm like, "Alright, here we go." She'll send me text messages sometimes that are so sweet. I'll respond back, and I'll say, "Thanks, Grok."

(Wade Foster at 00:17:33) That's amazing. I know. Now you just got to turn it around and have it critique your stuff.

(Joel Beasley at 00:17:37) Oh, I know.

(Wade Foster at 00:17:38) That's when it gets brutal, man. The funny thing I found is most people have this experience where the AI just agrees with them on everything and sort of tells them they're amazing, especially if you're just using kind of vanilla ChatGPT or something like that. And the tip I always have to share with people is when you build kind of your shared brain or your personal brain or your context, you have to use the AI's natural inclination to want to help you to your benefit. And so you basically have to instruct it to be like, "You're a coach, and the best way you can help me is to be hypercritical of my stuff." And if you do that, you start to get some really powerful coaching at the other side of these tools.

(Joel Beasley at 00:18:22) It's an amazing human psychology thing when I can tell that it's doing that, and I say, "No, no, no, I want to see the other side of this," and it'll just flip instantly. And it's like, "Thank you."

(Wade Foster at 00:18:31) Yep. It's like, that hurts, but, you know, alright, I'll go fix it.

(Joel Beasley at 00:18:37) The conversation coming up a little bit, changing gears here. Is SaaS dead? Vibe coding, everything that's going to be happening in the future. I know you had a little bit of a conversation about, I think, with Dharmesh on this. Give me the TLDR on that combo.

(Wade Foster at 00:18:54) Well, you know, I think Dharmesh had shared, absolutely not. Like, they're—I think HubSpot's a, I don't know, however many thousands of employees they have, and they're a software company through and through and through. And the number of SaaS tools that they have replaced with a vibe coding tool was zero. So, you know, I think it's a little more nuanced than that. But by and large, my take is that if you can buy software, you probably should. Like, if there's software available that solves your problem, it's generally just way cheaper to buy it and let somebody else maintain it and build all the workflows around it and all that sort of stuff. So you're probably better off doing that. But I think what a lot of us are increasingly realizing is that there are many places where we'd like software to tackle a problem for ourselves, but there's nothing to buy. And these tools are increasingly very good at helping you bridge that gap.

Then there's probably a third category, which is stuff that's kind of in the middle. You know, I think there was a lot of SaaS that kind of blossomed in the last, I don't know, decade or so that was kind of like these tiny little micro-SaaS, like these little gap filler things. Those, I think, are what folks start to feel like, "Do I really need to go spend another $50 a month on that or $100 a month on that? It really does only one tiny little thing for myself." That stuff, I'm like—you know, when people say, "Is SaaS dead?" I can start to go like, yeah, I can see that. But, you know, your CRM, your ERP, your financial tools that require proper governance, you know, regulatory issues, compliance, like, deterministic stuff—good freaking luck vibe coding your way to a solution on that.

(Joel Beasley at 00:20:46) Out of QuickBooks. You're getting audited, bro.

(Wade Foster at 00:20:50) I know. It's like, let's see what the IRS has to say about that.

(Joel Beasley at 00:20:53) No, it's interesting, though, because you've been out there talking to all these people. I'm out there talking to all these, and we're coming to the same conclusions, which I think is—those are the same sort of the way I looked at it was utility, which I think is what you were talking about with maybe just a small data transfer app you're paying for, and it's like, "Well, this is really basic. It does one single function type deal." We might be able to play with that. But I think, you know, it's easy to overreact and see the one-shot apps and just imagine SaaS is going to be dead. I think what's more likely to happen is what's always happened, and that's just margin compression. So you'll just get other startups that'll make something—now with less resources, they can make something just as good. And then a lot of it I see is stuff going forward, as you mentioned. So less about getting rid of all the stuff you're currently entangled in to run your business and more about, "Hey, we need to do that. Is this something we can build an app for with AI? Like, is it faster to put two engineers on it with some AI tools, or is it better to go hunt and go through procurement or whatever it is?" So, yeah, I think maybe there might be some velocity slowdown of new SaaS. I could be wrong. This could be the email thing all over again.

(Wade Foster at 00:22:07) Well, I think you're right. Like, I think we were already seeing this with the market correction that hit in 2023, where CFOs were hardcore hammering on seat usage. And, you know, a lot of these SaaS providers had huge numbers of seats deployed and charging exorbitant rates. And so, you know, CFOs were already laser focused on, "Hey, we're going to go scrutinize that spend, and we don't think that that spend is right-sized." Then you get into a world of AI where companies are just not growing headcount as fast, and so you don't see seats expand as much. And so, you know, to me, it's not that—for SaaS to, quote-unquote, have a problem, it doesn't have to be replaced. It just has to be compressed on the edges, where it's like, "Oh, there's not as many seats anymore. Okay, well, it's not growing as fast anymore." You know, people are scrutinizing the price tags a little bit more than they were. So, okay, you're not going to be able to command as much pricing and power. Software's still very useful, still very valuable. You don't want to go build it on your own. It's just you got a threat of a different alternative that, you know, maybe wasn't as prevalent before. And so on the margins, I do think that's kind of what we're feeling with SaaS right now.

(Joel Beasley at 00:23:26) You have instituted in May 2025 a "100% of new hires must be AI fluent." How do you do that?

(Wade Foster at 00:23:36) I look, I'm not going to say we're perfect at this, but what we practically did was we went function by function, and we just tried to understand how are the best people in those functions deploying AI. What are the use cases, and what are the benefits they're getting? And so then when we would go design hiring tests, we would just try and screen for those things and see, "Hey, are these people already doing some of this stuff, or have they figured out something more novel?" And, you know, we have a—it's on a scale of four things. There's unacceptable, acceptable, adaptive, and transformative. And, you know, it kind of comes back to the thing you were saying before where there's sort of two camps of AI users. You know, there's a lot of people when you go, like, say, "Hey, show me what you're doing with AI." You can kind of sniff it out pretty quick where they're like, "Okay, I'm kind of asking ChatGPT to rewrite this memo or summarize this thing." You're like, "Okay, well, you're not doing nothing, but there's a whole other world that you can go explore." And so you're mostly just trying to find people that are hyper-curious and, you know, have started to push the limits on these things. So I think that's the approach we generally take.

We got to go update that stuff every six months or so because of how fast the industry moves and how much the use cases are evolving really quickly. So, you know, the thing we released in mid-last year is directionally correct, but dated at this point in time. You have to update it for the new capabilities that have come out.

(Joel Beasley at 00:25:25) So let me repeat to make sure I got this right. You went function by function within the company and looked at how some of your best-performing people were using AI, and then you took some of that knowledge and somehow put it into the screening for when you're hiring people to break them into four levels, and those levels were unacceptable, acceptable—what were the other two?

(Wade Foster at 00:25:45) Adaptive and transformative.

(Joel Beasley at 00:25:47) Okay. And so that's you ranking what you think the candidate—where their AI maturity is?

(Wade Foster at 00:25:52) You got it.

(Joel Beasley at 00:25:53) Okay. Oh, and that's so smart. It's because it's by function.

(Wade Foster at 00:25:59) Oh, that's—

(Joel Beasley at 00:26:01) That's actually pretty cool.

(Wade Foster at 00:26:02) I mean, it makes sense. Like, the types of use cases a marketer has are different than the types of use cases that someone in finance has, and it's different than the types of use cases that somebody in engineering has. And yes, there's overlap. Like, generally you want a deep sense of curiosity in play. Like, that is something that is sort of ubiquitous across all the best users of AI. But the way they go deploy the tools are different, and you can kind of see—you can start to see they all sort of—there's some shared thinking bits, but there's novel things based on their own domain expertise they bring to the equation.

(Joel Beasley at 00:26:37) I got a question, and we can leave this part out if you want to. This is just for me. This is a just-for-me moment. When it's unacceptable, do you tell them that that's where their AI maturity is so that they can then go make a better choice in life? Or you just kind of like—

(Wade Foster at 00:26:55) We do give feedback to folks who get farther along in our interview process. So, but I don't think that we're giving just straight-up unacceptable stuff, "Out of here."

(Joel Beasley at 00:27:10) Get out of here. Go.

(Wade Foster at 00:27:12) It's not, you know, we're not your high school math teacher with a big red pen and an F at the top.

(Joel Beasley at 00:27:17) Isn't that so hard, though? Because you can create so much hate for your brand if you start being too honest with people too soon.

(Wade Foster at 00:27:25) You know, funny enough, this is where AI can help. It's like, let somebody down—I need you to be direct, but—and that's where Joel gives me a "Thanks, Grok."

(Joel Beasley at 00:27:38) Yeah. I need some help right now. Oh, this is fun. Tell me about this code—well, first of all, let's take a break for just two minutes. You've—how many kids do you have?

(Wade Foster at 00:27:46) I got two kids.

(Joel Beasley at 00:27:48) What are their ages?

(Wade Foster at 00:27:49) They're three and five.

(Joel Beasley at 00:27:49) What has being a father taught you about leading people?

(Wade Foster at 00:27:56) Honestly, I think communication. Like, you realize—gosh, how do I say this without being condescending? I think when you speak to a three-year-old, you realize that you can't just say, "Go get ready for bed." You have to say, "I need you to go upstairs. I need you to take your clothes upstairs. I need you to go to the potty. I need you to go wash your hands. I need you to get your toothbrush out. I need you to brush your teeth. I need you to put your jammies on." Like, you know, "go to bed" is what an adult hears. But for a three-year-old, they don't have the concept of any of these things.

Now, of course, in the workplace, no one is a three-year-old. But oftentimes, I've been working in this business for almost fifteen years. And so I can do the Zapier equivalent of, "Hey, go get ready for bed." And for someone who's brand new to the company, I might as well have said, you know, an entirely different language. And so you realize, oh, if I'm going to get the best out of people, I need to get really good about how I explain instructions to these things. Turns out, that's not only pretty useful for humans, that's also very useful for AI too. So these communication skills are—yeah, you sort of knew even pre-kids, you sort of know that this stuff is important. But raising kids is just a daily reminder of how crisply you have to communicate what you want. And if you want to get a good outcome, well, it's only going to be as good as the information you give them.

(Joel Beasley at 00:29:43) That is so good. Yes, exactly. And it's fascinating how you connected delivering communication and the correct-sized chunks with both children and AI, because I have that—that's the art with working with AI. You and I can easily give it way too much information and have it not have a quality result. It's knowing the exact—it's an art right now, and I would say that I'm unconsciously competent in doing it, but giving it just the right amount to get what I need for that specific situation.

(Wade Foster at 00:30:18) Totally. And the more you play with it, you start to get a sense of, "Oh, it's going to need this bit of detail. It's going to need this bit of detail, and it's going to need this bit of detail on it." And the order in which I give it to it, and sort of—you're almost priming the pump where it's like, "I've got this task that I'm leading up to it, but I'm not going to just go ask it to do the task first. I'm going to prime it with this and prime it with this and prime it with this and prime it with this." And then right when you sort of think you've got it all, you're like, "Okay, now that you've collected all this stuff, here's what I want you to go do." And you just get way better outcomes than saying, "Go to bed."

(Joel Beasley at 00:30:56) Tell me about this code red at Zapier. It sounds very intense. What was that?

(Wade Foster at 00:31:02) So we'd never had a code red before, but, you know, this would have been in spring of, I guess, '23. So, you know, we had seen the ChatGPT moment, and we were pretty excited about the ChatGPT launch. But that was kind of just a point in time. And so, you know, we didn't really mobilize the company with a differing set of urgency and energy at that moment in time. Rather, we were just like, "Hey, check this product out." We shared it in general in Slack. We're like, "We like cool products. You like cool products. Like, check out a cool product." Like, that was kind of the vibes. And it was very much just, "This might be interesting for your roadmap. So if it is, take a look. Maybe you'll want to change it." So very suggestive, very organic, et cetera.

(Wade Foster at 00:31:56) And, you know, over the course of the next six months, you know, we started applying pressure, you know, softly, fairly gently in a handful of places in the organization where we felt like AI was gonna have a bigger impact. But, again, not a lot of mandates. The thing that caused us to call the Code Red was the GPT-4 launch. Reason why is it happened roughly six months after ChatGPT came out, and the delta between 3.5 and 4, both in terms of model quality and costs, was pretty big. And so now you've got two data points.

(Wade Foster at 00:32:36) It's not exactly a trend line yet, but it didn't take a genius to look out and go, if any of this continues in any way, this has big implications on, you know, how we work, the products we offer, everything. And so we felt like the company sort of needed to meet that moment, and we just borrowed from, you know, medical terminology and just said, hey, Code Red. We're going on this stuff. We didn't even define Code Red internally up until that point.

(Wade Foster at 00:33:15) So it definitely was jarring for the company because a lot of folks were like, what does this even mean? It sounds intense.

(Joel Beasley at 00:33:23) Yeah.

(Wade Foster at 00:33:23) And, you know, we're like, yeah, it's meant to be intense. We need to wake up because I think this is the type of thing that is worth, you know, playing that card. Like, I think as a CEO, you have to be very careful because if you pull that lever all the time, people learn to ignore it. But if you pull it in the moments that matter, it's jarring.

(Wade Foster at 00:33:44) Yes, but it's jarring in the way that people are like, okay, this one matters. I think the most impactful thing we did right after we called that Code Red, we did a bunch of stuff, but the one that really mattered was the hackathon. Like, we stopped the company for a week.

(Wade Foster at 00:34:02) We had everybody put their hands on a keyboard and start building with this stuff. And we went from about, yeah, just under 10% of people who were using AI, you know, weekly in their job to over 50% in one week. And that was easily the singular most important thing that lurched the company ahead. And I think, you know, in spring of 2023, we were probably already one of the more sophisticated companies that were using AI just because of doing the hackathon. It took, you know, I think, another year for us to get that to, you know, 100% of folks using AI regularly.

(Wade Foster at 00:34:48) But that was a, yeah, I mean, it's just a big moment to reorient the company around what was coming.

(Joel Beasley at 00:34:56) That's awesome. Do you have a favorite model provider? Or do you, for different functions, do you like different ones?

(Wade Foster at 00:35:01) I use all sorts of stuff for different things. I like to have them critique each other. You know, I do think I've been using Opus 4.6 recently. Launched last week, it's like a faster o5 or 4.5. 4.5 was pretty good.

(Wade Foster at 00:35:17) The GPT 5.3 is really good for hard tasks, and it can think really, really long. So that's a pretty impactful model. You know, I use Gemini for Nano Banana. I use it to—

(Joel Beasley at 00:35:38) Nano Banana is pretty good.

(Wade Foster at 00:35:39) I know. It's pretty fun. I use it to—I have one use case where anytime I go to a restaurant, I always talk to Gemini. I don't know if it's actually better, but I'm just like, Google has Google reviews, so I just sort of feel like it should be better at that. I don't know.

(Wade Foster at 00:35:54) I try to keep all our AIs happy. Like, I want all the models to be on good terms with me, you know, just depending on how things roll.

(Joel Beasley at 00:36:03) Sometimes you just talk to them and say nice things to them.

(Wade Foster at 00:36:06) Yep. And I have—like I said, I switch back and forth and have them critique each other's work and, like, I assign the same tasks to multiple ones of them, and I just kind of like to get the differences. You know?

(Joel Beasley at 00:36:19) That is awesome. I'm curious. You get to see how people are using these Zaps across the entire, you know, organization, all these industries. Tell me about what people are doing in relation to hooking AI workloads up to it. Are you seeing anything interesting?

(Joel Beasley at 00:36:36) Are they using it for human-in-the-loop stuff? Tell me about that.

(Wade Foster at 00:36:40) Yeah. I think the most practical usage we see is for these workflows that require high reliability. And so one of the nice things about Zapier is it's a deterministic workflow builder. Works the same way every single time. And the nice thing about adding AI to that is you can get pretty high reliability out of a task.

(Wade Foster at 00:37:17) And so what that might look like is, hey, I've got a lead coming through the site. I wanna go hit an enrichment API or wanna have it go do deep research, and then I wanna go land it in the CRM. Pretty simple workflow. In the past, you just wouldn't do the AI step.

(Wade Foster at 00:37:31) You just take, you know, first name, last name, email, company, URL, plop it in the CRM. But now you can go do this AI enrichment step and just get a lot more interesting context for it. You could plug it into the Gamma use case we were talking about before. Go generate a custom slide deck for these things. And so there's these hybrid deterministic AI workflows that work really well in a setup like this.

(Wade Foster at 00:37:58) And, yeah, I think they're pretty powerful compared to a purely agentic use case that, you know, can create pretty magnificent demos and can work okay some of the time, but some of the time, it just doesn't, and you're not exactly sure why. You know, these tools do have a little bit of a black box feel to them. And, you know, you have to set up evals and all these systems to really eke out those gains. And I think a lot of workflows internally, that's just not something you're gonna wanna go do. Like, it makes sense for, you know, if you're building a product and you need something production grade, but for your marketing workflows, your sales workflows, things like that, you kinda wanna blur between the lines a little.

(Joel Beasley at 00:38:44) Does Zapier have any consultative part of the business where they actually help businesses build these workflows?

(Wade Foster at 00:38:51) Yeah. We have started an AI transformation program where we'll come in and work with, you know, either the whole company or a particular function in your company, and we'll help modernize the whole thing. And so, you know, a lot of that, of course, we're using Zapier, but, you know, we'll help you with a wide variety of tools. We integrate with 8,000 different tools. We integrate with all the major model providers.

(Wade Foster at 00:39:15) And so, you know, we can sort of help answer questions and advise on, you know, maybe the best way to sort of modernize your stack as well too, and, you know, do some lightweight support for those tools also along the way. And I think that's been pretty helpful for a lot of these companies because, you know, especially folks that either functions or teams that are just not super well versed in this, a lot of people just don't know where to start. You know, they kind of have used ChatGPT. They know how to summarize an email and, you know, write a cool poem or something like that. But that's pretty different from, you know, a production AI workflow.

(Joel Beasley at 00:40:01) If you are spending your free time using AI to write cool poems, please ask your AI for help. Social life things to do. No, I'm kidding. I'm kidding.

(Wade Foster at 00:40:14) You've probably seen this too. Like it's like, it's Kevin, what's going on?

(Joel Beasley at 00:40:19) I might do that, actually. I might do that for my wife. Just have—maybe I'll do that this afternoon. I'll email it to you and let you know how it went, Wade. Well, Wade, alright, two things I wanna make sure we touch on. They're very close to my heart before we wrap up here. The first one is, can you tell me the correct way to say the name of the company?

(Wade Foster at 00:40:40) Zapier makes you happier.

(Joel Beasley at 00:40:42) Okay. Thank god. Do you have any idea how many arguments I've had with my British friends that it's not Zapier? Oh my gosh. I finally have an official clip.

(Joel Beasley at 00:40:52) I can clip this, and I'm sending it to all my British friends.

(Wade Foster at 00:40:56) You know, I'd have debates with our old CMO about this. And the way I justified it to myself is that the naming confusion just was free word-of-mouth.

(Joel Beasley at 00:41:10) Yeah. Well, it's a zap. Your little actions are little zaps, like zap, zap, zap, like quick, like little zaps.

(Wade Foster at 00:41:18) Yeah. So it's just like things.

(Joel Beasley at 00:41:19) You don't zape things because you zap things. I've had that conversation more times in my life than I would like to admit.

(Wade Foster at 00:41:29) My old family has been like, could you not afford the other P? Like, come on.

(Joel Beasley at 00:41:35) No. It's efficient. It's efficient. I love API, dude. I would've never have thought that the API, right?

(Wade Foster at 00:41:40) We were more clever about it. If you go find our very first logo, the API was in gray, and the Z and the E and the R was in orange. And so we're a little too cute for school, I think.

(Joel Beasley at 00:41:52) Hey, your revenue, though. That's all—that's what matters. You're making an impact in the marketplace. And the next thing as we wrap up, I love this idea that I heard you have, and I want you to tell me about it, about you think people should automate themselves out of a job.

(Joel Beasley at 00:42:09) Because we share this view together, but I rarely run into people who articulate it. And so I wanna hear your version of it.

(Wade Foster at 00:42:16) Well, you know, I think there's this fear amongst the workplace where, you know, oh, if I automate myself out of a job, then I will just be let go. I'll become redundant. I certainly would never do this, and I know of no other CEO or leader where a person came to me and said, I have automated my entire job, whose reaction is not holy crap. Show me how, and I have nine other projects that I'm ready to go turn you loose on. Like, that is such—if you've legitimately figured out how to do that, you have instantly become one of the most valuable employees in the company because you know how to do something that is just very, very rare.

(Wade Foster at 00:43:02) And so I think the people that are thinking in systems and are figuring out how to break these down, yeah, you have one of the most valuable traits, and you should not be nervous about doing that. You should, you know, run full steam ahead and use more automation.

(Joel Beasley at 00:43:21) I love it. It's so hard, and it is so rare. Like, as a founder to see that if somebody came to me and did that, I'd be like, well, let's adjust your comp plan, and let's get you on some new projects because—

(Wade Foster at 00:43:32) Exactly.

(Joel Beasley at 00:43:33) We want you to—we want you to love this place. We're not gonna say let's go home.

(Wade Foster at 00:43:39) Well, I guess that's it. You know? Like, have a good life.

(Joel Beasley at 00:43:45) But I get it, though, because early on in your career, I was there. See, the reason why I have so much empathy for this is when I was very, very early in my career, I could understand how that could be. I could. And I can understand that feeling of, you know, let's be safe, and let's inject myself into this role versus fully automate it out. So I get it, and I have sympathy for it.

(Joel Beasley at 00:44:10) And I think that's why as an employer, now I just—tell me if you've come up with anything better, but I just say it out loud a lot just so that people hear me remind them. Because, like, you could say it once in their onboarding, but, like, if you don't build it into the culture of it being something that's said frequently, it doesn't have the same impact.

(Wade Foster at 00:44:28) We have a value at Zapier called don't be a robot, build the robot. And so I think we've just attracted people who are more predisposed to this mindset, probably in part because of the value, in part because of the products we build. Doesn't mean that we don't have folks that get a little fearful on this, that, or the other, but I think on the margins, we probably have less of this challenge than maybe your typical American corporation.

(Joel Beasley at 00:44:59) Well, I mean, it's at the root of what you guys do. You automate mundane things. You make it easy to do that.

(Wade Foster at 00:45:04) Yeah. Like, everyone here wants to take a machete to boring work. So—

(Joel Beasley at 00:45:11) I like it. Where could people learn more about the pod—how can they subscribe to your podcast?

(Wade Foster at 00:45:16) Well, check out Agents of Scale on any of your preferred podcast platforms, and you can follow me on socials, Wade Foster, on X or on LinkedIn. And check out Zapier if you haven't.

(Joel Beasley at 00:45:29) Check it out. I'm a customer. They've been awesome for over a decade. We did it, Wade. We made a podcast.

(Joel Beasley at 00:45:35) How do you feel?

(Wade Foster at 00:45:36) This was great. Super fun, Joel.

(Intro Narrator at 00:45:39) Before you go, check out a little bit of Wade's podcast, Agents of Scale. Here's the first few minutes of their episode with John from Gamma. Listen to the full episode wherever you get your podcasts.

(Wade Foster at 00:46:00) Hey, folks. Welcome to Agents of Scale. It's a show where we are exploring how the world's most forward thinking leaders are operationalizing AI across their companies. I'm your host, Wade Foster, cofounder, CEO of Zapier. And today, I'm joined by John.

(Wade Foster at 00:46:14) He's a cofounder and head of product at Gamma. If you haven't checked out Gamma, Gamma is reinventing the slide deck, and they're doing it with AI. Who doesn't want an AI to make all their slide decks? Sound awesome. And they've had some really impressive growth, scaled to 50 million users in just a year.

(Wade Foster at 00:46:36) But there is a big old story behind that. Not exactly overnight success as I think a lot of folks like to portray it. But before we talk about Gamma, John, I want you to tell me about PlateMATE. What's the story behind PlateMATE?

(John at 00:46:52) Well, happy to be here. Thanks for hosting me. And I love that we're going back in my history. This is great. So PlateMATE is a project that I actually worked on in college way before any of the generative AI would have made it actually good or easy.

(John at 00:47:06) But the basic concept, you could say it was my first entrepreneurial idea, but I just did it in college as a project, was what if you could just take a picture of your food, your lunch, your dinner, whatever, and have an app analyze it and tell you you had this much protein, this many carbs, this many calories, and add it all up instead of people having to count this up for themselves or whatever it might be. Great concept, but I would say before its time in terms of technology. But it was a neat one to try to build. At the time, AI just could not solve this problem. We were so far—this was probably, like, gosh, 15 years ago now.

(John at 00:47:37) AI was just not there to be able to do this. And so we actually used human intelligence. We used Amazon Mechanical Turk. If listeners don't know what Mechanical Turk is, it's this platform where you could basically pay people by the minute, maybe even by the second to complete tasks for you online for, like, I don't know, probably below minimum wage. And so you would just put tasks up there, and the task would be like, label—draw a box around all the foods you see on this plate.

(John at 00:48:01) And another person's task would be identify what's in box number three, and they'd be like, I don't know. I think that's mashed potatoes or something. And then person number three's job would be, look at the mashed potatoes and say, okay, match this to this USDA food category so that we can actually find out how many calories or carbs are in there. And then our app would stitch it all together.

(John at 00:48:21) It was a cool concept. This was when the wisdom of crowds was sort of like this big thing. I had, like, dreams of building a mobile app. Unfortunately, it never got beyond a research project because it never actually got to be cost feasible at the time. I think we came to the point where it would cost maybe, you know, like, a buck per meal to have someone label all your food for you.

(John at 00:48:41) So you're talking three meals a day, 30 days a month, like $100 a month subscription just to break even on somebody telling you what's in your food. I think there could be some medical applications for it or whatever, but hard to see it really taking off as a business. So unfortunately I had to put it back on the shelf, but I'm psyched to see that now their company is doing this with AI at way lower cost.

(Wade Foster at 00:49:00) Yeah. I mean, these products exist. Like, if you were to build it today, how quick do you think you could build a version of this today compared to Facebook?

(John at 00:49:07) Code it so fast.

(Wade Foster at 00:49:08) I really think, yeah. And I think the weekend's probably there's—

(John at 00:49:10) Actually no magic there now. It's like the vision models just do it. I think, you know, the hard part is like tuning the accuracy and making sure it's right and getting people to trust it. I also had this whole other idea of, like, what if it was a social network? What if we all posted our food pictures, which people have been doing since time immemorial on Instagram and Twitter or whatever.

(John at 00:49:29) But then I would label what's in your lunch, and you would label what's in mine, like for fun. I think there's still some idea there. I never figured out the formula for it. But yeah, just the version where AI does it, I'd code it in an afternoon. It'd be fun.

(Wade Foster at 00:49:40) Yeah. I've heard it's still actually pretty tricky because it's hard to see, like, what are the ingredients in this. Like, you need to have a quarter there for, like, you know, to—

(John at 00:49:50) To be like, oh, for scale.

(Wade Foster at 00:49:52) And then, yeah, you just don't know how much butter did the person actually put in this thing or not. Like, it's hard to tell.

(John at 00:49:58) But, you know, it's all about what your baseline is. So the thing we did in that paper that I thought was really interesting was we actually compared it to professional nutritionists and also to self-reporting of people on their own eating. And it turns out just averaging together a bunch of these Mechanical Turk people did at least as well as the professional nutritionists and often better than people self-reporting. The nutritionist because it was just one person working off the same limit of information. Whereas if you averaged five or 10 strangers, you actually could get a more calibrated average. And better than self-reporting because it turns out even when we know how much butter we put in, we lie to ourselves about it.

(John at 00:50:30) And so an impartial stranger without the information often does better than you yourself with the information. And so considering that, it might still work.

(Wade Foster at 00:50:38) Yeah. I've heard that, like, when people ask, like, why is restaurant food so much better than the homemade one? And it's like, well, you're paying me, the cook, so you don't have to see how much butter I actually put in this thing.

(John at 00:50:51) I love that. Please don't tell me. I couldn't possibly know. Then you make cookies at home, and you're like, dear lord. That's how you do this?

(Wade Foster at 00:50:58) It's like, yeah. So that's why it tastes so good. I love it. What do you think you learned about building products from Platemate? You know, since that time, you've had a pretty epic career across, you know, Microsoft and Optimizely and now Google. Like, have worked on some pretty impressive products.

(Wade Foster at 00:51:21) Like, what groundwork did you sort of pick up on working on Platemate?

(John at 00:51:26) Gosh. That's a good question. I mean, the first thing is, like, work on a universal problem that lots of people have.

(Joel Beasley at 00:51:33) Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email, [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.